Add snapshot evidence to cancel-cascade MTF issue doc

Append "Snapshot evidence (2026-05-13)" section to
`cancel_cascade_too_slow_under_main_thread_forkserver_issue.md`
documenting `fail_after_w_trace` diag capture results for
`test_nested_multierrors` under the MTF backend — reproduction cmd,
ptree analysis, observed hang signature, and updated triage plan.

(this commit msg was generated in some part by [`claude-code`][claude-code-gh])
[claude-code-gh]: https://github.com/anthropics/claude-code

(cherry picked from commit 5372fd178a)

.github/workflows/ci.yml

@@ -88,10 +88,27 @@ jobs:
 
 
   testing:
-    name: '${{ matrix.os }} Python${{ matrix.python-version }} spawn_backend=${{ matrix.spawn_backend }} tpt_proto=${{ matrix.tpt_proto }}'
+    name: '${{ matrix.os }} Python${{ matrix.python-version }} spawn_backend=${{ matrix.spawn_backend }} tpt_proto=${{ matrix.tpt_proto }} capture=${{ matrix.capture }}'
-    timeout-minutes: 16
+    timeout-minutes: 20
     runs-on: ${{ matrix.os }}
 
+    # NOTE on the matrix shape — the `capture=` mode follows
+    # `spawn_backend`:
+    #
+    # - `trio` / `mp_*` backends use `--capture=fd` (default)
+    #   for per-test attribution of subactor *raw-fd* output
+    #   in failure reports.
+    # - Fork-based backends (`main_thread_forkserver`,
+    #   `subint_forkserver`) REQUIRE `--capture=sys` because
+    #   fork-child × `--capture=fd` is a known deadlock
+    #   pattern. See the long NOTE in `tractor._testing.pytest`'s
+    #   `pytest_load_initial_conftests` for the mechanism +
+    #   tradeoff write-up.
+    #
+    # If a future matrix row adds a fork-spawn backend
+    # WITHOUT setting `capture: 'sys'`, the
+    # `pytest_load_initial_conftests` hook fail-fasts on `CI=1`
+    # with a clear error msg. So the matrix is self-policing.
     strategy:
       fail-fast: false
       matrix:
@@ -118,6 +135,26 @@ jobs:
           'tcp',
           'uds',
         ]
+        capture: [
+          'fd',  # default for non-fork backends
+        ]
+
+        # Fork-based backends — added via `include:` so each
+        # cell carries its REQUIRED `capture: 'sys'` mode.
+        # Linux-only for now; macOS coverage TBD pending
+        # local validation.
+        include:
+          - os: ubuntu-latest
+            python-version: '3.13'
+            spawn_backend: 'main_thread_forkserver'
+            tpt_proto: 'tcp'
+            capture: 'sys'
+          - os: ubuntu-latest
+            python-version: '3.13'
+            spawn_backend: 'main_thread_forkserver'
+            tpt_proto: 'uds'
+            capture: 'sys'
+
         # https://github.com/orgs/community/discussions/26253#discussioncomment-3250989
         exclude:
           # don't do UDS run on macOS (for now)
@@ -158,7 +195,11 @@ jobs:
           -rsx
           --spawn-backend=${{ matrix.spawn_backend }}
           --tpt-proto=${{ matrix.tpt_proto }}
-          --capture=fd
+          --capture=${{ matrix.capture }}
+        # NOTE: capture mode is matrix-driven — `fd` for
+        # non-fork backends (per-test fd attribution),
+        # `sys` for fork-based (avoids fork-child x
+        # capture-fd deadlock). See matrix-NOTE above.
 
   # XXX legacy NOTE XXX
   #

ai/conc-anal/cancel_cascade_too_slow_under_main_thread_forkserver_issue.md

@@ -0,0 +1,314 @@
+# Cancel-cascade `trio.TooSlowError` flakes under `main_thread_forkserver`
+
+## Symptom
+
+Running the full test suite under
+
+```bash
+./py313/bin/python -m pytest tests/ \
+  --tpt-proto=tcp \
+  --spawn-backend=main_thread_forkserver
+```
+
+surfaces a single, **rotating** `trio.TooSlowError`
+failure each run. The failure isn't deterministic on
+test identity — different test each run — but it
+ALWAYS looks like:
+
+```
+FAILED tests/<file>::test_<name> - trio.TooSlowError
+==== 1 failed, 373 passed, 17 skipped, 11–12 xfailed,
+       0–1 xpassed, ~550 warnings in ~6min ====
+```
+
+Pass rate: **~99.7%** (373 of 374 non-skip tests).
+Wall-clock per full run: 5–6 min.
+
+## Tests observed flaking so far
+
+Each row was the SOLE failure in a separate run:
+
+| run # | test |
+|---|---|
+| 1 | `tests/test_advanced_streaming.py::test_dynamic_pub_sub[KeyboardInterrupt]` |
+| 2 | `tests/test_infected_asyncio.py::test_context_spawns_aio_task_that_errors[parent_actor_cancels_child=False]` |
+
+Both share the same shape:
+
+- **Cancel cascade** of N subactors back to a parent root actor.
+- N ≥ `multiprocessing.cpu_count()` for `test_dynamic_pub_sub`
+  (it spawns `cpus - 1` consumers + publisher + dynamic-consumer).
+- N ≈ 2 for `test_context_spawns_aio_task_that_errors` —
+  but each subactor is `infect_asyncio=True`, so each
+  cancel involves the trio↔asyncio guest-run unwind
+  which is structurally heavier than pure-trio.
+- Test wraps the cascade in `trio.fail_after(N seconds)`
+  and the cap fires before the cascade completes.
+
+The exact failing test rotates because each test is
+independently close to the cap; whichever happens to
+be unlucky in scheduling/CPU-contention on a given run
+is the one that times out.
+
+## Root-cause family
+
+`hard_kill` (`tractor/spawn/_spawn.py:hard_kill`) runs
+the SC-graceful teardown ladder per subactor:
+
+1. `Portal.cancel_actor()` — graceful IPC cancel-req.
+2. Wait `terminate_after=1.6s` for sub to exit.
+3. If still alive: `proc.kill()` (SIGKILL).
+4. (NEW) `_unlink_uds_bind_addrs()` — post-mortem
+   sock-file cleanup for UDS leaks (issue #452 fix).
+
+For a cascade of N subactors, each pays steps 1–4. If
+graceful-cancel doesn't complete within 1.6s for ANY
+sub, that sub eats a full 1.6s of `move_on_after` plus
+the `proc.wait()` post-SIGKILL.
+
+Worst case under fork backend with N=cpus subs:
+- N × 1.6s = 16s+ on a 10-core box just for the
+  graceful timeout phase
+- Plus per-spawn fork-IPC handshake cost compounds
+  during teardown (each sub's IPC cleanup goes through
+  the same forkserver coordinator)
+- Plus the new autouse fixtures
+  (`_track_orphaned_uds_per_test`,
+  `_detect_runaway_subactors_per_test`,
+  `_reap_orphaned_subactors`) all run at test
+  teardown, adding small (10s of ms) but cumulative
+  overhead
+
+Current cap: 30s (`fail_after_s = 30 if
+is_forking_spawner else 12`). Empirically fits the
+median run but the tail breaks ~0.3% of the time.
+
+## NOT regressing
+
+To confirm this is a flake and not a regression:
+
+- Pre-`WakeupSocketpair`-patch baseline: tests
+  HUNG INDEFINITELY (busy-loop never released).
+- Post-patch: pass-or-fail-fast, ~99.7% pass, the
+  occasional cap-hit fails in bounded time (<60s for
+  the offending test).
+- Same test PASSES under `--spawn-backend=trio`
+  (no fork, no hard-kill compounding).
+
+So the suite is dramatically better than before; the
+remaining flake is a known-tolerable steady-state.
+
+## Possible mitigations (ranked)
+
+### A. Bump the cap further
+
+Cheapest. Change the per-test `fail_after_s` from 30
+to e.g. 60 for fork backends. Pros: trivial. Cons:
+masks any genuine slowness regression we'd want to
+catch.
+
+### B. CPU-count-aware cap
+
+For tests whose N scales with `cpu_count()`, scale
+the cap too:
+
+```python
+fail_after_s = (
+    max(30, cpu_count() * 3)  # 3s/actor floor
+    if is_forking_spawner
+    else 12
+)
+```
+
+Pros: scales with the actual cancel-cascade work.
+Cons: still arbitrary multiplier.
+
+### C. `pytest-rerunfailures` for these tests only
+
+Mark the known-flaky tests with
+`@pytest.mark.flaky(reruns=1)` (needs
+`pytest-rerunfailures` dep). Single retry hides
+genuine ~0.3% transient flakes.
+
+Pros: no cap change, surfaces persistent failures
+loudly. Cons: adds a dep, retries can mask real bugs
+if used widely.
+
+### D. Reduce `hard_kill`'s `terminate_after`
+
+Drop from 1.6s → 0.8s. Cuts the worst-case cascade
+time roughly in half. Risks: fewer subs get a chance
+to run their cleanup before SIGKILL → more orphaned
+state for the autouse reapers to handle (ironically,
+adds back overhead elsewhere).
+
+### E. Profile + targeted fix
+
+Add `log.devx()` markers in `hard_kill` to time each
+phase. Identify if any subactor is consistently
+hitting the 1.6s cap (vs. exiting in <0.1s). If so,
+that sub has a teardown bug worth fixing at source.
+Pros: actually fixes the underlying slowness. Cons:
+real investigation work, deferred from this round.
+
+## Recommendation
+
+Land this issue-doc as the tracker. Apply **(B)** as
+a small follow-up — cheap and proportional. If it
+still flakes, escalate to **(E)** with a `log.devx()`
+profile-pass.
+
+`(C)` is a backstop if `(B)` doesn't quite get there
+and we need green CI faster than (E) can deliver.
+
+## Verification protocol
+
+After applying any mitigation:
+
+```bash
+# Run the suite N times back-to-back, count failures.
+# A persistent failure on the SAME test == real bug.
+# Failures rotating across tests == still cap-related.
+
+for i in $(seq 1 5); do
+  ./py313/bin/python -m pytest tests/ \
+    --tpt-proto=tcp \
+    --spawn-backend=main_thread_forkserver \
+    -q 2>&1 | tail -2
+done
+```
+
+Target: 0 failures across 5 runs ⇒ ship. 1–2 failures
+still rotating ⇒ apply (C). Same test failing twice
+⇒ escalate to (E).
+
+## Snapshot evidence (2026-05-13)
+
+After landing the `fail_after_w_trace` /
+`afk_alarm_w_trace` capture-on-timeout helpers
+(`tractor._testing.trace`), `test_nested_multierrors`
+on the `main_thread_forkserver` backend produces
+**reproducible diag snapshots** at
+`$XDG_CACHE_HOME/tractor/hung-dumps/test_nested_multierrors_start_method_main_thread_forkserver__<iso-ts>/`.
+
+### Reproduction
+
+```bash
+pytest \
+  -v --verbose --durations=10 \
+  --spawn-backend=main_thread_forkserver \
+  --tpt-proto=uds \
+  --capture=sys --show-capture=stderr -rxX \
+  tests/test_cancellation.py::test_nested_multierrors
+```
+
+The test is `xfail(strict=False)` for MTF — it RUNS
+each invocation so snapshots accumulate, but doesn't
+break `--lf` workflow.
+
+### Consistent shape across runs
+
+5+ snapshots taken back-to-back show the SAME pattern:
+
+- **Timing:** ~10s wall-clock total. Inner
+  `fail_after_w_trace(10)` fires at exactly T=10s;
+  cascade's `nursery.__aexit__` takes ~0.6s more to
+  gather + propagate the resulting
+  `BaseExceptionGroup`. **Trio backend completes the
+  SAME test in <6s** — so the MTF cascade is ~2x
+  slower at minimum.
+
+- **`BaseExceptionGroup` shape:** mixed
+  `[RemoteActorError, Cancelled]`. The first
+  subactor's natural error-propagation (`assert 0`
+  raised → `RemoteActorError` portal-result)
+  arrives before T=10s; the OTHER subactor's
+  portal-wait is still in flight at T=10s, gets
+  cancelled by `fail_after_w_trace`'s scope-cancel
+  → returns `Cancelled` instead.
+
+- **Orphan-spawn skew:** snapshot's `orphans` bucket
+  (after the `_is_tractor_subactor` cgroup-slice
+  override fix) consistently shows 2-4 init-adopted
+  procs at `depth_3` and `depth_1` levels — these
+  are the leaves whose parent (`depth_2` spawner)
+  was killed mid-cascade but who hadn't yet seen
+  the cancel signal themselves.
+
+- **UDS sock-leak:** 2-6 dead-orphan socks per run
+  (varies with cascade timing). The
+  `track_orphaned_uds_per_test` fixture reaps them
+  post-test → contamination is isolated per-invocation.
+
+### Capture mechanism
+
+`fail_after_w_trace` covers two firing paths:
+
+1. **`trio.TooSlowError`** raised at scope-exit
+   (body returned cleanly past deadline) — direct
+   `except` handler captures.
+
+2. **Scope-cancel + body raises non-`Cancelled` exc**
+   (e.g. `nursery.__aexit__` wraps timeout-induced
+   `Cancelled` into a `BaseExceptionGroup` that
+   escapes before `trio.fail_after`'s exit-check
+   could fire `TooSlowError`) — body-raise `except`
+   handler checks `scope.cancel_called` and
+   captures if True. This path catches the
+   `test_nested_multierrors` shape specifically (see
+   "BaseExceptionGroup shape" above).
+
+The snapshot dir contains:
+- `trace.txt` — `ptree` + `hung_state` (kernel
+  `wchan`/`stack` + `py-spy dump --locals` when
+  sudo cached), with `include_strays=True`
+  surfacing any cross-test ghost subactor trees in
+  the `orphans` bucket.
+- `bindspace.txt` — UDS bindspace classification
+  (live-active / orphaned-alive / orphaned-dead).
+- `meta.json` — `{pid, label, captured_at, sudo_cached}`.
+
+The end-of-session `pytest_terminal_summary` hook
+in `tractor._testing.pytest` lists every snapshot
+dir from the run so you don't have to scroll back
+through captured-stderr lines:
+
+```
+========================= tractor hang-snapshot index ==========================
+N `fail_after_w_trace` / `afk_alarm_w_trace` snapshot(s) captured this session:
+  <test-id>
+    → /home/.../.cache/tractor/hung-dumps/<label>__<ts>
+```
+
+### Caveats
+
+The snapshot fires AFTER the body-raise (not at the
+exact moment of scope-cancel), so the parent's
+py-spy frames show `_do_capture_snapshot` itself
+running, NOT the cancel-cascade hang frame. To see
+the actual hang state, manual `acli.ptree` /
+`acli.hung_dump` from a second terminal at T=10s
+would be needed — **not currently possible**
+because per-test reaper fixtures clean up ~0.6s
+post-timeout. See follow-up TODO in
+`tractor/_testing/trace.py` for a
+`TRACTOR_TRACE_HOLD=1` env-var pause mode.
+
+## See also
+
+- [#452](https://github.com/goodboy/tractor/issues/452) —
+  UDS sock-file leak (related — `hard_kill`'s
+  cleanup phase contributes to cascade time)
+- `ai/conc-anal/trio_wakeup_socketpair_busy_loop_under_fork_issue.md`
+  — the upstream-trio fix that turned this from a
+  100% hang into a 0.3% flake
+- `ai/conc-anal/infected_asyncio_under_main_thread_forkserver_hang_issue.md`
+  — the asyncio variant which contributes to one of
+  the rotating failures
+- `tractor/spawn/_spawn.py::hard_kill` — the SIGKILL
+  cascade source
+- `tractor/_testing/_reap.py::_track_orphaned_uds_per_test`,
+  `_detect_runaway_subactors_per_test`,
+  `_reap_orphaned_subactors` — autouse cleanup
+  fixtures whose cumulative teardown overhead
+  contributes to the cascade time

ai/conc-anal/subint_cancel_delivery_hang_issue.md

@@ -0,0 +1,161 @@
+# `subint` backend: parent trio loop parks after subint teardown (Ctrl-C works; not a CPython-level issue)
+
+Follow-up to the Phase B subint spawn-backend PR (see
+`tractor.spawn._subint`, issue #379). Distinct from the
+`subint_sigint_starvation_issue.md` (SIGINT-unresponsive
+starvation hang): this one is **Ctrl-C-able**, which means
+it's *not* the shared-GIL-hostage class and is ours to fix
+from inside tractor rather than waiting on upstream CPython
+/ msgspec progress.
+
+## TL;DR
+
+After a stuck-subint subactor is torn down via the
+hard-kill path, a parent-side trio task parks on an
+*orphaned resource* (most likely a `chan.recv()` /
+`process_messages` loop on the now-dead subint's IPC
+channel) and waits forever for bytes that can't arrive —
+because the channel was torn down without emitting a clean
+EOF/`BrokenResourceError` to the waiting receiver.
+
+Unlike `subint_sigint_starvation_issue.md`, the main trio
+loop **is** iterating normally — SIGINT delivers cleanly
+and the test unhangs. But absent Ctrl-C, the test suite
+wedges indefinitely.
+
+## Symptom
+
+Running `test_subint_non_checkpointing_child` under
+`--spawn-backend=subint` (in
+`tests/test_subint_cancellation.py`):
+
+1. Test spawns a subactor whose main task runs
+   `threading.Event.wait(1.0)` in a loop — releases the
+   GIL but never inserts a trio checkpoint.
+2. Parent does `an.cancel_scope.cancel()`. Our
+   `subint_proc` cancel path fires: soft-kill sends
+   `Portal.cancel_actor()` over the live IPC channel →
+   subint's trio loop *should* process the cancel msg on
+   its IPC dispatcher task (since the GIL releases are
+   happening).
+3. Expected: subint's `trio.run()` unwinds, driver thread
+   exits naturally, parent returns.
+4. Actual: parent `trio.run()` never completes. Test
+   hangs past its `trio.fail_after()` deadline.
+
+## Evidence
+
+### `strace` on the hung pytest process during SIGINT
+
+```
+--- SIGINT {si_signo=SIGINT, si_code=SI_KERNEL} ---
+write(17, "\2", 1)                      = 1
+```
+
+Contrast with the SIGINT-starvation hang (see
+`subint_sigint_starvation_issue.md`) where that same
+`write()` returned `EAGAIN`. Here the SIGINT byte is
+written successfully → Python's signal handler pipe is
+being drained → main trio loop **is** iterating → SIGINT
+gets turned into `trio.Cancelled` → the test unhangs (if
+the operator happens to be there to hit Ctrl-C).
+
+### Stack dump (via `tractor.devx.dump_on_hang`)
+
+Single main thread visible, parked in
+`trio._core._io_epoll.get_events` inside `trio.run` at the
+test's `trio.run(...)` call site. No subint driver thread
+(subint was destroyed successfully — this is *after* the
+hard-kill path, not during it).
+
+## Root cause hypothesis
+
+Most consistent with the evidence: a parent-side trio
+task is awaiting a `chan.recv()` / `process_messages` loop
+on the dead subint's IPC channel. The sequence:
+
+1. Soft-kill in `subint_proc` sends `Portal.cancel_actor()`
+   over the channel. The subint's trio dispatcher *may* or
+   may not have processed the cancel msg before the subint
+   was destroyed — timing-dependent.
+2. Hard-kill timeout fires (because the subint's main
+   task was in `threading.Event.wait()` with no trio
+   checkpoint — cancel-msg processing couldn't race the
+   timeout).
+3. Driver thread abandoned, `_interpreters.destroy()`
+   runs. Subint is gone.
+4. But the parent-side trio task holding a
+   `chan.recv()` / `process_messages` loop against that
+   channel was **not** explicitly cancelled. The channel's
+   underlying socket got torn down, but without a clean
+   EOF delivered to the waiting recv, the task parks
+   forever on `trio.lowlevel.wait_readable` (or similar).
+
+This matches the "main loop fine, task parked on
+orphaned I/O" signature.
+
+## Why this is ours to fix (not CPython's)
+
+- Main trio loop iterates normally → GIL isn't starved.
+- SIGINT is deliverable → not a signal-pipe-full /
+  wakeup-fd contention scenario.
+- The hang is in *our* supervision code, specifically in
+  how `subint_proc` tears down its side of the IPC when
+  the subint is abandoned/destroyed.
+
+## Possible fix directions
+
+1. **Explicit parent-side channel abort on subint
+   abandon.** In `subint_proc`'s teardown block, after the
+   hard-kill timeout fires, explicitly close the parent's
+   end of the IPC channel to the subint. Any waiting
+   `chan.recv()` / `process_messages` task sees
+   `BrokenResourceError` (or `ClosedResourceError`) and
+   unwinds.
+2. **Cancel parent-side RPC tasks tied to the dead
+   subint's channel.** The `Actor._rpc_tasks` / nursery
+   machinery should have a handle on any
+   `process_messages` loops bound to a specific peer
+   channel. Iterate those and cancel explicitly.
+3. **Bound the top-level `await actor_nursery
+   ._join_procs.wait()` shield in `subint_proc`** (same
+   pattern as the other bounded shields the hard-kill
+   patch added). If the nursery never sets `_join_procs`
+   because a child task is parked, the bound would at
+   least let the teardown proceed.
+
+Of these, (1) is the most surgical and directly addresses
+the root cause. (2) is a defense-in-depth companion. (3)
+is a band-aid but cheap to add.
+
+## Current workaround
+
+None in-tree. The test's `trio.fail_after()` bound
+currently fires and raises `TooSlowError`, so the test
+visibly **fails** rather than hangs — which is
+intentional (an unbounded cancellation-audit test would
+defeat itself). But in interactive test runs the operator
+has to hit Ctrl-C to move past the parked state before
+pytest reports the failure.
+
+## Reproducer
+
+```
+./py314/bin/python -m pytest \
+  tests/test_subint_cancellation.py::test_subint_non_checkpointing_child \
+  --spawn-backend=subint --tb=short --no-header -v
+```
+
+Expected: hangs until `trio.fail_after(15)` fires, or
+Ctrl-C unwedges it manually.
+
+## References
+
+- `tractor.spawn._subint.subint_proc` — current subint
+  teardown code; see the `_HARD_KILL_TIMEOUT` bounded
+  shields + `daemon=True` driver-thread abandonment
+  (commit `b025c982`).
+- `ai/conc-anal/subint_sigint_starvation_issue.md` — the
+  sibling CPython-level hang (GIL-starvation,
+  SIGINT-unresponsive) which is **not** this issue.
+- Phase B tracking: issue #379.

ai/conc-anal/subint_fork_blocked_by_cpython_post_fork_issue.md

@@ -0,0 +1,337 @@
+# `os.fork()` from a non-main sub-interpreter aborts the child (CPython refuses post-fork cleanup)
+
+Third `subint`-class analysis in this project. Unlike its
+two siblings (`subint_sigint_starvation_issue.md`,
+`subint_cancel_delivery_hang_issue.md`), this one is not a
+hang — it's a **hard CPython-level refusal** of an
+experimental spawn strategy we wanted to try.
+
+## TL;DR
+
+An in-process sub-interpreter cannot be used as a
+"launchpad" for `os.fork()` on current CPython. The fork
+syscall succeeds in the parent, but the forked CHILD
+process is aborted immediately by CPython's post-fork
+cleanup with:
+
+```
+Fatal Python error: _PyInterpreterState_DeleteExceptMain: not main interpreter
+```
+
+This is enforced by a hard `PyStatus_ERR` gate in
+`Python/pystate.c`. The CPython devs acknowledge the
+fragility with an in-source comment (`// Ideally we could
+guarantee tstate is running main.`) but provide no
+mechanism to satisfy the precondition from user code.
+
+**Implication for tractor**: the `subint_fork` backend
+sketched in `tractor.spawn._subint_fork` is structurally
+dead on current CPython. The submodule is kept as
+documentation of the attempt; `--spawn-backend=subint_fork`
+raises `NotImplementedError` pointing here.
+
+## Context — why we tried this
+
+The motivation is issue #379's "Our own thoughts, ideas
+for `fork()`-workaround/hacks..." section. The existing
+trio-backend (`tractor.spawn._trio.trio_proc`) spawns
+subactors via `trio.lowlevel.open_process()` → ultimately
+`posix_spawn()` or `fork+exec`, from the parent's main
+interpreter that is currently running `trio.run()`. This
+brushes against a known-fragile interaction between
+`trio` and `fork()` tracked in
+[python-trio/trio#1614](https://github.com/python-trio/trio/issues/1614)
+and siblings — mostly mitigated in `tractor`'s case only
+incidentally (we `exec()` immediately post-fork).
+
+The idea was:
+
+1. Create a subint that has *never* imported `trio`.
+2. From a worker thread in that subint, call `os.fork()`.
+3. In the child, `execv()` back into
+   `python -m tractor._child` — same as `trio_proc` does.
+4. The fork is from a trio-free context → trio+fork
+   hazards avoided regardless of downstream behavior.
+
+The parent-side orchestration (`ipc_server.wait_for_peer`,
+`SpawnSpec`, `Portal` yield) would reuse
+`trio_proc`'s flow verbatim, with only the subproc-spawn
+mechanics swapped.
+
+## Symptom
+
+Running the prototype (`tractor.spawn._subint_fork.subint_fork_proc`,
+see git history prior to the stub revert) on py3.14:
+
+```
+Fatal Python error: _PyInterpreterState_DeleteExceptMain: not main interpreter
+Python runtime state: initialized
+
+Current thread 0x00007f6b71a456c0 [subint-fork-lau] (most recent call first):
+  File "<script>", line 2 in <module>
+<script>:2: DeprecationWarning: This process (pid=802985) is multi-threaded, use of fork() may lead to deadlocks in the child.
+```
+
+Key clues:
+
+- The **`DeprecationWarning`** fires in the parent (before
+  fork completes) — fork *is* executing, we get that far.
+- The **`Fatal Python error`** comes from the child — it
+  aborts during CPython's post-fork C initialization
+  before any user Python runs in the child.
+- The thread name `subint-fork-lau[nchpad]` is ours —
+  confirms the fork is being called from the launchpad
+  subint's driver thread.
+
+## CPython source walkthrough
+
+### Call site — `Modules/posixmodule.c:728-793`
+
+The post-fork-child hook CPython runs in the child process:
+
+```c
+void
+PyOS_AfterFork_Child(void)
+{
+    PyStatus status;
+    _PyRuntimeState *runtime = &_PyRuntime;
+
+    // re-creates runtime->interpreters.mutex (HEAD_UNLOCK)
+    status = _PyRuntimeState_ReInitThreads(runtime);
+    ...
+
+    PyThreadState *tstate = _PyThreadState_GET();
+    _Py_EnsureTstateNotNULL(tstate);
+
+    ...
+
+    // Ideally we could guarantee tstate is running main.   ← !!!
+    _PyInterpreterState_ReinitRunningMain(tstate);
+
+    status = _PyEval_ReInitThreads(tstate);
+    ...
+
+    status = _PyInterpreterState_DeleteExceptMain(runtime);
+    if (_PyStatus_EXCEPTION(status)) {
+        goto fatal_error;
+    }
+    ...
+
+fatal_error:
+    Py_ExitStatusException(status);
+}
+```
+
+The `// Ideally we could guarantee tstate is running
+main.` comment is a flashing warning sign — the CPython
+devs *know* this path is fragile when fork is called from
+a non-main subint, but they've chosen to abort rather than
+silently corrupt state. Arguably the right call.
+
+### The refusal — `Python/pystate.c:1035-1075`
+
+```c
+/*
+ * Delete all interpreter states except the main interpreter.  If there
+ * is a current interpreter state, it *must* be the main interpreter.
+ */
+PyStatus
+_PyInterpreterState_DeleteExceptMain(_PyRuntimeState *runtime)
+{
+    struct pyinterpreters *interpreters = &runtime->interpreters;
+
+    PyThreadState *tstate = _PyThreadState_Swap(runtime, NULL);
+    if (tstate != NULL && tstate->interp != interpreters->main) {
+        return _PyStatus_ERR("not main interpreter");       ← our error
+    }
+
+    HEAD_LOCK(runtime);
+    PyInterpreterState *interp = interpreters->head;
+    interpreters->head = NULL;
+    while (interp != NULL) {
+        if (interp == interpreters->main) {
+            interpreters->main->next = NULL;
+            interpreters->head = interp;
+            interp = interp->next;
+            continue;
+        }
+
+        // XXX Won't this fail since PyInterpreterState_Clear() requires
+        // the "current" tstate to be set?
+        PyInterpreterState_Clear(interp);  // XXX must activate?
+        zapthreads(interp);
+        ...
+    }
+    ...
+}
+```
+
+The comment in the docstring (`If there is a current
+interpreter state, it *must* be the main interpreter.`) is
+the formal API contract. The `XXX` comments further in
+suggest the CPython team is already aware this function
+has latent issues even in the happy path.
+
+## Chain summary
+
+1. Our launchpad subint's driver OS-thread calls
+   `os.fork()`.
+2. `fork()` succeeds. Child wakes up with:
+   - The parent's full memory image (including all
+     subints).
+   - Only the *calling* thread alive (the driver thread).
+   - `_PyThreadState_GET()` on that thread returns the
+     **launchpad subint's tstate**, *not* main's.
+3. CPython runs `PyOS_AfterFork_Child()`.
+4. It reaches `_PyInterpreterState_DeleteExceptMain()`.
+5. Gate check fails: `tstate->interp != interpreters->main`.
+6. `PyStatus_ERR("not main interpreter")` → `fatal_error`
+   goto → `Py_ExitStatusException()` → child aborts.
+
+Parent-side consequence: `os.fork()` in the subint
+bootstrap returned successfully with the child's PID, but
+the child died before connecting back. Our parent's
+`ipc_server.wait_for_peer(uid)` would hang forever — the
+child never gets to `_actor_child_main`.
+
+## Definitive answer to "Open Question 1"
+
+From the (now-stub) `subint_fork_proc` docstring:
+
+> Does CPython allow `os.fork()` from a non-main
+> sub-interpreter under the legacy config?
+
+**No.** Not in a usable-by-user-code sense. The fork
+syscall is not blocked, but the child cannot survive
+CPython's post-fork initialization. This is enforced, not
+accidental, and the CPython devs have acknowledged the
+fragility in-source.
+
+## What we'd need from CPython to unblock
+
+Any one of these, from least-to-most invasive:
+
+1. **A pre-fork hook mechanism** that lets user code (or
+   tractor itself via `os.register_at_fork(before=...)`)
+   swap the current tstate to main before fork runs. The
+   swap would need to work across the subint→main
+   boundary, which is the actual hard part —
+   `_PyThreadState_Swap()` exists but is internal.
+
+2. **A `_PyInterpreterState_DeleteExceptFor(tstate->interp)`
+   variant** that cleans up all *other* subints while
+   preserving the calling subint's state. Lets the child
+   continue executing in the subint after fork; a
+   subsequent `execv()` clears everything at the OS
+   level anyway.
+
+3. **A cleaner error** than `Fatal Python error` aborting
+   the child. Even without fixing the underlying
+   capability, a raised Python-level exception in the
+   parent's `fork()` call (rather than a silent child
+   abort) would at least make the failure mode
+   debuggable.
+
+## Upstream-report draft (for CPython issue tracker)
+
+### Title
+
+> `os.fork()` from a non-main sub-interpreter aborts the
+> child with a fatal error in `PyOS_AfterFork_Child`; can
+> we at least make it a clean `RuntimeError` in the
+> parent?
+
+### Body
+
+> **Version**: Python 3.14.x
+>
+> **Summary**: Calling `os.fork()` from a thread currently
+> executing inside a sub-interpreter causes the forked
+> child process to abort during CPython's post-fork
+> cleanup, with the following output in the child:
+>
+> ```
+> Fatal Python error: _PyInterpreterState_DeleteExceptMain: not main interpreter
+> ```
+>
+> From the **parent's** point of view the fork succeeded
+> (returned a valid child PID). The failure is completely
+> opaque to parent-side Python code — unless the parent
+> does `os.waitpid()` it won't even notice the child
+> died.
+>
+> **Root cause** (as I understand it from reading sources):
+> `Modules/posixmodule.c::PyOS_AfterFork_Child()` calls
+> `_PyInterpreterState_DeleteExceptMain()` with a
+> precondition that `_PyThreadState_GET()->interp` be the
+> main interpreter. When `fork()` is called from a thread
+> executing inside a subinterpreter, the child wakes up
+> with its tstate still pointing at the subint, and the
+> gate in `Python/pystate.c:1044-1047` fails.
+>
+> A comment in the source
+> (`Modules/posixmodule.c:753` — `// Ideally we could
+> guarantee tstate is running main.`) suggests this is a
+> known-fragile path rather than an intentional
+> invariant.
+>
+> **Use case**: I was experimenting with using a
+> sub-interpreter as a "fork launchpad" — have a subint
+> that has never imported `trio`, call `os.fork()` from
+> that subint's thread, and in the child `execv()` back
+> into a fresh Python interpreter process. The goal was
+> to sidestep known issues with `trio` + `fork()`
+> interaction (see
+> [python-trio/trio#1614](https://github.com/python-trio/trio/issues/1614))
+> by guaranteeing the forking context had never been
+> "contaminated" by trio's imports or globals. This
+> approach would allow `trio`-using applications to
+> combine `fork`-based subprocess spawning with
+> per-worker `trio.run()` runtimes — a fairly common
+> pattern that currently requires workarounds.
+>
+> **Request**:
+>
+> Ideally: make fork-from-subint work (e.g., by swapping
+> the caller's tstate to main in the pre-fork hook), or
+> provide a `_PyInterpreterState_DeleteExceptFor(interp)`
+> variant that permits the caller's subint to survive
+> post-fork so user code can subsequently `execv()`.
+>
+> Minimally: convert the fatal child-side abort into a
+> clean `RuntimeError` (or similar) raised in the
+> parent's `fork()` call. Even if the capability isn't
+> expanded, the failure mode should be debuggable by
+> user-code in the parent — right now it's a silent
+> child death with an error message buried in the
+> child's stderr that parent code can't programmatically
+> see.
+>
+> **Related**: PEP 684 (per-interpreter GIL), PEP 734
+> (`concurrent.interpreters` public API). The private
+> `_interpreters` module is what I used to create the
+> launchpad — behavior is the same whether using
+> `_interpreters.create('legacy')` or
+> `concurrent.interpreters.create()` (the latter was not
+> tested but the gate is identical).
+>
+> Happy to contribute a minimal reproducer + test case if
+> this is something the team wants to pursue.
+
+## References
+
+- `Modules/posixmodule.c:728` —
+  [`PyOS_AfterFork_Child`](https://github.com/python/cpython/blob/main/Modules/posixmodule.c#L728)
+- `Python/pystate.c:1040` —
+  [`_PyInterpreterState_DeleteExceptMain`](https://github.com/python/cpython/blob/main/Python/pystate.c#L1040)
+- PEP 684 (per-interpreter GIL):
+  <https://peps.python.org/pep-0684/>
+- PEP 734 (`concurrent.interpreters` public API):
+  <https://peps.python.org/pep-0734/>
+- [python-trio/trio#1614](https://github.com/python-trio/trio/issues/1614)
+  — the original motivation for the launchpad idea.
+- tractor issue #379 — "Our own thoughts, ideas for
+  `fork()`-workaround/hacks..." section where this was
+  first sketched.
+- `tractor.spawn._subint_fork` — in-tree stub preserving
+  the attempted impl's shape in git history.

ai/conc-anal/subint_fork_from_main_thread_smoketest.py

@@ -0,0 +1,375 @@
+#!/usr/bin/env python3
+'''
+Standalone CPython-level feasibility check for the "main-interp
+worker-thread forkserver + subint-hosted trio" architecture
+proposed as a workaround to the CPython-level refusal
+documented in
+`ai/conc-anal/subint_fork_blocked_by_cpython_post_fork_issue.md`.
+
+Purpose
+-------
+Deliberately NOT a `tractor` test. Zero `tractor` imports.
+Uses `_interpreters` (private stdlib) + `os.fork()` directly so
+the signal is unambiguous — pass/fail here is a property of
+CPython alone, independent of our runtime.
+
+Run each scenario in isolation; the child's fate is observable
+only via `os.waitpid()` of the parent and the scenario's own
+status prints.
+
+Scenarios (pick one with `--scenario <name>`)
+---------------------------------------------
+
+- `control_subint_thread_fork` — the KNOWN-BROKEN case we
+  documented in `subint_fork_blocked_by_cpython_post_fork_issue.md`:
+  drive a subint from a thread, call `os.fork()` inside its
+  `_interpreters.exec()`, watch the child abort. **Included as
+  a control** — if this scenario DOESN'T abort the child, our
+  analysis is wrong and we should re-check everything.
+
+- `main_thread_fork` — baseline sanity. Call `os.fork()` from
+  the process's main thread. Must always succeed; if this
+  fails something much bigger is broken.
+
+- `worker_thread_fork` — the architectural assertion. Spawn a
+  regular `threading.Thread` (attached to main interp, NOT a
+  subint), have IT call `os.fork()`. Child should survive
+  post-fork cleanup.
+
+- `full_architecture` — end-to-end: main-interp worker thread
+  forks. In the child, fork-thread (still main-interp) creates
+  a subint, drives a second worker thread inside it that runs
+  a trivial `trio.run()`. Validates the "root runtime lives in
+  a subint in the child" piece of the proposed arch.
+
+All scenarios print a self-contained pass/fail banner. Exit
+code 0 on expected outcome (which for `control_*` means "child
+aborted", not "child succeeded"!).
+
+Requires Python 3.14+.
+
+Usage
+-----
+::
+
+    python subint_fork_from_main_thread_smoketest.py \\
+        --scenario main_thread_fork
+
+    python subint_fork_from_main_thread_smoketest.py \\
+        --scenario full_architecture
+
+'''
+from __future__ import annotations
+import argparse
+import os
+import sys
+import threading
+import time
+
+
+# Hard-require py3.14 for the public `concurrent.interpreters`
+# API (we still drop to `_interpreters` internally, same as
+# `tractor.spawn._subint`).
+try:
+    from concurrent import interpreters as _public_interpreters  # noqa: F401
+    import _interpreters  # type: ignore
+except ImportError:
+    print(
+        'FAIL (setup): requires Python 3.14+ '
+        '(missing `concurrent.interpreters`)',
+        file=sys.stderr,
+    )
+    sys.exit(2)
+
+
+# The actual primitives this script exercises live in
+# `tractor.spawn._subint_forkserver` — we re-import them here
+# rather than inlining so the module and the validation stay
+# in sync. (Early versions of this file had them inline for
+# the "zero tractor imports" isolation guarantee; now that
+# CPython-level feasibility is confirmed, the validated
+# primitives have moved into tractor proper.)
+from tractor.spawn._main_thread_forkserver import (
+    fork_from_worker_thread,
+    wait_child,
+)
+from tractor.spawn._subint_forkserver import (
+    run_subint_in_worker_thread,
+)
+
+
+# ----------------------------------------------------------------
+# small observability helpers (test-harness only)
+# ----------------------------------------------------------------
+
+
+def _banner(title: str) -> None:
+    line = '=' * 60
+    print(f'\n{line}\n{title}\n{line}', flush=True)
+
+
+def _report(
+    label: str,
+    *,
+    ok: bool,
+    status_str: str,
+    expect_exit_ok: bool,
+) -> None:
+    verdict: str = 'PASS' if ok else 'FAIL'
+    expected_str: str = (
+        'normal exit (rc=0)'
+        if expect_exit_ok
+        else 'abnormal death (signal or nonzero exit)'
+    )
+    print(
+        f'[{verdict}] {label}: '
+        f'expected {expected_str}; observed {status_str}',
+        flush=True,
+    )
+
+
+# ----------------------------------------------------------------
+# scenario: `control_subint_thread_fork` (known-broken)
+# ----------------------------------------------------------------
+
+
+def scenario_control_subint_thread_fork() -> int:
+    _banner(
+        '[control] fork from INSIDE a subint (expected: child aborts)'
+    )
+    interp_id = _interpreters.create('legacy')
+    print(f'  created subint {interp_id}', flush=True)
+
+    # Shared flag: child writes a sentinel file we can detect from
+    # the parent. If the child manages to write this, CPython's
+    # post-fork refusal is NOT happening → analysis is wrong.
+    sentinel = '/tmp/subint_fork_smoketest_control_child_ran'
+    try:
+        os.unlink(sentinel)
+    except FileNotFoundError:
+        pass
+
+    bootstrap = (
+        'import os\n'
+        'pid = os.fork()\n'
+        'if pid == 0:\n'
+        # child — if CPython's refusal fires this code never runs
+        f'    with open({sentinel!r}, "w") as f:\n'
+        '        f.write("ran")\n'
+        '    os._exit(0)\n'
+        'else:\n'
+        # parent side (inside the launchpad subint) — stash the
+        # forked PID on a shareable dict so we can waitpid()
+        # from the outer main interp. We can't just return it;
+        # _interpreters.exec() returns nothing useful.
+        '    import builtins\n'
+        '    builtins._forked_child_pid = pid\n'
+    )
+
+    # NOTE, we can't easily pull state back from the subint.
+    # For the CONTROL scenario we just time-bound the fork +
+    # check the sentinel. If sentinel exists → child ran →
+    # analysis wrong. If not → child aborted → analysis
+    # confirmed.
+    done = threading.Event()
+
+    def _drive() -> None:
+        try:
+            _interpreters.exec(interp_id, bootstrap)
+        except Exception as err:
+            print(
+                f'  subint bootstrap raised (expected on some '
+                f'CPython versions): {type(err).__name__}: {err}',
+                flush=True,
+            )
+        finally:
+            done.set()
+
+    t = threading.Thread(
+        target=_drive,
+        name='control-subint-fork-launchpad',
+        daemon=True,
+    )
+    t.start()
+    done.wait(timeout=5.0)
+    t.join(timeout=2.0)
+
+    # Give the (possibly-aborted) child a moment to die.
+    time.sleep(0.5)
+
+    sentinel_present = os.path.exists(sentinel)
+    verdict = (
+        # "PASS" for our analysis means sentinel NOT present.
+        'PASS' if not sentinel_present else 'FAIL (UNEXPECTED)'
+    )
+    print(
+        f'[{verdict}] control: sentinel present={sentinel_present} '
+        f'(analysis predicts False — child should abort before '
+        f'writing)',
+        flush=True,
+    )
+    if sentinel_present:
+        os.unlink(sentinel)
+
+    try:
+        _interpreters.destroy(interp_id)
+    except _interpreters.InterpreterError:
+        pass
+
+    return 0 if not sentinel_present else 1
+
+
+# ----------------------------------------------------------------
+# scenario: `main_thread_fork` (baseline sanity)
+# ----------------------------------------------------------------
+
+
+def scenario_main_thread_fork() -> int:
+    _banner(
+        '[baseline] fork from MAIN thread (expected: child exits normally)'
+    )
+
+    pid = os.fork()
+    if pid == 0:
+        os._exit(0)
+
+    return 0 if _wait_child(
+        pid,
+        label='main_thread_fork',
+        expect_exit_ok=True,
+    ) else 1
+
+
+# ----------------------------------------------------------------
+# scenario: `worker_thread_fork` (architectural assertion)
+# ----------------------------------------------------------------
+
+
+def _run_worker_thread_fork_scenario(
+    label: str,
+    *,
+    child_target=None,
+) -> int:
+    '''
+    Thin wrapper: delegate the actual fork to the
+    `tractor.spawn._subint_forkserver` primitive, then wait
+    on the child and render a pass/fail banner.
+
+    '''
+    try:
+        pid: int = fork_from_worker_thread(
+            child_target=child_target,
+            thread_name=f'worker-fork-thread[{label}]',
+        )
+    except RuntimeError as err:
+        print(f'[FAIL] {label}: {err}', flush=True)
+        return 1
+    print(f'  forked child pid={pid}', flush=True)
+    ok, status_str = wait_child(pid, expect_exit_ok=True)
+    _report(
+        label,
+        ok=ok,
+        status_str=status_str,
+        expect_exit_ok=True,
+    )
+    return 0 if ok else 1
+
+
+def scenario_worker_thread_fork() -> int:
+    _banner(
+        '[arch] fork from MAIN-INTERP WORKER thread '
+        '(expected: child exits normally — this is the one '
+        'that matters)'
+    )
+    return _run_worker_thread_fork_scenario(
+        'worker_thread_fork',
+    )
+
+
+# ----------------------------------------------------------------
+# scenario: `full_architecture`
+# ----------------------------------------------------------------
+
+
+_CHILD_TRIO_BOOTSTRAP: str = (
+    'import trio\n'
+    'async def _main():\n'
+    '    await trio.sleep(0.05)\n'
+    '    return 42\n'
+    'result = trio.run(_main)\n'
+    'assert result == 42, f"trio.run returned {result}"\n'
+    'print("  CHILD subint: trio.run OK, result=42", '
+    'flush=True)\n'
+)
+
+
+def _child_trio_in_subint() -> int:
+    '''
+    CHILD-side `child_target`: drive a trivial `trio.run()`
+    inside a fresh legacy-config subint on a worker thread,
+    using the `tractor.spawn._subint_forkserver.run_subint_in_worker_thread`
+    primitive. Returns 0 on success.
+
+    '''
+    try:
+        run_subint_in_worker_thread(
+            _CHILD_TRIO_BOOTSTRAP,
+            thread_name='child-subint-trio-thread',
+        )
+    except RuntimeError as err:
+        print(
+            f'  CHILD: run_subint_in_worker_thread timed out / thread '
+            f'never returned: {err}',
+            flush=True,
+        )
+        return 3
+    except BaseException as err:
+        print(
+            f'  CHILD: subint bootstrap raised: '
+            f'{type(err).__name__}: {err}',
+            flush=True,
+        )
+        return 4
+    return 0
+
+
+def scenario_full_architecture() -> int:
+    _banner(
+        '[arch-full] worker-thread fork + child runs trio in a '
+        'subint (end-to-end proposed arch)'
+    )
+    return _run_worker_thread_fork_scenario(
+        'full_architecture',
+        child_target=_child_trio_in_subint,
+    )
+
+
+# ----------------------------------------------------------------
+# main
+# ----------------------------------------------------------------
+
+
+SCENARIOS: dict[str, Callable[[], int]] = {
+    'control_subint_thread_fork': scenario_control_subint_thread_fork,
+    'main_thread_fork': scenario_main_thread_fork,
+    'worker_thread_fork': scenario_worker_thread_fork,
+    'full_architecture': scenario_full_architecture,
+}
+
+
+def main() -> int:
+    ap = argparse.ArgumentParser(
+        description=__doc__,
+        formatter_class=argparse.RawDescriptionHelpFormatter,
+    )
+    ap.add_argument(
+        '--scenario',
+        choices=sorted(SCENARIOS.keys()),
+        required=True,
+    )
+    args = ap.parse_args()
+    return SCENARIOS[args.scenario]()
+
+
+if __name__ == '__main__':
+    sys.exit(main())

ai/conc-anal/subint_forkserver_orphan_sigint_hang_issue.md

@@ -0,0 +1,385 @@
+# `subint_forkserver` backend: orphaned-subactor SIGINT wedged in `epoll_wait`
+
+Follow-up to the Phase C `subint_forkserver` spawn-backend
+PR (see `tractor.spawn._subint_forkserver`, issue #379).
+Surfaced by the xfail'd
+`tests/spawn/test_subint_forkserver.py::test_orphaned_subactor_sigint_cleanup_DRAFT`.
+
+Related-but-distinct from
+`subint_cancel_delivery_hang_issue.md` (orphaned-channel
+park AFTER subint teardown) and
+`subint_sigint_starvation_issue.md` (GIL-starvation,
+SIGINT never delivered): here the SIGINT IS delivered,
+trio's handler IS installed, but trio's event loop never
+wakes — so the KBI-at-checkpoint → `_trio_main` catch path
+(which is the runtime's *intentional* OS-cancel design)
+never fires.
+
+## TL;DR
+
+When a `subint_forkserver`-spawned subactor is orphaned
+(parent `SIGKILL`'d, no IPC cancel path available) and then
+externally `SIGINT`'d, the subactor hangs in
+`trio/_core/_io_epoll.py::get_events` (epoll_wait)
+indefinitely — even though:
+
+1. `threading.current_thread() is threading.main_thread()`
+   post-fork (CPython 3.14 re-designates correctly).
+2. Trio's SIGINT handler IS installed in the subactor
+   (`signal.getsignal(SIGINT)` returns
+   `<function KIManager.install.<locals>.handler at 0x...>`).
+3. The kernel does deliver SIGINT — the signal arrives at
+   the only thread in the process (the fork-inherited
+   worker which IS now "main" per Python).
+
+Yet `epoll_wait` does not return. Trio's wakeup-fd mechanism
+— the machinery that turns SIGINT into an epoll-wake — is
+somehow not firing the wakeup. Until that's fixed, the
+intentional "KBI-as-OS-cancel" path in
+`tractor/spawn/_entry.py::_trio_main:164` is unreachable
+for forkserver-spawned subactors whose parent dies.
+
+## Symptom
+
+Test: `tests/spawn/test_subint_forkserver.py::test_orphaned_subactor_sigint_cleanup_DRAFT`
+(currently marked `@pytest.mark.xfail(strict=True)`).
+
+1. Harness subprocess brings up a tractor root actor +
+   one `run_in_actor(_sleep_forever)` subactor via
+   `try_set_start_method('subint_forkserver')`.
+2. Harness prints `CHILD_PID` (subactor) and
+   `PARENT_READY` (root actor) markers to stdout.
+3. Test `os.kill(parent_pid, SIGKILL)` + `proc.wait()`
+   to fully reap the root-actor harness.
+4. Child (now reparented to pid 1) is still alive.
+5. Test `os.kill(child_pid, SIGINT)` and polls
+   `os.kill(child_pid, 0)` for up to 10s.
+6. **Observed**: the child is still alive at deadline —
+   SIGINT did not unwedge the trio loop.
+
+## What the "intentional" cancel path IS
+
+`tractor/spawn/_entry.py::_trio_main:157-186` —
+
+```python
+try:
+    if infect_asyncio:
+        actor._infected_aio = True
+        run_as_asyncio_guest(trio_main)
+    else:
+        trio.run(trio_main)
+
+except KeyboardInterrupt:
+    logmeth = log.cancel
+    exit_status: str = (
+        'Actor received KBI (aka an OS-cancel)\n'
+        ...
+    )
+```
+
+The "KBI == OS-cancel" mapping IS the runtime's
+deliberate, documented design. An OS-level SIGINT should
+flow as: kernel → trio handler → KBI at trio checkpoint
+→ unwinds `async_main` → surfaces at `_trio_main`'s
+`except KeyboardInterrupt:` → `log.cancel` + clean `rc=0`.
+
+**So fixing this hang is not "add a new SIGINT behavior" —
+it's "make the existing designed behavior actually fire in
+this backend config".** That's why option (B) ("fix root
+cause") is aligned with existing design intent, not a
+scope expansion.
+
+## Evidence
+
+### Positive control: standalone fork-from-worker + `trio.run(sleep_forever)` + SIGINT WORKS
+
+```python
+import os, signal, time, trio
+from tractor.spawn._subint_forkserver import (
+    fork_from_worker_thread, wait_child,
+)
+
+def child_target() -> int:
+    async def _main():
+        try:
+            await trio.sleep_forever()
+        except KeyboardInterrupt:
+            print('CHILD: caught KBI — trio SIGINT works!')
+            return
+    trio.run(_main)
+    return 0
+
+pid = fork_from_worker_thread(child_target, thread_name='trio-sigint-test')
+time.sleep(1.0)
+os.kill(pid, signal.SIGINT)
+wait_child(pid)
+```
+
+Result: `CHILD: caught KBI — trio SIGINT works!` + clean
+exit. So the fork-child + trio signal plumbing IS healthy
+in isolation. The hang appears only with the full tractor
+subactor runtime on top.
+
+### Negative test: full tractor subactor + orphan-SIGINT
+
+Equivalent to the xfail test. Traceback dump via
+`faulthandler.register(SIGUSR1, all_threads=True)` at the
+stuck moment:
+
+```
+Current thread 0x00007... [subint-forkserv] (most recent call first):
+  File ".../trio/_core/_io_epoll.py", line 245 in get_events
+  File ".../trio/_core/_run.py", line 2415 in run
+  File "tractor/spawn/_entry.py", line 162 in _trio_main
+  File "tractor/_child.py", line 72 in _actor_child_main
+  File "tractor/spawn/_subint_forkserver.py", line 650 in _child_target
+  File "tractor/spawn/_subint_forkserver.py", line 308 in _worker
+  File ".../threading.py", line 1024 in run
+```
+
+### Thread + signal-mask inventory of the stuck subactor
+
+Single thread (`tid == pid`, comm `'subint-forkserv'`,
+which IS `threading.main_thread()` post-fork):
+
+```
+SigBlk:  0000000000000000  # nothing blocked
+SigIgn:  0000000001001000  # SIGPIPE etc (Python defaults)
+SigCgt:  0000000108000202  # bit 1 = SIGINT caught
+```
+
+Bit 1 set in `SigCgt` → SIGINT handler IS installed. So
+trio's handler IS in place at the kernel level — not a
+"handler missing" situation.
+
+### Handler identity
+
+Inside the subactor's RPC body, `signal.getsignal(SIGINT)`
+returns `<function KIManager.install.<locals>.handler at
+0x...>` — trio's own `KIManager` handler. tractor's only
+SIGINT touches are `signal.getsignal()` *reads* (to stash
+into `debug.DebugStatus._trio_handler`); nothing writes
+over trio's handler outside the debug-REPL shielding path
+(`devx/debug/_tty_lock.py::shield_sigint`) which isn't
+engaged here (no debug_mode).
+
+## Ruled out
+
+- **GIL starvation / signal-pipe-full** (class A,
+  `subint_sigint_starvation_issue.md`): subactor runs on
+  its own GIL (separate OS process), not sharing with the
+  parent → no cross-process GIL contention. And `strace`-
+  equivalent in the signal mask shows SIGINT IS caught,
+  not queued.
+- **Orphaned channel park** (`subint_cancel_delivery_hang_issue.md`):
+  different failure mode — that one has trio iterating
+  normally and getting wedged on an orphaned
+  `chan.recv()` AFTER teardown. Here trio's event loop
+  itself never wakes.
+- **Tractor explicitly catching + swallowing KBI**:
+  greppable — the one `except KeyboardInterrupt:` in the
+  runtime is the INTENTIONAL cancel-path catch at
+  `_trio_main:164`. `async_main` uses `except Exception`
+  (not BaseException), so KBI should propagate through
+  cleanly if it ever fires.
+- **Missing `signal.set_wakeup_fd` (main-thread
+  restriction)**: post-fork, the fork-worker thread IS
+  `threading.main_thread()`, so trio's main-thread check
+  passes and its wakeup-fd install should succeed.
+
+## Root cause hypothesis (unverified)
+
+The SIGINT handler fires but trio's wakeup-fd write does
+not wake `epoll_wait`. Candidate causes, ranked by
+plausibility:
+
+1. **Wakeup-fd lifecycle race around tractor IPC setup.**
+   `async_main` spins up an IPC server + `process_messages`
+   loops early. Somewhere in that path the wakeup-fd that
+   trio registered with its epoll instance may be
+   closed/replaced/clobbered, so subsequent SIGINT writes
+   land on an fd that's no longer in the epoll set.
+   Evidence needed: compare
+   `signal.set_wakeup_fd(-1)` return value inside a
+   post-tractor-bringup RPC body vs. a pre-bringup
+   equivalent. If they differ, that's it.
+2. **Shielded cancel scope around `process_messages`.**
+   The RPC message loop is likely wrapped in a trio cancel
+   scope; if that scope is `shield=True` at any outer
+   layer, KBI scheduled at a checkpoint could be absorbed
+   by the shield and never bubble out to `_trio_main`.
+3. **Pre-fork wakeup-fd inheritance.** trio in the PARENT
+   process registered a wakeup-fd with its own epoll. The
+   child inherits the fd number but not the parent's
+   epoll instance — if tractor/trio re-uses the parent's
+   stale fd number anywhere, writes would go to a no-op
+   fd. (This is the least likely — `trio.run()` on the
+   child calls `KIManager.install` which should install a
+   fresh wakeup-fd from scratch.)
+
+## Cross-backend scope question
+
+**Untested**: does the same orphan-SIGINT hang reproduce
+against the `trio_proc` backend (stock subprocess + exec)?
+If yes → pre-existing tractor bug, independent of
+`subint_forkserver`. If no → something specific to the
+fork-from-worker path (e.g. inherited fds, mid-epoll-setup
+interference).
+
+**Quick repro for trio_proc**:
+
+```python
+# save as /tmp/trio_proc_orphan_sigint_repro.py
+import os, sys, signal, time, glob
+import subprocess as sp
+
+SCRIPT = '''
+import os, sys, trio, tractor
+async def _sleep_forever():
+    print(f"CHILD_PID={os.getpid()}", flush=True)
+    await trio.sleep_forever()
+
+async def _main():
+    async with (
+        tractor.open_root_actor(registry_addrs=[("127.0.0.1", 12350)]),
+        tractor.open_nursery() as an,
+    ):
+        await an.run_in_actor(_sleep_forever, name="sf-child")
+        print(f"PARENT_READY={os.getpid()}", flush=True)
+        await trio.sleep_forever()
+
+trio.run(_main)
+'''
+
+proc = sp.Popen(
+    [sys.executable, '-c', SCRIPT],
+    stdout=sp.PIPE, stderr=sp.STDOUT,
+)
+# parse CHILD_PID + PARENT_READY off proc.stdout ...
+# SIGKILL parent, SIGINT child, poll.
+```
+
+If that hangs too, open a broader issue; if not, this is
+`subint_forkserver`-specific (likely fd-inheritance-related).
+
+## Why this is ours to fix (not CPython's)
+
+- Signal IS delivered (`SigCgt` bitmask confirms).
+- Handler IS installed (trio's `KIManager`).
+- Thread identity is correct post-fork.
+- `_trio_main` already has the intentional KBI→clean-exit
+  path waiting to fire.
+
+Every CPython-level precondition is met. Something in
+tractor's runtime or trio's integration with it is
+breaking the SIGINT→wakeup→event-loop-wake pipeline.
+
+## Possible fix directions
+
+1. **Audit the wakeup-fd across tractor's IPC bringup.**
+   Add a trio startup hook that captures
+   `signal.set_wakeup_fd(-1)` at `_trio_main` entry,
+   after `async_main` enters, and periodically — assert
+   it's unchanged. If it moves, track down the writer.
+2. **Explicit `signal.set_wakeup_fd` reset after IPC
+   setup.** Brute force: re-install a fresh wakeup-fd
+   mid-bringup. Band-aid, but fast to try.
+3. **Ensure no `shield=True` cancel scope envelopes the
+   RPC-message-loop / IPC-server task.** If one does,
+   KBI-at-checkpoint never escapes.
+4. **Once fixed, the `child_sigint='trio'` mode on
+   `subint_forkserver_proc`** becomes effectively a no-op
+   or a doc-only mode — trio's natural handler already
+   does the right thing. Might end up removing the flag
+   entirely if there's no behavioral difference between
+   modes.
+
+## Current workaround
+
+None; `child_sigint` defaults to `'ipc'` (IPC cancel is
+the only reliable cancel path today), and the xfail test
+documents the gap. Operators hitting orphan-SIGINT get a
+hung process that needs `SIGKILL`.
+
+## Reproducer
+
+Inline, standalone (no pytest):
+
+```python
+# save as /tmp/orphan_sigint_repro.py  (py3.14+)
+import os, sys, signal, time, glob, trio
+import tractor
+from tractor.spawn._subint_forkserver import (
+    fork_from_worker_thread,
+)
+
+async def _sleep_forever():
+    print(f'SUBACTOR[{os.getpid()}]', flush=True)
+    await trio.sleep_forever()
+
+async def _main():
+    async with (
+        tractor.open_root_actor(
+            registry_addrs=[('127.0.0.1', 12349)],
+        ),
+        tractor.open_nursery() as an,
+    ):
+        await an.run_in_actor(_sleep_forever, name='sf-child')
+        await trio.sleep_forever()
+
+def child_target() -> int:
+    from tractor.spawn._spawn import try_set_start_method
+    try_set_start_method('subint_forkserver')
+    trio.run(_main)
+    return 0
+
+pid = fork_from_worker_thread(child_target, thread_name='repro')
+time.sleep(3.0)
+
+# find the subactor pid via /proc
+children = []
+for path in glob.glob(f'/proc/{pid}/task/*/children'):
+    with open(path) as f:
+        children.extend(int(x) for x in f.read().split() if x)
+subactor_pid = children[0]
+
+# SIGKILL root → orphan the subactor
+os.kill(pid, signal.SIGKILL)
+os.waitpid(pid, 0)
+time.sleep(0.3)
+
+# SIGINT the orphan — should cause clean trio exit
+os.kill(subactor_pid, signal.SIGINT)
+
+# poll for exit
+for _ in range(100):
+    try:
+        os.kill(subactor_pid, 0)
+        time.sleep(0.1)
+    except ProcessLookupError:
+        print('HARNESS: subactor exited cleanly ✔')
+        sys.exit(0)
+os.kill(subactor_pid, signal.SIGKILL)
+print('HARNESS: subactor hung — reproduced')
+sys.exit(1)
+```
+
+Expected (current): `HARNESS: subactor hung — reproduced`.
+
+After fix: `HARNESS: subactor exited cleanly ✔`.
+
+## References
+
+- `tractor/spawn/_entry.py::_trio_main:157-186` — the
+  intentional KBI→clean-exit path this bug makes
+  unreachable.
+- `tractor/spawn/_subint_forkserver` — the backend whose
+  orphan cancel-robustness this blocks.
+- `tests/spawn/test_subint_forkserver.py::test_orphaned_subactor_sigint_cleanup_DRAFT`
+  — the xfail'd reproducer in the test suite.
+- `ai/conc-anal/subint_cancel_delivery_hang_issue.md` —
+  sibling "orphaned channel park" hang (different class).
+- `ai/conc-anal/subint_sigint_starvation_issue.md` —
+  sibling "GIL starvation SIGINT drop" hang (different
+  class).
+- tractor issue #379 — subint backend tracking.

ai/conc-anal/subint_forkserver_test_cancellation_leak_issue.md

@@ -0,0 +1,851 @@
+# `subint_forkserver` backend: `test_cancellation.py` multi-level cancel cascade hang
+
+> **Tracked at:** [#449](https://github.com/goodboy/tractor/issues/449)
+
+Follow-up tracker: surfaced while wiring the new
+`subint_forkserver` spawn backend into the full tractor
+test matrix (step 2 of the post-backend-lands plan).
+See also
+`ai/conc-anal/subint_forkserver_orphan_sigint_hang_issue.md`
+— sibling tracker for a different forkserver-teardown
+class which probably shares the same fundamental root
+cause (fork-FD-inheritance across nested spawns).
+
+## TL;DR
+
+`tests/test_cancellation.py::test_nested_multierrors[subint_forkserver]`
+hangs indefinitely under our new backend. The hang is
+**inside the graceful IPC cancel cascade** — every actor
+in the multi-level tree parks in `epoll_wait` waiting
+for IPC messages that never arrive. Not a hard-kill /
+tree-reap issue (we don't reach the hard-kill fallback
+path at all).
+
+Working hypothesis (unverified): **`os.fork()` from a
+subactor inherits the root parent's IPC listener socket
+FDs**. When a first-level subactor forkserver-spawns a
+grandchild, that grandchild inherits both its direct
+spawner's FDs AND the root's FDs — IPC message routing
+becomes ambiguous (or silently sends to the wrong
+channel), so the cancel cascade can't reach its target.
+
+## Corrected diagnosis vs. earlier draft
+
+An earlier version of this doc claimed the root cause
+was **"forkserver teardown doesn't tree-kill
+descendants"** (SIGKILL only reaches the direct child,
+grandchildren survive and hold TCP `:1616`). That
+diagnosis was **wrong**, caused by conflating two
+observations:
+
+1. *5-zombie leak holding :1616* — happened in my own
+   workflow when I aborted a bg pytest task with
+   `pkill` (SIGTERM/SIGKILL, not SIGINT). The abrupt
+   kill skipped the graceful `ActorNursery.__aexit__`
+   cancel cascade entirely, orphaning descendants to
+   init. **This was my cleanup bug, not a forkserver
+   teardown bug.** Codified the fix (SIGINT-first +
+   bounded wait before SIGKILL) in
+   `feedback_sc_graceful_cancel_first.md` +
+   `.claude/skills/run-tests/SKILL.md`.
+2. *`test_nested_multierrors` hangs indefinitely* —
+   the real, separate, forkserver-specific bug
+   captured by this doc.
+
+The two symptoms are unrelated. The tree-kill / setpgrp
+fix direction proposed earlier would not help (1) (SC-
+graceful-cleanup is the right answer there) and would
+not help (2) (the hang is in the cancel cascade, not
+in the hard-kill fallback).
+
+## Symptom
+
+Reproducer (py3.14, clean env):
+
+```sh
+# preflight: ensure clean env
+ss -tlnp 2>/dev/null | grep ':1616' && echo 'FOUL — cleanup first!' || echo 'clean'
+
+./py314/bin/python -m pytest --spawn-backend=subint_forkserver \
+  'tests/test_cancellation.py::test_nested_multierrors[subint_forkserver]' \
+  --timeout=30 --timeout-method=thread --tb=short -v
+```
+
+Expected: `pytest-timeout` fires at 30s with a thread-
+dump banner, but the process itself **remains alive
+after timeout** and doesn't unwedge on subsequent
+SIGINT. Requires SIGKILL to reap.
+
+## Evidence (tree structure at hang point)
+
+All 5 processes are kernel-level `S` (sleeping) in
+`do_epoll_wait` (trio's event loop waiting on I/O):
+
+```
+PID     PPID    THREADS  NAME             ROLE
+333986  1       2        subint-forkserv  pytest main (the test body)
+333993  333986  3        subint-forkserv  "child 1" spawner subactor
+  334003 333993 1        subint-forkserv  grandchild errorer under child-1
+  334014 333993 1        subint-forkserv  grandchild errorer under child-1
+333999  333986  1        subint-forkserv  "child 2" spawner subactor (NO grandchildren!)
+```
+
+### Asymmetric tree depth
+
+The test's `spawn_and_error(breadth=2, depth=3)` should
+have BOTH direct children spawning 2 grandchildren
+each, going 3 levels deep. Reality:
+
+- Child 1 (333993, 3 threads) DID spawn its two
+  grandchildren as expected — fully booted trio
+  runtime.
+- Child 2 (333999, 1 thread) did NOT spawn any
+  grandchildren — clearly never completed its
+  nursery's first `run_in_actor`. Its 1-thread state
+  suggests the runtime never fully booted (no trio
+  worker threads for `waitpid`/IPC).
+
+This asymmetry is the key clue: the two direct
+children started identically but diverged. Probably a
+race around fork-inherited state (listener FDs,
+subactor-nursery channel state) that happens to land
+differently depending on spawn ordering.
+
+### Parent-side state
+
+Thread-dump of pytest main (333986) at the hang:
+
+- Main trio thread — parked in
+  `trio._core._io_epoll.get_events` (epoll_wait on
+  its event loop). Waiting for IPC from children.
+- Two trio-cache worker threads — each parked in
+  `outcome.capture(sync_fn)` calling
+  `os.waitpid(child_pid, 0)`. These are our
+  `_ForkedProc.wait()` off-loads. They're waiting for
+  the direct children to exit — but children are
+  stuck in their own epoll_wait waiting for IPC from
+  the parent.
+
+**It's a deadlock, not a leak:** the parent is
+correctly running `soft_kill(proc, _ForkedProc.wait,
+portal)` (graceful IPC cancel via
+`Portal.cancel_actor()`), but the children never
+acknowledge the cancel message (or the message never
+reaches them through the tangled post-fork IPC).
+
+## What's NOT the cause (ruled out)
+
+- **`_ForkedProc.kill()` only SIGKILLs direct pid /
+  missing tree-kill**: doesn't apply — we never reach
+  the hard-kill path. The deadlock is in the graceful
+  cancel cascade.
+- **Port `:1616` contention**: ruled out after the
+  `reg_addr` fixture-wiring fix; each test session
+  gets a unique port now.
+- **GIL starvation / SIGINT pipe filling** (class-A,
+  `subint_sigint_starvation_issue.md`): doesn't apply
+  — each subactor is its own OS process with its own
+  GIL (not legacy-config subint).
+- **Child-side `_trio_main` absorbing KBI**: grep
+  confirmed; `_trio_main` only catches KBI at the
+  `trio.run()` callsite, which is reached only if the
+  trio loop exits normally. The children here never
+  exit trio.run() — they're wedged inside.
+
+## Hypothesis: FD inheritance across nested forks
+
+`subint_forkserver_proc` calls
+`fork_from_worker_thread()` which ultimately does
+`os.fork()` from a dedicated worker thread. Standard
+Linux/POSIX fork semantics: **the child inherits ALL
+open FDs from the parent**, including listener
+sockets, epoll fds, trio wakeup pipes, and the
+parent's IPC channel sockets.
+
+At root-actor fork-spawn time, the root's IPC server
+listener FDs are open in the parent. Those get
+inherited by child 1. Child 1 then forkserver-spawns
+its OWN subactor (grandchild). The grandchild
+inherits FDs from child 1 — but child 1's address
+space still contains **the root's IPC listener FDs
+too** (inherited at first fork). So the grandchild
+has THREE sets of FDs:
+
+1. Its own (created after becoming a subactor).
+2. Its direct parent child-1's.
+3. The ROOT's (grandparent's) — inherited transitively.
+
+IPC message routing may be ambiguous in this tangled
+state. Or a listener socket that the root thinks it
+owns is actually open in multiple processes, and
+messages sent to it go to an arbitrary one. That
+would exactly match the observed "graceful cancel
+never propagates".
+
+This hypothesis predicts the bug **scales with fork
+depth**: single-level forkserver spawn
+(`test_subint_forkserver_spawn_basic`) works
+perfectly, but any test that spawns a second level
+deadlocks. Matches observations so far.
+
+## Fix directions (to validate)
+
+### 1. `close_fds=True` equivalent in `fork_from_worker_thread()`
+
+`subprocess.Popen` / `trio.lowlevel.open_process` have
+`close_fds=True` by default on POSIX — they
+enumerate open FDs in the child post-fork and close
+everything except stdio + any explicitly-passed FDs.
+Our raw `os.fork()` doesn't. Adding the equivalent to
+our `_worker` prelude would isolate each fork
+generation's FD set.
+
+Implementation sketch in
+`tractor.spawn._subint_forkserver.fork_from_worker_thread._worker`:
+
+```python
+def _worker() -> None:
+    pid: int = os.fork()
+    if pid == 0:
+        # CHILD: close inherited FDs except stdio + the
+        # pid-pipe we just opened.
+        keep: set[int] = {0, 1, 2, rfd, wfd}
+        import resource
+        soft, _ = resource.getrlimit(resource.RLIMIT_NOFILE)
+        os.closerange(3, soft)  # blunt; or enumerate /proc/self/fd
+        # ... then child_target() as before
+```
+
+Problem: overly aggressive — closes FDs the
+grandchild might legitimately need (e.g. its parent's
+IPC channel for the spawn-spec handshake, if we rely
+on that). Needs thought about which FDs are
+"inheritable and safe" vs. "inherited by accident".
+
+### 2. Cloexec on tractor's own FDs
+
+Set `FD_CLOEXEC` on tractor-created sockets (listener
+sockets, IPC channel sockets, pipes). This flag
+causes automatic close on `execve`, but since we
+`fork()` without `exec()`, this alone doesn't help.
+BUT — combined with a child-side explicit close-
+non-cloexec loop, it gives us a way to mark "my
+private FDs" vs. "safe to inherit". Most robust, but
+requires tractor-wide audit.
+
+### 3. Explicit FD cleanup in `_ForkedProc`/`_child_target`
+
+Have `subint_forkserver_proc`'s `_child_target`
+closure explicitly close the parent-side IPC listener
+FDs before calling `_actor_child_main`. Requires
+being able to enumerate "the parent's listener FDs
+that the child shouldn't keep" — plausible via
+`Actor.ipc_server`'s socket objects.
+
+### 4. Use `os.posix_spawn` with explicit `file_actions`
+
+Instead of raw `os.fork()`, use `os.posix_spawn()`
+which supports explicit file-action specifications
+(close this FD, dup2 that FD). Cleaner semantics, but
+probably incompatible with our "no exec" requirement
+(subint_forkserver is a fork-without-exec design).
+
+**Likely correct answer: (3) — targeted FD cleanup
+via `actor.ipc_server` handle.** (1) is too blunt,
+(2) is too wide-ranging, (4) changes the spawn
+mechanism.
+
+## Reproducer (standalone, no pytest)
+
+```python
+# save as /tmp/forkserver_nested_hang_repro.py  (py3.14+)
+import trio, tractor
+
+async def assert_err():
+    assert 0
+
+async def spawn_and_error(breadth: int = 2, depth: int = 1):
+    async with tractor.open_nursery() as n:
+        for i in range(breadth):
+            if depth > 0:
+                await n.run_in_actor(
+                    spawn_and_error,
+                    breadth=breadth,
+                    depth=depth - 1,
+                    name=f'spawner_{i}_{depth}',
+                )
+            else:
+                await n.run_in_actor(
+                    assert_err,
+                    name=f'errorer_{i}',
+                )
+
+async def _main():
+    async with tractor.open_nursery() as n:
+        for i in range(2):
+            await n.run_in_actor(
+                spawn_and_error,
+                name=f'top_{i}',
+                breadth=2,
+                depth=1,
+            )
+
+if __name__ == '__main__':
+    from tractor.spawn._spawn import try_set_start_method
+    try_set_start_method('subint_forkserver')
+    with trio.fail_after(20):
+        trio.run(_main)
+```
+
+Expected (current): hangs on `trio.fail_after(20)`
+— children never ack the error-propagation cancel
+cascade. Pattern: top 2 direct children, 4
+grandchildren, 1 errorer deadlocks while trying to
+unwind through its parent chain.
+
+After fix: `trio.TooSlowError`-free completion; the
+root's `open_nursery` receives the
+`BaseExceptionGroup` containing the `AssertionError`
+from the errorer and unwinds cleanly.
+
+## Update — 2026-04-23: partial fix landed, deeper layer surfaced
+
+Three improvements landed as separate commits in the
+`subint_forkserver_backend` branch (see `git log`):
+
+1. **`_close_inherited_fds()` in fork-child prelude**
+   (`tractor/spawn/_subint_forkserver.py`). POSIX
+   close-fds-equivalent enumeration via
+   `/proc/self/fd` (or `RLIMIT_NOFILE` fallback), keep
+   only stdio. This is fix-direction (1) from the list
+   above — went with the blunt form rather than the
+   targeted enum-via-`actor.ipc_server` form, turns
+   out the aggressive close is safe because every
+   inheritable resource the fresh child needs
+   (IPC-channel socket, etc.) is opened AFTER the
+   fork anyway.
+2. **`_ForkedProc.wait()` via `os.pidfd_open()` +
+   `trio.lowlevel.wait_readable()`** — matches the
+   `trio.Process.wait` / `mp.Process.sentinel` pattern
+   used by `trio_proc` and `proc_waiter`. Gives us
+   fully trio-cancellable child-wait (prior impl
+   blocked a cache thread on a sync `os.waitpid` that
+   was NOT trio-cancellable due to
+   `abandon_on_cancel=False`).
+3. **`_parent_chan_cs` wiring** in
+   `tractor/runtime/_runtime.py`: capture the shielded
+   `loop_cs` for the parent-channel `process_messages`
+   task in `async_main`; explicitly cancel it in
+   `Actor.cancel()` teardown. This breaks the shield
+   during teardown so the parent-chan loop exits when
+   cancel is issued, instead of parking on a parent-
+   socket EOF that might never arrive under fork
+   semantics.
+
+**Concrete wins from (1):** the sibling
+`subint_forkserver_orphan_sigint_hang_issue.md` class
+is **now fixed** — `test_orphaned_subactor_sigint_cleanup_DRAFT`
+went from strict-xfail to pass. The xfail mark was
+removed; the test remains as a regression guard.
+
+**test_nested_multierrors STILL hangs** though.
+
+### Updated diagnosis (narrowed)
+
+DIAGDEBUG instrumentation of `process_messages` ENTER/
+EXIT pairs + `_parent_chan_cs.cancel()` call sites
+showed (captured during a 20s-timeout repro):
+
+- 80 `process_messages` ENTERs, 75 EXITs → 5 stuck.
+- **All 40 `shield=True` ENTERs matched EXIT** — every
+  shielded parent-chan loop exits cleanly. The
+  `_parent_chan_cs` wiring works as intended.
+- **The 5 stuck loops are all `shield=False`** — peer-
+  channel handlers (inbound connections handled by
+  `handle_stream_from_peer` in stream_handler_tn).
+- After our `_parent_chan_cs.cancel()` fires, NEW
+  shielded process_messages loops start (on the
+  session reg_addr port — probably discovery-layer
+  reconnection attempts). These don't block teardown
+  (they all exit) but indicate the cancel cascade has
+  more moving parts than expected.
+
+### Remaining unknown
+
+Why don't the 5 peer-channel loops exit when
+`service_tn.cancel_scope.cancel()` fires? They're in
+`stream_handler_tn` which IS `service_tn` in the
+current configuration (`open_ipc_server(parent_tn=
+service_tn, stream_handler_tn=service_tn)`). A
+standard nursery-scope-cancel should propagate through
+them — no shield, no special handler. Something
+specific to the fork-spawned configuration keeps them
+alive.
+
+Candidate follow-up experiments:
+
+- Dump the trio task tree at the hang point (via
+  `stackscope` or direct trio introspection) to see
+  what each stuck loop is awaiting. `chan.__anext__`
+  on a socket recv? An inner lock? A shielded sub-task?
+- Compare peer-channel handler lifecycle under
+  `trio_proc` vs `subint_forkserver` with equivalent
+  logging to spot the divergence.
+- Investigate whether the peer handler is caught in
+  the `except trio.Cancelled:` path at
+  `tractor/ipc/_server.py:448` that re-raises — but
+  re-raise means it should still exit. Unless
+  something higher up swallows it.
+
+### Attempted fix (DID NOT work) — hypothesis (3)
+
+Tried: in `_serve_ipc_eps` finally, after closing
+listeners, also iterate `server._peers` and
+sync-close each peer channel's underlying stream
+socket fd:
+
+```python
+for _uid, _chans in list(server._peers.items()):
+    for _chan in _chans:
+        try:
+            _stream = _chan._transport.stream if _chan._transport else None
+            if _stream is not None:
+                _stream.socket.close()  # sync fd close
+        except (AttributeError, OSError):
+            pass
+```
+
+Theory: closing the socket fd from outside the stuck
+recv task would make the recv see EBADF /
+ClosedResourceError and unblock.
+
+Result: `test_nested_multierrors[subint_forkserver]`
+still hangs identically. Either:
+- The sync `socket.close()` doesn't propagate into
+  trio's in-flight `recv_some()` the way I expected
+  (trio may hold an internal reference that keeps the
+  fd open even after an external close), or
+- The stuck recv isn't even the root blocker and the
+  peer handlers never reach the finally for some
+  reason I haven't understood yet.
+
+Either way, the sync-close hypothesis is **ruled
+out**. Reverted the experiment, restored the skip-
+mark on the test.
+
+### Aside: `-s` flag does NOT change `test_nested_multierrors` behavior
+
+Tested explicitly: both with and without `-s`, the
+test hangs identically. So the capture-pipe-fill
+hypothesis is **ruled out** for this test.
+
+The earlier `test_context_stream_semantics.py` `-s`
+observation was most likely caused by a competing
+pytest run in my session (confirmed via process list
+— my leftover pytest was alive at that time and
+could have been holding state on the default
+registry port).
+
+## Update — 2026-04-23 (late): cancel delivery ruled in, nursery-wait ruled BLOCKER
+
+**New diagnostic run** instrumented
+`handle_stream_from_peer` at ENTER / `except
+trio.Cancelled:` / finally, plus `Actor.cancel()`
+just before `self._parent_chan_cs.cancel()`. Result:
+
+- **40 `handle_stream_from_peer` ENTERs**.
+- **0 `except trio.Cancelled:` hits** — cancel
+  never fires on any peer-handler.
+- **35 finally hits** — those handlers exit via
+  peer-initiated EOF (normal return), NOT cancel.
+- **5 handlers never reach finally** — stuck forever.
+- **`Actor.cancel()` fired in 12 PIDs** — but the
+  PIDs with peer handlers that DIDN'T fire
+  Actor.cancel are exactly **root + 2 direct
+  spawners**. These 3 actors have peer handlers
+  (for their own subactors) that stay stuck because
+  **`Actor.cancel()` at these levels never runs**.
+
+### The actual deadlock shape
+
+`Actor.cancel()` lives in
+`open_root_actor.__aexit__` / `async_main` teardown.
+That only runs when the enclosing `async with
+tractor.open_nursery()` exits. The nursery's
+`__aexit__` calls the backend `*_proc` spawn target's
+teardown, which does `soft_kill() →
+_ForkedProc.wait()` on its child PID. That wait is
+trio-cancellable via pidfd now (good) — but nothing
+CANCELS it because the outer scope only cancels when
+`Actor.cancel()` runs, which only runs when the
+nursery completes, which waits on the child.
+
+It's a **multi-level mutual wait**:
+
+```
+root              blocks on spawner.wait()
+  spawner         blocks on grandchild.wait()
+    grandchild    blocks on errorer.wait()
+      errorer     Actor.cancel() ran, but process
+                  may not have fully exited yet
+                  (something in root_tn holding on?)
+```
+
+Each level waits for the level below. The bottom
+level (errorer) reaches Actor.cancel(), but its
+process may not fully exit — meaning its pidfd
+doesn't go readable, meaning the grandchild's
+waitpid doesn't return, meaning the grandchild's
+nursery doesn't unwind, etc. all the way up.
+
+### Refined question
+
+**Why does an errorer process not exit after its
+`Actor.cancel()` completes?**
+
+Possibilities:
+1. `_parent_chan_cs.cancel()` fires (shielded
+   parent-chan loop unshielded), but the task is
+   stuck INSIDE the shielded loop's recv in a way
+   that cancel still can't break.
+2. After `Actor.cancel()` returns, `async_main`
+   still has other tasks in `root_tn` waiting for
+   something that never arrives (e.g. outbound
+   IPC reply delivery).
+3. The `os._exit(rc)` in `_worker` (at
+   `_subint_forkserver.py`) doesn't run because
+   `_child_target` never returns.
+
+Next-session candidate probes (in priority order):
+
+1. **Instrument `_worker`'s fork-child branch** to
+   confirm whether `child_target()` returns (and
+   thus `os._exit(rc)` is reached) for errorer
+   PIDs. If yes → process should die; if no →
+   trace back into `_actor_child_main` /
+   `_trio_main` / `async_main` to find the stuck
+   spot.
+2. **Instrument `async_main`'s final unwind** to
+   see which await in the teardown doesn't
+   complete.
+3. **Compare under `trio_proc` backend** at the
+   same `_worker`-equivalent level to see where
+   the flows diverge.
+
+### Rule-out: NOT a stuck peer-chan recv
+
+Earlier hypothesis was that the 5 stuck peer-chan
+loops were blocked on a socket recv that cancel
+couldn't interrupt. This pass revealed the real
+cause: cancel **never reaches those tasks** because
+their owning actor's `Actor.cancel()` never runs.
+The recvs are fine — they're just parked because
+nothing is telling them to stop.
+
+## Update — 2026-04-23 (very late): leaves exit, middle actors stuck in `trio.run`
+
+Yet another instrumentation pass — this time
+printing at:
+
+- `_worker` child branch: `pre child_target()` /
+  `child_target RETURNED rc=N` / `about to
+  os._exit(rc)` 
+- `_trio_main`: `about to trio.run` /
+  `trio.run RETURNED NORMALLY` / `FINALLY`
+
+**Fresh-run results** (`test_nested_multierrors[
+subint_forkserver]`, depth=1/breadth=2, 1 root + 14
+forked = 15 actors total):
+
+- **9 processes completed the full flow** —
+  `trio.run RETURNED NORMALLY` → `child_target
+  RETURNED rc=0` → `about to os._exit(0)`. These
+  are the LEAVES of the tree (errorer actors) plus
+  their direct parents (depth-0 spawners). They
+  actually exit their processes.
+- **5 processes are stuck INSIDE `trio.run(trio_main)`**
+  — they hit "about to trio.run" but NEVER see
+  "trio.run RETURNED NORMALLY". These are root +
+  top-level spawners + one intermediate.
+
+**What this means:** `async_main` itself is the
+deadlock holder, not the peer-channel loops.
+Specifically, the outer `async with root_tn:` in
+`async_main` never exits for the 5 stuck actors.
+Their `trio.run` never returns → `_trio_main`
+catch/finally never runs → `_worker` never reaches
+`os._exit(rc)` → the PROCESS never dies → its
+parent's `_ForkedProc.wait()` blocks → parent's
+nursery hangs → parent's `async_main` hangs → ...
+
+### The new precise question
+
+**What task in the 5 stuck actors' `async_main`
+never completes?** Candidates:
+
+1. The shielded parent-chan `process_messages`
+   task in `root_tn` — but we explicitly cancel it
+   via `_parent_chan_cs.cancel()` in `Actor.cancel()`.
+   However, `Actor.cancel()` only runs during
+   `open_root_actor.__aexit__`, which itself runs
+   only after `async_main`'s outer unwind — which
+   doesn't happen. So the shield isn't broken.
+
+2. `await actor_nursery._join_procs.wait()` or
+   similar in the inline backend `*_proc` flow.
+
+3. `_ForkedProc.wait()` on a grandchild that
+   actually DID exit — but the pidfd_open watch
+   didn't fire for some reason (race between
+   pidfd_open and the child exiting?).
+
+The most specific next probe: **add DIAG around
+`_ForkedProc.wait()` enter/exit** to see whether
+the pidfd-based wait returns for every grandchild
+exit. If a stuck parent's `_ForkedProc.wait()`
+NEVER returns despite its child exiting, the
+pidfd mechanism has a race bug under nested
+forkserver.
+
+Alternative probe: instrument `async_main`'s outer
+nursery exits to find which nursery's `__aexit__`
+is stuck, drilling down from `trio.run` to the
+specific `async with` that never completes.
+
+### Cascade summary (updated tree view)
+
+```
+ROOT (pytest)                       STUCK in trio.run
+├── top_0 (spawner, d=1)            STUCK in trio.run
+│   ├── spawner_0_d1_0 (d=0)        exited (os._exit 0)
+│   │   ├── errorer_0_0             exited (os._exit 0)
+│   │   └── errorer_0_1             exited (os._exit 0)
+│   └── spawner_0_d1_1 (d=0)        exited (os._exit 0)
+│       ├── errorer_0_2             exited (os._exit 0)
+│       └── errorer_0_3             exited (os._exit 0)
+└── top_1 (spawner, d=1)            STUCK in trio.run
+    ├── spawner_1_d1_0 (d=0)        STUCK in trio.run (sibling race?)
+    │   ├── errorer_1_0             exited
+    │   └── errorer_1_1             exited
+    └── spawner_1_d1_1 (d=0)        STUCK in trio.run
+        ├── errorer_1_2             exited
+        └── errorer_1_3             exited
+```
+
+Grandchildren (d=0 spawners) exit OR stick —
+asymmetric. Not purely depth-determined. Some race
+condition in nursery teardown when multiple
+siblings error simultaneously.
+
+## Update — 2026-04-23 (late, probe iteration 3): hang pinpointed to `wait_for_no_more_peers()`
+
+Further DIAGDEBUG at every milestone in `async_main`
+(runtime UP / EXITED service_tn / EXITED root_tn /
+FINALLY ENTER / RETURNING) plus `_ForkedProc.wait`
+ENTER/RETURNED per-pidfd. Result:
+
+**Every stuck actor reaches `async_main: FINALLY
+ENTER` but NOT `async_main: RETURNING`.**
+
+That isolates the hang to a specific await in
+`async_main`'s finally block at
+`tractor/runtime/_runtime.py:1837+`. The suspect:
+
+```python
+# Ensure all peers (actors connected to us as clients) are finished
+if ipc_server := actor.ipc_server and ipc_server.has_peers(check_chans=True):
+    ...
+    await ipc_server.wait_for_no_more_peers()  # ← UNBOUNDED, blocks forever
+```
+
+`_no_more_peers` is an `Event` set only when
+`server._peers` empties (see
+`ipc/_server.py:526-530`). If ANY peer-handler is
+stuck (the 5 unclosed loops from the earlier pass),
+it keeps its channel in `server._peers`, so the
+event never fires, so the wait hangs.
+
+### Applied fix (partial, landed as defensive-in-depth)
+
+`tractor/runtime/_runtime.py:1981` —
+`wait_for_no_more_peers()` call now wrapped in
+`trio.move_on_after(3.0)` + a warning log when the
+timeout fires. Commented with the full rationale.
+
+**Verified:** with this fix, ALL 15 actors reach
+`async_main: RETURNING` cleanly (up from 10/15
+reaching end before).
+
+**Unfortunately:** the test still hangs past 45s
+total — meaning there's YET ANOTHER unbounded wait
+downstream of `async_main`. The bounded
+`wait_for_no_more_peers` unblocks one level, but
+the cascade has another level above it.
+
+### Candidates for the remaining hang
+
+1. `open_root_actor`'s own finally / post-
+   `async_main` flow in `_root.py` — specifically
+   `await actor.cancel(None)` which has its own
+   internal waits.
+2. The `trio.run()` itself doesn't return even
+   after the root task completes because trio's
+   nursery still has background tasks running.
+3. Maybe `_serve_ipc_eps`'s finally has an await
+   that blocks when peers aren't clearing.
+
+### Current stance
+
+- Defensive `wait_for_no_more_peers` bound landed
+  (good hygiene regardless). Revealing a real
+  deadlock-avoidance gap in tractor's cleanup.
+- Test still hangs → skip-mark restored on
+  `test_nested_multierrors[subint_forkserver]`.
+- The full chain of unbounded waits needs another
+  session of drilling, probably at
+  `open_root_actor` / `actor.cancel` level.
+
+### Summary of this investigation's wins
+
+1. **FD hygiene fix** (`_close_inherited_fds`) —
+   correct, closed orphan-SIGINT sibling issue.
+2. **pidfd-based `_ForkedProc.wait`** — cancellable,
+   matches trio_proc pattern.
+3. **`_parent_chan_cs` wiring** —
+   `Actor.cancel()` now breaks the shielded parent-
+   chan `process_messages` loop.
+4. **`wait_for_no_more_peers` bounded** —
+   prevents the actor-level finally hang.
+5. **Ruled-out hypotheses:** tree-kill missing
+   (wrong), stuck socket recv (wrong).
+6. **Pinpointed remaining unknown:** at least one
+   more unbounded wait in the teardown cascade
+   above `async_main`. Concrete candidates
+   enumerated above.
+
+## Update — 2026-04-23 (VERY late): pytest capture pipe IS the final gate
+
+After landing fixes 1-4 and instrumenting every
+layer down to `tractor_test`'s `trio.run(_main)`:
+
+**Empirical result: with `pytest -s` the test PASSES
+in 6.20s.** Without `-s` (default `--capture=fd`) it
+hangs forever.
+
+DIAG timeline for the root pytest PID (with `-s`
+implied from later verification):
+
+```
+tractor_test: about to trio.run(_main)
+open_root_actor: async_main task started, yielding to test body
+_main: about to await wrapped test fn
+_main: wrapped RETURNED cleanly        ← test body completed!
+open_root_actor: about to actor.cancel(None)
+Actor.cancel ENTER req_chan=False
+Actor.cancel RETURN
+open_root_actor: actor.cancel RETURNED
+open_root_actor: outer FINALLY
+open_root_actor: finally END (returning from ctxmgr)
+tractor_test: trio.run FINALLY (returned or raised)  ← trio.run fully returned!
+```
+
+`trio.run()` fully returns. The test body itself
+completes successfully (pytest.raises absorbed the
+expected `BaseExceptionGroup`). What blocks is
+**pytest's own stdout/stderr capture** — under
+`--capture=fd` default, pytest replaces the parent
+process's fd 1,2 with pipe write-ends it's reading
+from. Fork children inherit those pipe fds
+(because `_close_inherited_fds` correctly preserves
+stdio). High-volume subactor error-log tracebacks
+(7+ actors each logging multiple
+`RemoteActorError`/`ExceptionGroup` tracebacks on
+the error-propagation cascade) fill the 64KB Linux
+pipe buffer. Subactor writes block. Subactor can't
+progress. Process doesn't exit. Parent's
+`_ForkedProc.wait` (now pidfd-based and
+cancellable, but nothing's cancelling here since
+the test body already completed) keeps the pipe
+reader alive... but pytest isn't draining its end
+fast enough because test-teardown/fixture-cleanup
+is in progress.
+
+**Actually** the exact mechanism is slightly
+different: pytest's capture fixture MIGHT be
+actively reading, but faster-than-writer subactors
+overflow its internal buffer. Or pytest might be
+blocked itself on the finalization step.
+
+Either way, `-s` conclusively fixes it.
+
+### Why I ruled this out earlier (and shouldn't have)
+
+Earlier in this investigation I tested
+`test_nested_multierrors` with/without `-s` and
+both hung. That's because AT THAT TIME, fixes 1-4
+weren't all in place yet. The test was hanging at
+multiple deeper levels long before reaching the
+"generate lots of error-log output" phase. Once
+the cascade actually tore down cleanly, enough
+output was produced to hit the capture-pipe limit.
+
+**Classic order-of-operations mistake in
+debugging:** ruling something out too early based
+on a test that was actually failing for a
+different reason.
+
+### Fix direction (next session)
+
+Redirect subactor stdout/stderr to `/dev/null` (or
+a session-scoped log file) in the fork-child
+prelude, right after `_close_inherited_fds()`. This
+severs the inherited pytest-capture pipes and lets
+subactor output flow elsewhere. Under normal
+production use (non-pytest), stdout/stderr would
+be the TTY — we'd want to keep that. So the
+redirect should be conditional or opt-in via the
+`child_sigint`/proc_kwargs flag family.
+
+Alternative: document as a gotcha and recommend
+`pytest -s` for any tests using the
+`subint_forkserver` backend with multi-level actor
+trees. Simpler, user-visible, no code change.
+
+### Current state
+
+- Skip-mark on `test_nested_multierrors[subint_forkserver]`
+  restored with reason pointing here.
+- Test confirmed passing with `-s` after all 4
+  cascade fixes applied.
+- The 4 cascade fixes are NOT wasted — they're
+  correct hardening regardless of the capture-pipe
+  issue, AND without them we'd never reach the
+  "actually produces enough output to fill the
+  pipe" state.
+
+## Stopgap (landed)
+
+`test_nested_multierrors` skip-marked under
+`subint_forkserver` via
+`@pytest.mark.skipon_spawn_backend('subint_forkserver',
+reason='...')`, cross-referenced to this doc. Mark
+should be dropped once the peer-channel-loop exit
+issue is fixed.
+
+## References
+
+- `tractor/spawn/_subint_forkserver.py::fork_from_worker_thread`
+  — the primitive whose post-fork FD hygiene is
+  probably the culprit.
+- `tractor/spawn/_subint_forkserver.py::subint_forkserver_proc`
+  — the backend function that orchestrates the
+  graceful cancel path hitting this bug.
+- `tractor/spawn/_subint_forkserver.py::_ForkedProc`
+  — the `trio.Process`-compatible shim; NOT the
+  failing component (confirmed via thread-dump).
+- `tests/test_cancellation.py::test_nested_multierrors`
+  — the test that surfaced the hang.
+- `ai/conc-anal/subint_forkserver_orphan_sigint_hang_issue.md`
+  — sibling hang class; probably same underlying
+  fork-FD-inheritance root cause.
+- tractor issue #379 — subint backend tracking.

ai/conc-anal/subint_forkserver_thread_constraints_on_pep684_issue.md

@@ -0,0 +1,186 @@
+# Revisit `subint_forkserver` thread-cache constraints once msgspec PEP 684 support lands
+
+> **Tracked at:** [#450](https://github.com/goodboy/tractor/issues/450)
+
+Follow-up tracker for cleanup work gated on the msgspec
+PEP 684 adoption upstream ([jcrist/msgspec#563](https://github.com/jcrist/msgspec/issues/563)).
+
+Context — why this exists
+-------------------------
+
+The `tractor.spawn._subint_forkserver` submodule currently
+carries two "non-trio" thread-hygiene constraints whose
+necessity is tangled with issues that *should* dissolve
+under PEP 684 isolated-mode subinterpreters:
+
+1. `fork_from_worker_thread()` / `run_subint_in_worker_thread()`
+   internally allocate a **dedicated `threading.Thread`**
+   rather than using `trio.to_thread.run_sync()`.
+2. The test helper is named
+   `run_fork_in_non_trio_thread()` — the
+   `non_trio` qualifier is load-bearing today.
+
+This doc catalogs *why* those constraints exist, which of
+them isolated-mode would fix, and what the
+audit-and-cleanup path looks like once msgspec #563 is
+resolved.
+
+The three reasons the constraints exist
+---------------------------------------
+
+### 1. GIL-starvation class → fixed by PEP 684 isolated mode
+
+The class-A hang documented in
+`subint_sigint_starvation_issue.md` is entirely about
+legacy-config subints **sharing the main GIL**. Once
+msgspec #563 lands and tractor flips
+`tractor.spawn._subint` to
+`concurrent.interpreters.create()` (isolated config), each
+subint gets its own GIL. Abandoned subint threads can't
+contend for main's GIL → can't starve the main trio loop
+→ signal-wakeup-pipe drains normally → no SIGINT-drop.
+
+This class of hazard **dissolves entirely**. The
+non-trio-thread requirement for *this reason* disappears.
+
+### 2. Destroy race / tstate-recycling → orthogonal; unclear
+
+The `subint_proc` dedicated-thread fix (commit `26fb8206`)
+addressed a different issue: `_interpreters.destroy(interp_id)`
+was blocking on a trio-cache worker that had run an
+earlier `interp.exec()` for that subint. Working
+hypothesis at the time was "the cached thread retains the
+subint's tstate".
+
+But tstate-handling is **not specific to GIL mode** —
+`_PyXI_Enter` / `_PyXI_Exit` (the C-level machinery both
+configs use to enter/leave a subint from a thread) should
+restore the caller's tstate regardless of GIL config. So
+isolated mode **doesn't obviously fix this**. It might be:
+
+- A py3.13 bug fixed in later versions — we saw the race
+  first on 3.13 and never re-tested on 3.14 after moving
+  to dedicated threads.
+- A genuine CPython quirk around cached threads that
+  exec'd into a subint, persisting across GIL modes.
+- Something else we misdiagnosed — the empirical fix
+  (dedicated thread) worked but the analysis may have
+  been incomplete.
+
+Only way to know: once we're on isolated mode, empirically
+retry `trio.to_thread.run_sync(interp.exec, ...)` and see
+if `destroy()` still blocks. If it does, keep the
+dedicated thread; if not, one constraint relaxed.
+
+### 3. Fork-from-main-interp-tstate (the constraint in this module's helper names)
+
+The fork-from-main-interp-tstate invariant — CPython's
+`PyOS_AfterFork_Child` →
+`_PyInterpreterState_DeleteExceptMain` gate documented in
+`subint_fork_blocked_by_cpython_post_fork_issue.md` — is
+about the calling thread's **current** tstate at the
+moment `os.fork()` runs. If trio's cache threads never
+enter subints at all, their tstate is plain main-interp,
+and fork from them would be fine.
+
+The reason the smoke test +
+`run_fork_in_non_trio_thread` test helper
+currently use a dedicated `threading.Thread` is narrow:
+**we don't want to risk a trio cache thread that has
+previously been used as a subint driver being the one that
+picks up the fork job**. If cached tstate doesn't get
+cleared (back to reason #2), the fork's child-side
+post-init would see the wrong interp and abort.
+
+In an isolated-mode world where msgspec works:
+
+- `subint_proc` would use the public
+  `concurrent.interpreters.create()` + `Interpreter.exec()`
+  / `Interpreter.close()` — which *should* handle tstate
+  cleanly (they're the "blessed" API).
+- If so, trio's cache threads are safe to fork from
+  regardless of whether they've previously driven subints.
+- → the `non_trio` qualifier in
+  `run_fork_in_non_trio_thread` becomes
+  *overcautious* rather than load-bearing, and the
+  dedicated-thread primitives in `_subint_forkserver.py`
+  can likely be replaced with straight
+  `trio.to_thread.run_sync()` wrappers.
+
+TL;DR
+-----
+
+| constraint | fixed by isolated mode? |
+|---|---|
+| GIL-starvation (class A) | **yes** |
+| destroy race on cached worker | unclear — empirical test on py3.14 + isolated API required |
+| fork-from-main-tstate requirement on worker | **probably yes, conditional on the destroy-race question above** |
+
+If #2 also resolves on py3.14+ with isolated mode,
+tractor could drop the `non_trio` qualifier from the fork
+helper's name and just use `trio.to_thread.run_sync(...)`
+for everything. But **we shouldn't do that preemptively**
+— the current cautious design is cheap (one dedicated
+thread per fork / per subint-exec) and correct.
+
+Audit plan when msgspec #563 lands
+----------------------------------
+
+Assuming msgspec grows `Py_mod_multiple_interpreters`
+support:
+
+1. **Flip `tractor.spawn._subint` to isolated mode.** Drop
+   the `_interpreters.create('legacy')` call in favor of
+   the public API (`concurrent.interpreters.create()` +
+   `Interpreter.exec()` / `Interpreter.close()`). Run the
+   three `ai/conc-anal/subint_*_issue.md` reproducers —
+   class-A (`test_stale_entry_is_deleted` etc.) should
+   pass without the `skipon_spawn_backend('subint')` marks
+   (revisit the marker inventory).
+
+2. **Empirical destroy-race retest.** In `subint_proc`,
+   swap the dedicated `threading.Thread` back to
+   `trio.to_thread.run_sync(Interpreter.exec, ...,
+   abandon_on_cancel=False)` and run the full subint test
+   suite. If `Interpreter.close()` (or the backing
+   destroy) blocks the same way as the legacy version
+   did, revert and keep the dedicated thread.
+
+3. **If #2 clean**, audit `_subint_forkserver.py`:
+   - Rename `run_fork_in_non_trio_thread` → drop the
+     `_non_trio_` qualifier (e.g. `run_fork_in_thread`) or
+     inline the two-line `trio.to_thread.run_sync` call at
+     the call sites and drop the helper entirely.
+   - Consider whether `fork_from_worker_thread` +
+     `run_subint_in_worker_thread` still warrant being
+     separate module-level primitives or whether they
+     collapse into a compound
+     `trio.to_thread.run_sync`-driven pattern inside the
+     (future) `subint_forkserver_proc` backend.
+
+4. **Doc fallout.** `subint_sigint_starvation_issue.md`
+   and `subint_cancel_delivery_hang_issue.md` both cite
+   the legacy-GIL-sharing architecture as the root cause.
+   Close them with commit-refs to the isolated-mode
+   migration. This doc itself should get a closing
+   post-mortem section noting which of #1/#2/#3 actually
+   resolved vs persisted.
+
+References
+----------
+
+- `tractor.spawn._subint_forkserver` — the in-tree module
+  whose constraints this doc catalogs.
+- `ai/conc-anal/subint_sigint_starvation_issue.md` — the
+  GIL-starvation class.
+- `ai/conc-anal/subint_cancel_delivery_hang_issue.md` —
+  sibling Ctrl-C-able hang class.
+- `ai/conc-anal/subint_fork_blocked_by_cpython_post_fork_issue.md`
+  — why fork-from-subint is blocked (this drives the
+  forkserver-via-non-subint-thread workaround).
+- `ai/conc-anal/subint_fork_from_main_thread_smoketest.py`
+  — empirical validation for the workaround.
+- [PEP 684 — per-interpreter GIL](https://peps.python.org/pep-0684/)
+- [PEP 734 — `concurrent.interpreters` public API](https://peps.python.org/pep-0734/)
+- [jcrist/msgspec#563 — PEP 684 support tracker](https://github.com/jcrist/msgspec/issues/563)
+- tractor issue #379 — subint backend tracking.

ai/conc-anal/subint_sigint_starvation_issue.md

@@ -0,0 +1,350 @@
+# `subint` backend: abandoned-subint thread can wedge main trio event loop (Ctrl-C unresponsive)
+
+Follow-up to the Phase B subint spawn-backend PR (see
+`tractor.spawn._subint`, issue #379). The hard-kill escape
+hatch we landed (`_HARD_KILL_TIMEOUT`, bounded shields,
+`daemon=True` driver-thread abandonment) handles *most*
+stuck-subint scenarios cleanly, but there's one class of
+hang that can't be fully escaped from within tractor: a
+still-running abandoned sub-interpreter can starve the
+**parent's** trio event loop to the point where **SIGINT is
+effectively dropped by the kernel ↔ Python boundary** —
+making the pytest process un-Ctrl-C-able.
+
+## Symptom
+
+Running `test_stale_entry_is_deleted[subint]` under
+`--spawn-backend=subint`:
+
+1. Test spawns a subactor (`transport_fails_actor`) which
+   kills its own IPC server and then
+   `trio.sleep_forever()`.
+2. Parent tries `Portal.cancel_actor()` → channel
+   disconnected → fast return.
+3. Nursery teardown triggers our `subint_proc` cancel path.
+   Portal-cancel fails (dead channel),
+   `_HARD_KILL_TIMEOUT` fires, driver thread is abandoned
+   (`daemon=True`), `_interpreters.destroy(interp_id)`
+   raises `InterpreterError` (because the subint is still
+   running).
+4. Test appears to hang indefinitely at the *outer*
+   `async with tractor.open_nursery() as an:` exit.
+5. `Ctrl-C` at the terminal does nothing. The pytest
+   process is un-interruptable.
+
+## Evidence
+
+### `strace` on the hung pytest process
+
+```
+--- SIGINT {si_signo=SIGINT, si_code=SI_KERNEL} ---
+write(37, "\2", 1) = -1 EAGAIN (Resource temporarily unavailable)
+rt_sigreturn({mask=[WINCH]}) = 140585542325792
+```
+
+Translated:
+
+- Kernel delivers `SIGINT` to pytest.
+- CPython's C-level signal handler fires and tries to
+  write the signal number byte (`0x02` = SIGINT) to fd 37
+  — the **Python signal-wakeup fd** (set via
+  `signal.set_wakeup_fd()`, which trio uses to wake its
+  event loop on signals).
+- Write returns `EAGAIN` — **the pipe is full**. Nothing
+  is draining it.
+- `rt_sigreturn` with the signal masked off — signal is
+  "handled" from the kernel's perspective but the actual
+  Python-level handler (and therefore trio's
+  `KeyboardInterrupt` delivery) never runs.
+
+### Stack dump (via `tractor.devx.dump_on_hang`)
+
+At 20s into the hang, only the **main thread** is visible:
+
+```
+Thread 0x...7fdca0191780 [python] (most recent call first):
+  File ".../trio/_core/_io_epoll.py", line 245 in get_events
+  File ".../trio/_core/_run.py", line 2415 in run
+  File ".../tests/discovery/test_registrar.py", line 575 in test_stale_entry_is_deleted
+  ...
+```
+
+No driver thread shows up. The abandoned-legacy-subint
+thread still exists from the OS's POV (it's still running
+inside `_interpreters.exec()` driving the subint's
+`trio.run()` on `trio.sleep_forever()`) but the **main
+interp's faulthandler can't see threads currently executing
+inside a sub-interpreter's tstate**. Concretely: the thread
+is alive, holding state we can't introspect from here.
+
+## Root cause analysis
+
+The most consistent explanation for both observations:
+
+1. **Legacy-config subinterpreters share the main GIL.**
+   PEP 734's public `concurrent.interpreters.create()`
+   defaults to `'isolated'` (per-interp GIL), but tractor
+   uses `_interpreters.create('legacy')` as a workaround
+   for C extensions that don't yet support PEP 684
+   (notably `msgspec`, see
+   [jcrist/msgspec#563](https://github.com/jcrist/msgspec/issues/563)).
+   Legacy-mode subints share process-global state
+   including the GIL.
+
+2. **Our abandoned subint thread never exits.** After our
+   hard-kill timeout, `driver_thread.join()` is abandoned
+   via `abandon_on_cancel=True` and the thread is
+   `daemon=True` so proc-exit won't block on it — but the
+   thread *itself* is still alive inside
+   `_interpreters.exec()`, driving a `trio.run()` that
+   will never return (the subint actor is in
+   `trio.sleep_forever()`).
+
+3. **`_interpreters.destroy()` cannot force-stop a running
+   subint.** It raises `InterpreterError` on any
+   still-running subinterpreter; there is no public
+   CPython API to force-destroy one.
+
+4. **Shared-GIL + non-terminating subint thread → main
+   trio loop starvation.** Under enough load (the subint's
+   trio event loop iterating in the background, IPC-layer
+   tasks still in the subint, etc.) the main trio event
+   loop can fail to iterate frequently enough to drain its
+   wakeup pipe. Once that pipe fills, `SIGINT` writes from
+   the C signal handler return `EAGAIN` and signals are
+   silently dropped — exactly what `strace` shows.
+
+The shielded
+`await actor_nursery._join_procs.wait()` at the top of
+`subint_proc` (inherited unchanged from the `trio_proc`
+pattern) is structurally involved too: if main trio *does*
+get a schedule slice, it'd find the `subint_proc` task
+parked on `_join_procs` under shield — which traps whatever
+`Cancelled` arrives. But that's a second-order effect; the
+signal-pipe-full condition is the primary "Ctrl-C doesn't
+work" cause.
+
+## Why we can't fix this from inside tractor
+
+- **No force-destroy API.** CPython provides neither a
+  `_interpreters.force_destroy()` nor a thread-
+  cancellation primitive (`pthread_cancel` is actively
+  discouraged and unavailable on Windows). A subint stuck
+  in pure-Python loops (or worse, C code that doesn't poll
+  for signals) is structurally unreachable from outside.
+- **Shared GIL is the root scheduling issue.** As long as
+  we're forced into legacy-mode subints for `msgspec`
+  compatibility, the abandoned-thread scenario is
+  fundamentally a process-global GIL-starvation window.
+- **`signal.set_wakeup_fd()` is process-global.** Even if
+  we wanted to put our own drainer on the wakeup pipe,
+  only one party owns it at a time.
+
+## Current workaround
+
+- **Fixture-side SIGINT loop on the `daemon` subproc** (in
+  this test's `daemon: subprocess.Popen` fixture in
+  `tests/conftest.py`). The daemon dying closes its end of
+  the registry IPC, which unblocks a pending recv in main
+  trio's IPC-server task, which lets the event loop
+  iterate, which drains the wakeup pipe, which finally
+  delivers the test-harness SIGINT.
+- **Module-level skip on py3.13**
+  (`pytest.importorskip('concurrent.interpreters')`) — the
+  private `_interpreters` C module exists on 3.13 but the
+  multi-trio-task interaction hangs silently there
+  independently of this issue.
+
+## Path forward
+
+1. **Primary**: upstream `msgspec` PEP 684 adoption
+   ([jcrist/msgspec#563](https://github.com/jcrist/msgspec/issues/563)).
+   Unlocks `concurrent.interpreters.create()` isolated
+   mode → per-interp GIL → abandoned subint threads no
+   longer starve the parent's main trio loop. At that
+   point we can flip `_subint.py` back to the public API
+   (`create()` / `Interpreter.exec()` / `Interpreter.close()`)
+   and drop the private `_interpreters` path.
+
+2. **Secondary**: watch CPython for a public
+   force-destroy primitive. If something like
+   `Interpreter.close(force=True)` lands, we can use it as
+   a hard-kill final stage and actually tear down
+   abandoned subints.
+
+3. **Harness-level**: document the fixture-side SIGINT
+   loop pattern as the "known workaround" for subint-
+   backend tests that can leave background state holding
+   the main event loop hostage.
+
+## References
+
+- PEP 734 (`concurrent.interpreters`):
+  <https://peps.python.org/pep-0734/>
+- PEP 684 (per-interpreter GIL):
+  <https://peps.python.org/pep-0684/>
+- `msgspec` PEP 684 tracker:
+  <https://github.com/jcrist/msgspec/issues/563>
+- CPython `_interpretersmodule.c` source:
+  <https://github.com/python/cpython/blob/main/Modules/_interpretersmodule.c>
+- `tractor.spawn._subint` module docstring (in-tree
+  explanation of the legacy-mode choice and its
+  tradeoffs).
+
+## Reproducer
+
+```
+./py314/bin/python -m pytest \
+  tests/discovery/test_registrar.py::test_stale_entry_is_deleted \
+  --spawn-backend=subint \
+  --tb=short --no-header -v
+```
+
+Hangs indefinitely without the fixture-side SIGINT loop;
+with the loop, the test completes (albeit with the
+abandoned-thread warning in logs).
+
+## Additional known-hanging tests (same class)
+
+All three tests below exhibit the same
+signal-wakeup-fd-starvation fingerprint (`write() → EAGAIN`
+on the wakeup pipe after enough SIGINT attempts) and
+share the same structural cause — abandoned legacy-subint
+driver threads contending with the main interpreter for
+the shared GIL until the main trio loop can no longer
+drain its wakeup pipe fast enough to deliver signals.
+
+They're listed separately because each exposes the class
+under a different load pattern worth documenting.
+
+### `tests/discovery/test_registrar.py::test_stale_entry_is_deleted[subint]`
+
+Original exemplar — see the **Symptom** and **Evidence**
+sections above. One abandoned subint
+(`transport_fails_actor`, stuck in `trio.sleep_forever()`
+after self-cancelling its IPC server) is sufficient to
+tip main into starvation once the harness's `daemon`
+fixture subproc keeps its half of the registry IPC alive.
+
+### `tests/test_cancellation.py::test_cancel_while_childs_child_in_sync_sleep[subint-False]`
+
+Cancel a grandchild that's in sync Python sleep from 2
+nurseries up. The test's own docstring declares the
+dependency: "its parent should issue a 'zombie reaper' to
+hard kill it after sufficient timeout" — which for
+`trio`/`mp_*` is an OS-level `SIGKILL` of the grandchild
+subproc. **Under `subint` there's no equivalent** (no
+public CPython API to force-destroy a running
+sub-interpreter), so the grandchild's sync-sleeping
+`trio.run()` persists inside its abandoned driver thread
+indefinitely. The nested actor-tree (parent → child →
+grandchild, all subints) means a single cancel triggers
+multiple concurrent hard-kill abandonments, each leaving
+a live driver thread.
+
+This test often only manifests the starvation under
+**full-suite runs** rather than solo execution —
+earlier-in-session subint tests also leave abandoned
+driver threads behind, and the combined population is
+what actually tips main trio into starvation. Solo runs
+may stay Ctrl-C-able with fewer abandoned threads in the
+mix.
+
+### `tests/test_cancellation.py::test_multierror_fast_nursery[subint-25-0.5]`
+
+Nursery-error-path throughput stress-test parametrized
+for **25 concurrent subactors**. When the multierror
+fires and the nursery cancels, every subactor goes
+through our `subint_proc` teardown. The bounded
+hard-kills run in parallel (all `subint_proc` tasks are
+sibling trio tasks), so the timeout budget is ~3s total
+rather than 3s × 25. After that, **25 abandoned
+`daemon=True` driver threads are simultaneously alive** —
+an extreme pressure multiplier on the same mechanism.
+
+The `strace` fingerprint is striking under this load: six
+or more **successful** `write(16, "\2", 1) = 1` calls
+(main trio getting brief GIL slices, each long enough to
+drain exactly one wakeup-pipe byte) before finally
+saturating with `EAGAIN`:
+
+```
+--- SIGINT {si_signo=SIGINT, si_code=SI_KERNEL} ---
+write(16, "\2", 1)                      = 1
+rt_sigreturn({mask=[WINCH]})            = 140141623162400
+--- SIGINT {si_signo=SIGINT, si_code=SI_KERNEL} ---
+write(16, "\2", 1)                      = 1
+rt_sigreturn({mask=[WINCH]})            = 140141623162400
+--- SIGINT {si_signo=SIGINT, si_code=SI_KERNEL} ---
+write(16, "\2", 1)                      = 1
+rt_sigreturn({mask=[WINCH]})            = 140141623162400
+--- SIGINT {si_signo=SIGINT, si_code=SI_KERNEL} ---
+write(16, "\2", 1)                      = 1
+rt_sigreturn({mask=[WINCH]})            = 140141623162400
+--- SIGINT {si_signo=SIGINT, si_code=SI_KERNEL} ---
+write(16, "\2", 1)                      = 1
+rt_sigreturn({mask=[WINCH]})            = 140141623162400
+--- SIGINT {si_signo=SIGINT, si_code=SI_KERNEL} ---
+write(16, "\2", 1)                      = 1
+rt_sigreturn({mask=[WINCH]})            = 140141623162400
+--- SIGINT {si_signo=SIGINT, si_code=SI_KERNEL} ---
+write(16, "\2", 1)                      = -1 EAGAIN (Resource temporarily unavailable)
+rt_sigreturn({mask=[WINCH]})            = 140141623162400
+```
+
+Those successful writes indicate CPython's
+`sys.getswitchinterval()`-based GIL round-robin *is*
+giving main brief slices — just never long enough to run
+the Python-level signal handler through to the point
+where trio converts the delivered SIGINT into a
+`Cancelled` on the appropriate scope. Once the
+accumulated write rate outpaces main's drain rate, the
+pipe saturates and subsequent signals are silently
+dropped.
+
+The `pstree` below (pid `530060` = hung `pytest`) shows
+the subint-driver thread population at the moment of
+capture. Even with fewer than the full 25 shown (pstree
+truncates thread names to `subint-driver[<interp_id>` —
+interpreters `3` and `4` visible across 16 thread
+entries), the GIL-contender count is more than enough to
+explain the starvation:
+
+```
+>>> pstree -snapt 530060
+systemd,1 --switched-root --system --deserialize=40
+  └─login,1545 --
+      └─bash,1872
+          └─sway,2012
+              └─alacritty,70471 -e xonsh
+                  └─xonsh,70487 .../bin/xonsh
+                      └─uv,70955 run xonsh
+                          └─xonsh,70959 .../py314/bin/xonsh
+                              └─python,530060 .../py314/bin/pytest -v tests/test_cancellation.py --spawn-backend=subint
+                                  ├─{subint-driver[3},531857
+                                  ├─{subint-driver[3},531860
+                                  ├─{subint-driver[3},531862
+                                  ├─{subint-driver[3},531866
+                                  ├─{subint-driver[3},531877
+                                  ├─{subint-driver[3},531882
+                                  ├─{subint-driver[3},531884
+                                  ├─{subint-driver[3},531945
+                                  ├─{subint-driver[3},531950
+                                  ├─{subint-driver[3},531952
+                                  ├─{subint-driver[4},531956
+                                  ├─{subint-driver[4},531959
+                                  ├─{subint-driver[4},531961
+                                  ├─{subint-driver[4},531965
+                                  ├─{subint-driver[4},531968
+                                  └─{subint-driver[4},531979
+```
+
+(`pstree` uses `{...}` to denote threads rather than
+processes — these are all the **driver OS-threads** our
+`subint_proc` creates with name
+`f'subint-driver[{interp_id}]'`. Every one of them is
+still alive, executing `_interpreters.exec()` inside a
+sub-interpreter our hard-kill has abandoned. At 16+
+abandoned driver threads competing for the main GIL, the
+main-interpreter trio loop gets starved and signal
+delivery stalls.)

ai/prompt-io/claude/20260417T124437Z_5cd6df5_prompt_io.md

@@ -0,0 +1,97 @@
+---
+model: claude-opus-4-7[1m]
+service: claude
+session: subints-spawner-phase-b1-impl
+timestamp: 2026-04-17T12:44:37Z
+git_ref: 5cd6df5
+scope: code
+substantive: true
+raw_file: 20260417T124437Z_5cd6df5_prompt_io.raw.md
+---
+
+## Prompt
+
+Continuing the Phase B work from the design sessions
+logged in `20260417T034918Z_9703210_prompt_io.md` and
+the Phase A impl in `20260417T035800Z_61a73ba_...`.
+
+User direction at this step:
+- "ok now let's continue with the subints part of the
+  plan ya?" → kick off Phase B.
+- After I proposed a 4-sub-phase breakdown (B.1
+  scaffolding, B.2 min-viable spawn, B.3 lifecycle,
+  B.4 test enablement), user: "ok can we continue
+  with B" → approving the phasing and design
+  defaults (thread-per-subint, UDS default,
+  code-string bootstrap via `interp.exec()`).
+- User switched the worktree to
+  `subint_spawner_backend` (instead of opening a
+  fresh worktree as I'd proposed): "i already
+  switched branches in the wkt, you should work off
+  `subint_spawner_backend` now".
+
+Scope of this turn: **B.1 scaffolding only** —
+wire the `'subint'` name through the spawn-method
+registry + harness, add a feature-detected stub
+module, bump pyproject. No actual spawn flow.
+
+## Response summary
+
+Produced the Phase B.1 scaffolding patch, landed as
+commit `5cd6df58` (subject: `Add `'subint'` spawn
+backend scaffold (#379)`). Four files changed,
++124 / -2 lines.
+
+Key pieces (all generated by claude, reviewed by
+human before commit):
+- `tractor/spawn/_subint.py` — **new**; feature-
+  detects `concurrent.interpreters`; `subint_proc()`
+  stub raises `RuntimeError` on py<3.14 or
+  `NotImplementedError` with issue-#379 URL on
+  py≥3.14. Signature mirrors `trio_proc`/`mp_proc`
+  so B.2 can drop the impl in without touching
+  `_methods`.
+- `tractor/spawn/_spawn.py` — adds `'subint'` to
+  `SpawnMethodKey`, grows a `case 'subint'` arm in
+  `try_set_start_method()` with feature-gate, re-
+  imports `sys` for the gate-error msg, adds late
+  `from ._subint import subint_proc` import and
+  `_methods` entry.
+- `tractor/_testing/pytest.py` — converts the
+  gate-error into `pytest.UsageError` via a
+  `try/except` around `try_set_start_method()` so
+  `--spawn-backend=subint` on py<3.14 prints a
+  clean banner instead of a traceback.
+- `pyproject.toml` — pin `requires-python` `<3.14`
+  → `<3.15`, add `3.14` trove classifier.
+
+## Files changed
+
+See `git diff 5cd6df5~1..5cd6df5 --stat`:
+
+```
+ pyproject.toml             |   3 +-
+ tractor/_testing/pytest.py |   8 +++-
+ tractor/spawn/_spawn.py    |  15 ++++++
+ tractor/spawn/_subint.py   | 100 +++++++++++++++++++++++++
+ 4 files changed, 124 insertions(+), 2 deletions(-)
+```
+
+Validation on py3.13:
+- registry/gate probe — OK; `_has_subints = False`,
+  `subint_proc()` raises `RuntimeError` as expected.
+- `try_set_start_method('subint')` — raises cleanly.
+- `pytest --spawn-backend=subint` — exits with
+  `pytest.UsageError` banner.
+- Spawn-relevant test subset — 69 passed, 1 skipped.
+- Full suite on py3.13 deferred (scaffolding is
+  no-op on <3.14).
+- Full-suite on py3.14 pending user setting up the
+  venv.
+
+## Human edits
+
+None — committed as generated by claude. `uv.lock`
+was intentionally left unstaged by the user (pin-
+bump regenerated cp314 wheel entries but they were
+deferred to a follow-up).

ai/prompt-io/claude/20260417T124437Z_5cd6df5_prompt_io.raw.md

@@ -0,0 +1,168 @@
+---
+model: claude-opus-4-7[1m]
+service: claude
+timestamp: 2026-04-17T12:44:37Z
+git_ref: 5cd6df5
+diff_cmd: git diff 5cd6df5~1..5cd6df5
+---
+
+Code generated in this turn was committed verbatim as
+`5cd6df58` ("Add `'subint'` spawn backend scaffold
+(#379)"). Per diff-ref mode, per-file code is captured
+via the pointers below, each followed by a prose
+summary. Non-code output (sanity-check results,
+design rationale) is included verbatim.
+
+## Per-file generated content
+
+### `tractor/spawn/_subint.py` (new, 100 lines)
+
+> `git diff 5cd6df5~1..5cd6df5 -- tractor/spawn/_subint.py`
+
+New scaffolding module for the PEP 734 subinterpreter
+backend. Contents:
+- AGPL header + module docstring (describes backend
+  intent, 3.14+ availability gate, and explicit
+  "SCAFFOLDING STUB" status pointing at issue #379).
+- Top-level `try/except ImportError` wrapping
+  `from concurrent import interpreters as
+  _interpreters` → sets module-global
+  `_has_subints: bool`. This lets the registry stay
+  introspectable on py<3.14 while spawn-time still
+  fails cleanly.
+- `subint_proc()` coroutine with signature matching
+  `trio_proc`/`mp_proc` exactly (same param names,
+  defaults, and `TaskStatus[Portal]` typing) —
+  intentional so Phase B.2 can drop the impl in
+  without touching `_methods` or changing call-site
+  binding.
+- Body raises `RuntimeError` on py<3.14 (with
+  `sys.version` printed) or `NotImplementedError`
+  with issue-#379 URL on py≥3.14.
+
+### `tractor/spawn/_spawn.py` (modified, +15 LOC)
+
+> `git diff 5cd6df5~1..5cd6df5 -- tractor/spawn/_spawn.py`
+
+- `import sys` re-added (pruned during Phase A, now
+  needed again for the py-version string in the
+  `'subint'` gate-error).
+- `SpawnMethodKey = Literal[...]` grows `'subint'` as
+  the 4th member, with inline comment `# py3.14+ via
+  `concurrent.interpreters` (PEP 734)`.
+- `try_set_start_method()` match-block grows a new
+  `case 'subint':` arm that imports
+  `from ._subint import _has_subints` lazily and
+  raises `RuntimeError` with a multi-line gate msg
+  if unavailable.
+- Bottom-of-module late-import section grows
+  `from ._subint import subint_proc` alongside the
+  existing `_trio` / `_mp` imports.
+- `_methods` dict grows `'subint': subint_proc`.
+
+### `tractor/_testing/pytest.py` (modified, +8 LOC)
+
+> `git diff 5cd6df5~1..5cd6df5 -- tractor/_testing/pytest.py`
+
+`pytest_configure()` wraps the
+`try_set_start_method(backend)` call in a
+`try/except RuntimeError` that re-raises as
+`pytest.UsageError(str(err))`. Rationale: the gate
+error on py<3.14 is legitimately a configuration
+problem, not a test failure, so pytest's UsageError
+path (exit code 4) gives a clean single-line banner
+instead of a traceback.
+
+### `pyproject.toml` (modified, +2 / -1)
+
+> `git diff 5cd6df5~1..5cd6df5 -- pyproject.toml`
+
+- `requires-python` pin relaxed `>=3.12, <3.14` →
+  `>=3.12, <3.15` to admit 3.14 as a valid target.
+- Added `"Programming Language :: Python :: 3.14"`
+  to the trove classifiers.
+
+## Non-code output (verbatim)
+
+### Design choices captured in the prior turn
+
+(Previously confirmed by the user on April 17 in this
+session — relevant excerpts captured here for
+provenance since they drove the shape of the impl):
+
+1. **Python pin**: `<3.15` + feature-gate the subint
+   backend at spawn time — user answered option (a)
+   from the design triage.
+2. **Deferred `fork()`-via-subint hack** from issue
+   #379's "Our own thoughts" section.
+3. **Phase B phasing**: user approved the B.1 / B.2 /
+   B.3 / B.4 breakdown — this commit is strictly B.1
+   (scaffolding only, no spawn-flow impl).
+4. **Option (B) worktree strategy**: new worktree
+   branched from `wkt/spawn_modularize`. *(Amended by
+   user at runtime: user switched the existing
+   `spawn_modularize` worktree to the
+   `subint_spawner_backend` branch instead.)*
+
+### Sanity-check output (verbatim terminal excerpts)
+
+Registry / feature-gate verification on py3.13:
+```
+SpawnMethodKey values: ('trio', 'mp_spawn',
+                       'mp_forkserver', 'subint')
+_methods keys: ['trio', 'mp_spawn',
+                'mp_forkserver', 'subint']
+_has_subints: False (py version: (3, 13) )
+[expected] RuntimeError: The 'subint' spawn backend
+requires Python 3.14+ (stdlib
+`concurrent.interpreters`, PEP 734).
+```
+
+`try_set_start_method('subint')` gate on py3.13:
+```
+[expected] RuntimeError: Spawn method 'subint'
+requires Python 3.14+ (stdlib
+`concurrent.interpreters`, PEP 734).
+```
+
+Pytest `--spawn-backend=subint` on py3.13 (the new
+UsageError wrapper kicking in):
+```
+ERROR: Spawn method 'subint' requires Python 3.14+
+(stdlib `concurrent.interpreters`, PEP 734).
+Current runtime: 3.13.11 (main, Dec  5 2025,
+16:06:33) [GCC 15.2.0]
+```
+
+Collection probe: `404 tests collected in 0.18s`
+(no import errors from the new module).
+
+Spawn-relevant test subset (`tests/test_local.py
+test_rpc.py test_spawning.py test_multi_program.py
+tests/discovery/`):
+```
+69 passed, 1 skipped, 10 warnings in 61.38s
+```
+
+Full suite was **not** run on py3.13 for this commit
+— the scaffolding is no-op on <3.14 and full-suite
+validation under py3.14 is pending that venv being
+set up by the user.
+
+### Commit message
+
+Also AI-drafted (via `/commit-msg`, with the prose
+rewrapped through `/home/goodboy/.claude/skills/pr-msg/
+scripts/rewrap.py --width 67`) — the 33-line message
+on commit `5cd6df58` itself. Not reproduced here; see
+`git log -1 5cd6df58`.
+
+### Known follow-ups flagged to user
+
+- **`uv.lock` deferred**: pin-bump regenerated cp314
+  wheel entries in `uv.lock`, but the user chose to
+  not stage `uv.lock` for this commit. Warned
+  explicitly.
+- **Phase B.2 needs py3.14 venv** — running the
+  actual subint impl requires it; user said they'd
+  set it up separately.

ai/prompt-io/claude/20260418T042526Z_26fb820_prompt_io.md

@@ -0,0 +1,117 @@
+---
+model: claude-opus-4-7[1m]
+service: claude
+session: subints-phase-b2-destroy-race-fix
+timestamp: 2026-04-18T04:25:26Z
+git_ref: 26fb820
+scope: code
+substantive: true
+raw_file: 20260418T042526Z_26fb820_prompt_io.raw.md
+---
+
+## Prompt
+
+Follow-up to Phase B.2 (`5cd6df58`) after the user
+observed intermittent mid-suite hangs when running
+the tractor test suite under `--spawn-backend=subint`
+on py3.14. The specific sequence of prompts over
+several turns:
+
+1. User pointed at the `test_context_stream_semantics.py`
+   suite as the first thing to make run clean under
+   `--spawn-backend=subint`.
+2. After a series of `timeout`-terminated runs that
+   gave no diagnostic info, user nudged me to stop
+   relying on `timeout` and get actual runtime
+   diagnostics ("the suite hangs indefinitely, so i
+   don't think this `timeout 30` is helping you at
+   all.."). Switched to
+   `faulthandler.dump_traceback_later(...)` and a
+   resource-tracker fixture to rule out leaks.
+3. Captured a stack pinning the hang on
+   `_interpreters.destroy(interp_id)` in the subint
+   teardown finally block.
+4. Proposed dedicated-OS-thread fix. User greenlit.
+5. Implemented + verified on-worktree; user needed
+   to be pointed at the *worktree*'s `./py313` venv
+   because bare `pytest` was picking up the main
+   repo's venv (running un-patched `_subint.py`) and
+   still hanging.
+
+Running theme over the whole exchange: this patch
+only closes the *destroy race*. The user and I also
+traced through the deeper cancellation story — SIGINT
+can't reach subints, legacy-mode shares the GIL,
+portal-cancel dies when the IPC channel is already
+broken — and agreed the next step is a bounded
+hard-kill in `subint_proc`'s teardown plus a
+dedicated cancellation test suite. Those land as
+separate commits.
+
+## Response summary
+
+Produced the `tractor/spawn/_subint.py` patch landed
+as commit `26fb8206` ("Fix subint destroy race via
+dedicated OS thread"). One file, +110/-84 LOC.
+
+Mechanism: swap `trio.to_thread.run_sync(_interpreters
+.exec, ...)` for a plain `threading.Thread(target=...
+, daemon=False)`. The trio thread cache recycles
+workers — so the OS thread that ran `_interpreters
+.exec()` remained alive in the cache holding a
+stale subint tstate, blocking
+`_interpreters.destroy()` in the finally indefinitely.
+A dedicated one-shot thread exits naturally after
+the sync target returns, releasing tstate and
+unblocking destroy.
+
+Coordination across the trio↔thread boundary:
+- `trio.lowlevel.current_trio_token()` captured at
+  `subint_proc` entry
+- driver thread signals `subint_exited.set()` back
+  to parent trio via `trio.from_thread.run_sync(...,
+  trio_token=token)` (synchronous from the thread's
+  POV; the call returns after trio has run `.set()`)
+- `trio.RunFinishedError` swallowed in that path for
+  the process-teardown case where parent trio already
+  exited
+- teardown `finally` off-loads the sync
+  `driver_thread.join()` via `to_thread.run_sync` (a
+  cache thread carries no subint tstate — safe)
+
+## Files changed
+
+See `git diff 26fb820~1..26fb820 --stat`:
+
+```
+ tractor/spawn/_subint.py | 194 +++++++++++++++++++------------
+ 1 file changed, 110 insertions(+), 84 deletions(-)
+```
+
+Validation:
+- `test_parent_cancels[chk_ctx_result_before_exit=True-
+  cancel_method=ctx-child_returns_early=False]`
+  (the specific test that was hanging for the user)
+  — passed in 1.06s.
+- Full `tests/test_context_stream_semantics.py` under
+  subint — 61 passed in 100.35s (clean-cache re-run:
+  100.82s).
+- Trio backend regression subset — 69 passed / 1
+  skipped / 89.19s — no regressions from this change.
+
+## Files changed
+
+Beyond the `_subint.py` patch, the raw log also
+records the cancellation-semantics research that
+spanned this conversation but did not ship as code
+in *this* commit. Preserving it inline under "Non-
+code output" because it directly informs the
+Phase B.3 hard-kill impl that will follow (and any
+upstream CPython bug reports we end up filing).
+
+## Human edits
+
+None — committed as generated. The commit message
+itself was also AI-drafted via `/commit-msg` and
+rewrapped via the project's `rewrap.py --width 67`
+tooling; user landed it without edits.

ai/prompt-io/claude/20260418T042526Z_26fb820_prompt_io.raw.md

@@ -0,0 +1,220 @@
+---
+model: claude-opus-4-7[1m]
+service: claude
+timestamp: 2026-04-18T04:25:26Z
+git_ref: 26fb820
+diff_cmd: git diff 26fb820~1..26fb820
+---
+
+Code generated in this turn was committed as `26fb8206`
+("Fix subint destroy race via dedicated OS thread").
+Per diff-ref mode, per-file code is captured via the
+pointer below; the rest is verbatim non-code output
+— test-session diagnostics, CPython-internals
+reasoning, and the design synthesis that's worth
+keeping alongside the code since it informs the
+still-pending Phase B.3 hard-kill work.
+
+## Per-file generated content
+
+### `tractor/spawn/_subint.py` (modified, +110/-84 LOC)
+
+> `git diff 26fb820~1..26fb820 -- tractor/spawn/_subint.py`
+
+Rewrites the subint driver-thread strategy:
+- replaces `trio.to_thread.run_sync(_interpreters.exec, ...)`
+  with a plain `threading.Thread(target=_subint_target,
+  daemon=False)` so the OS thread truly exits after
+  `_interpreters.exec()` returns
+- captures a `trio.lowlevel.current_trio_token()` at
+  `subint_proc` entry; the driver thread signals
+  completion back via `trio.from_thread.run_sync(
+  subint_exited.set, trio_token=...)`
+- swallows `trio.RunFinishedError` in the signal path
+  for the case where the parent trio loop has already
+  exited (process teardown)
+- in the teardown `finally` off-loads the sync
+  `driver_thread.join()` call to `trio.to_thread.run_sync`
+  (a cache thread w/ no subint tstate — so no cache
+  conflict) to wait for the driver thread to fully
+  exit before calling `_interpreters.destroy()`
+
+## Non-code output (verbatim) — the CPython-internals research
+
+### What went wrong before this commit
+
+Under `--spawn-backend=subint` on py3.14, most single
+tests passed but longer runs hung intermittently. The
+position of the hang moved between runs (test #22 on
+one run, test #53 on another) suggesting a timing-
+dependent race rather than a deterministic bug.
+
+`faulthandler.dump_traceback_later()` eventually
+caught a stack with the main thread blocked in
+`_interpreters.destroy(interp_id)` at `_subint.py:293`.
+Only 2 threads were alive:
+- main thread waiting in `_interpreters.destroy()`
+- one idle trio thread-cache worker in
+  `trio._core._thread_cache._work`
+
+No subint was still running (`_interpreters.list_all()`
+showed only the main interp). A resource-tracker
+pytest fixture confirmed threads/subints did NOT
+accumulate across tests — this was not a leak but a
+specific "destroy blocks on cached thread w/ stale
+tstate" race.
+
+### Why the race exists
+
+`trio.to_thread.run_sync` uses a thread *cache* to
+avoid OS-thread creation overhead. When the sync
+callable returns, the OS thread is NOT terminated —
+it's parked in `_thread_cache._work` waiting for the
+next job. CPython's subinterpreter implementation
+attaches a **tstate** (thread-state object) to each
+OS thread that ever entered a subint via
+`_interpreters.exec()`. That tstate is released
+lazily — either when the thread picks up a new job
+(which re-attaches a new tstate, evicting the old
+one) or when the thread truly exits.
+
+`_interpreters.destroy(interp_id)` waits for *all*
+tstates associated w/ that subint to be released
+before it can proceed. If the cached worker is idle
+holding the stale tstate, destroy blocks indefinitely.
+Whether the race manifests depends on timing — if
+the cached thread happens to pick up another job
+quickly, destroy unblocks; if it sits idle, we hang.
+
+### Why a dedicated `threading.Thread` fixes it
+
+A plain `threading.Thread(target=_subint_target,
+daemon=False)` runs its target once and exits. When
+the target returns, OS-thread teardown (`_bootstrap_inner`
+→ `_bootstrap`) fires and CPython releases the
+tstate for that thread. `_interpreters.destroy()`
+then has no blocker.
+
+### Diagnostic tactics that actually helped
+
+1. `faulthandler.dump_traceback_later(n, repeat=False,
+   file=open(path, 'w'))` for captured stack dumps on
+   hang. Critically, pipe to a `file=` not stderr —
+   pytest captures stderr weirdly and the dump is
+   easy to miss.
+2. A resource-tracker autouse fixture printing
+   per-test `threading.active_count()` +
+   `len(_interpreters.list_all())` deltas → ruled out
+   leak-accumulation theories quickly.
+3. Running the hanging test *solo* vs in-suite —
+   when solo passes but in-suite hangs, you know
+   it's a cross-test state-transfer bug rather than
+   a test-internal bug.
+
+### Design synthesis — SIGINT + subints + SC
+
+The user and I walked through the cancellation
+semantics of PEP 684/734 subinterpreters in detail.
+Key findings we want to preserve:
+
+**Signal delivery in subints (stdlib limitation).**
+CPython's signal machinery only delivers signals
+(SIGINT included) to the *main thread of the main
+interpreter*. Subints cannot install signal handlers
+that will ever fire. This is an intentional design
+choice in PEP 684 and not expected to change. For
+tractor's subint actors, this means:
+
+- Ctrl-C never reaches a subint directly.
+- `trio.run()` running on a worker thread (as we do
+  for subints) already skips SIGINT handler install
+  because `signal.signal()` raises on non-main
+  threads.
+- The only cancellation surface into a subint is
+  our IPC `Portal.cancel_actor()`.
+
+**Legacy-mode subints share the main GIL** (which
+our impl uses since `msgspec` lacks PEP 684 support
+per `jcrist/msgspec#563`). This means a stuck subint
+thread can starve the parent's trio loop during
+cancellation — the parent can't even *start* its
+teardown handling until the subint yields the GIL.
+
+**Failure modes identified for Phase B.3 audit:**
+
+1. Portal cancel lands cleanly → subint unwinds →
+   thread exits → destroy succeeds. (Happy path.)
+2. IPC channel is already broken when we try to
+   send cancel (e.g., `test_ipc_channel_break_*`)
+   → cancel raises `BrokenResourceError` → subint
+   keeps running unaware → parent hangs waiting for
+   `subint_exited`. This is what breaks
+   `test_advanced_faults.py` under subint.
+3. Subint is stuck in non-checkpointing Python code
+   → portal-cancel msg queued but never processed.
+4. Subint is in a shielded cancel scope when cancel
+   arrives → delay until shield exits.
+
+**Current teardown has a shield-bug too:**
+`trio.CancelScope(shield=True)` wrapping the `finally`
+block absorbs Ctrl-C, so even when the user tries
+to break out they can't. This is the reason
+`test_ipc_channel_break_during_stream[break_parent-...
+no_msgstream_aclose]` locks up unkillable.
+
+**B.3 hard-kill fix plan (next commit):**
+
+1. Bound `driver_thread.join()` with
+   `trio.move_on_after(HARD_KILL_TIMEOUT)`.
+2. If it times out, log a warning naming the
+   `interp_id` and switch the driver thread to
+   `daemon=True` mode (not actually possible after
+   start — so instead create as daemon=True upfront
+   and accept the tradeoff of proc-exit not waiting
+   for a stuck subint).
+3. Best-effort `_interpreters.destroy()`; catch the
+   `InterpreterError` if the subint is still running.
+4. Document that the leak is real and the only
+   escape hatch we have without upstream cooperation.
+
+**Test plan for Phase B.3:**
+
+New `tests/test_subint_cancellation.py` covering:
+- SIGINT at spawn
+- SIGINT mid-portal-RPC
+- SIGINT during shielded section in subint
+- Dead-channel cancel (mirror of `test_ipc_channel_
+  break_during_stream` minimized)
+- Non-checkpointing subint (tight `while True` in
+  user code)
+- Per-test `pytest-timeout`-style bounds so the
+  tests visibly fail instead of wedging the runner
+
+### Sanity-check output (verbatim terminal excerpts)
+
+Post-fix single-test validation:
+```
+1 passed, 1 warning in 1.06s
+```
+(same test that was hanging pre-fix:
+`test_parent_cancels[...cancel_method=ctx-...False]`)
+
+Full `tests/test_context_stream_semantics.py`
+under subint:
+```
+61 passed, 1 warning in 100.35s (0:01:40)
+```
+and a clean-cache re-run:
+```
+61 passed, 1 warning in 100.82s (0:01:40)
+```
+
+No regressions on trio backend (same subset):
+```
+69 passed, 1 skipped, 3 warnings in 89.19s
+```
+
+### Commit msg
+
+Also AI-drafted via `/commit-msg` + `rewrap.py
+--width 67`. See `git log -1 26fb820`.

ai/prompt-io/claude/20260420T192739Z_5e8cd8b2_prompt_io.md

@@ -0,0 +1,111 @@
+---
+model: claude-opus-4-7[1m]
+service: claude
+session: subint-phase-b-hang-classification
+timestamp: 2026-04-20T19:27:39Z
+git_ref: HEAD (pre-commit; on branch `subint_spawner_backend`)
+scope: docs
+substantive: true
+raw_file: 20260420T192739Z_5e8cd8b2_prompt_io.raw.md
+---
+
+## Prompt
+
+Collab between user (`goodboy`) and `claude` to
+disambiguate two distinct hang modes hit during
+Phase B subint-spawn-backend bringup (issue #379).
+The user ran the failing suites, captured `strace`
+evidence on hung pytest pids, and set the framing:
+
+> "These cannot be the same bug. Different fix
+> paths. Write them up separately or we'll keep
+> conflating them."
+
+Follow-on asks:
+- Cross-link each doc from its reproducer test so
+  a future triager lands on the analysis without
+  needing to grep `ai/conc-anal/`.
+- On `test_stale_entry_is_deleted`: arm
+  `dump_on_hang(seconds=20, path=...)` so a
+  regression captures a stack dump (pytest stderr
+  capture otherwise eats `faulthandler` output).
+- Keep `test_stale_entry_is_deleted` un-`skip`ped
+  so the dump file remains inspectable.
+
+Defer the actual fix for the class-2 hang
+(Ctrl-C-able, tractor-side bug) to a later commit
+per the incremental-commits preference.
+
+## Response summary
+
+Wrote two new `ai/conc-anal/` classification docs
+and cross-linked them from the reproducer tests:
+
+1. **`subint_sigint_starvation_issue.md`** —
+   class 1: abandoned-legacy-subint thread +
+   shared GIL starves main trio loop →
+   signal-wakeup-fd pipe fills → SIGINT silently
+   dropped (`write() = EAGAIN`). Pytest process
+   un-Ctrl-C-able. Structurally a CPython limit;
+   blocked on `msgspec` PEP 684 support
+   (jcrist/msgspec#563). Reproducer:
+   `test_stale_entry_is_deleted[subint]`.
+
+2. **`subint_cancel_delivery_hang_issue.md`** —
+   class 2: parent-side trio task parks on an
+   orphaned IPC channel after subint teardown;
+   no clean EOF delivered to waiting receiver.
+   Ctrl-C-able (main trio loop iterating fine).
+   OUR bug to fix. Candidate fix: explicit
+   parent-side channel abort in `subint_proc`'s
+   hard-kill teardown. Reproducer:
+   `test_subint_non_checkpointing_child`.
+
+Test-side cross-links:
+- `tests/discovery/test_registrar.py`:
+  `test_stale_entry_is_deleted` → `trio.run(main)`
+  wrapped in `dump_on_hang(seconds=20,
+  path=<per-method-tmp>)`; long inline comment
+  summarizes `strace` evidence + root-cause chain
+  and points at both docs.
+- `tests/test_subint_cancellation.py`:
+  `test_subint_non_checkpointing_child` docstring
+  extended with "KNOWN ISSUE (Ctrl-C-able hang)"
+  section pointing at the class-2 doc + noting
+  the class-1 doc is NOT what this test hits.
+
+## Files changed
+
+- `ai/conc-anal/subint_sigint_starvation_issue.md`
+  — new, 205 LOC
+- `ai/conc-anal/subint_cancel_delivery_hang_issue.md`
+  — new, 161 LOC
+- `tests/discovery/test_registrar.py` — +52/-1
+  (arm `dump_on_hang`, inline-comment cross-link)
+- `tests/test_subint_cancellation.py` — +26
+  (docstring "KNOWN ISSUE" block)
+
+## Human edits
+
+Substantive collab — prose was jointly iterated:
+
+- User framed the two-doc split, set the
+  classification criteria (Ctrl-C-able vs not),
+  and provided the `strace` evidence.
+- User decided to keep `test_stale_entry_is_deleted`
+  un-`skip`ped (my initial suggestion was
+  `pytestmark.skipif(spawn_backend=='subint')`).
+- User chose the candidate fix ordering for
+  class 2 and marked "explicit parent-side channel
+  abort" as the surgical preferred fix.
+- User picked the file naming convention
+  (`subint_<hang-shape>_issue.md`) over my initial
+  `hang_class_{1,2}.md`.
+- Assistant drafted the prose, aggregated prior-
+  session root-cause findings from Phase B.2/B.3
+  bringup, and wrote the test-side cross-linking
+  comments.
+
+No further mechanical edits expected before
+commit; user may still rewrap via
+`scripts/rewrap.py` if preferred.

ai/prompt-io/claude/20260420T192739Z_5e8cd8b2_prompt_io.raw.md

@@ -0,0 +1,198 @@
+---
+model: claude-opus-4-7[1m]
+service: claude
+timestamp: 2026-04-20T19:27:39Z
+git_ref: HEAD (pre-commit; will land on branch `subint_spawner_backend`)
+diff_cmd: git diff HEAD~1..HEAD
+---
+
+Collab between `goodboy` (user) and `claude` (this
+assistant) spanning multiple test-run iterations on
+branch `subint_spawner_backend`. The user ran the
+failing suites, captured `strace` evidence on the
+hung pytest pids, and set the direction ("these are
+two different hangs — write them up separately so
+we don't re-confuse ourselves later"). The assistant
+aggregated prior-session findings (Phase B.2/B.3
+bringup) into two classification docs + test-side
+cross-links. All prose was jointly iterated; the
+user had final say on framing and decided which
+candidate fix directions to list.
+
+## Per-file generated content
+
+### `ai/conc-anal/subint_sigint_starvation_issue.md` (new, 205 LOC)
+
+> `git diff HEAD~1..HEAD -- ai/conc-anal/subint_sigint_starvation_issue.md`
+
+Writes up the "abandoned-legacy-subint thread wedges
+the parent trio loop" class. Key sections:
+
+- **Symptom** — `test_stale_entry_is_deleted[subint]`
+  hangs indefinitely AND is un-Ctrl-C-able.
+- **Evidence** — annotated `strace` excerpt showing
+  SIGINT delivered to pytest, C-level signal handler
+  tries to write to the signal-wakeup-fd pipe, gets
+  `write() = -1 EAGAIN (Resource temporarily
+  unavailable)`. Pipe is full because main trio loop
+  isn't iterating often enough to drain it.
+- **Root-cause chain** — our hard-kill abandons the
+  `daemon=True` driver OS thread after
+  `_HARD_KILL_TIMEOUT`; the subint *inside* that
+  thread is still running `trio.run()`;
+  `_interpreters.destroy()` cannot force-stop a
+  running subint (raises `InterpreterError`); legacy
+  subints share the main GIL → abandoned subint
+  starves main trio loop → wakeup-fd fills → SIGINT
+  silently dropped.
+- **Why it's structurally a CPython limit** — no
+  public force-destroy primitive for a running
+  subint; the only escape is per-interpreter GIL
+  isolation, gated on msgspec PEP 684 adoption
+  (jcrist/msgspec#563).
+- **Current escape hatch** — harness-side SIGINT
+  loop in the `daemon` fixture teardown that kills
+  the bg registrar subproc, eventually unblocking
+  a parent-side recv enough for the main loop to
+  drain the wakeup pipe.
+
+### `ai/conc-anal/subint_cancel_delivery_hang_issue.md` (new, 161 LOC)
+
+> `git diff HEAD~1..HEAD -- ai/conc-anal/subint_cancel_delivery_hang_issue.md`
+
+Writes up the *sibling* hang class — same subint
+backend, distinct root cause:
+
+- **TL;DR** — Ctrl-C-able, so NOT the SIGINT-
+  starvation class; main trio loop iterates fine;
+  ours to fix.
+- **Symptom** — `test_subint_non_checkpointing_child`
+  hangs past the expected `_HARD_KILL_TIMEOUT`
+  budget even after the subint is torn down.
+- **Diagnosis** — a parent-side trio task (likely
+  a `chan.recv()` in `process_messages`) parks on
+  an orphaned IPC channel; channel was torn down
+  without emitting a clean EOF /
+  `BrokenResourceError` to the waiting receiver.
+- **Candidate fix directions** — listed in rough
+  order of preference:
+  1. Explicit parent-side channel abort in
+     `subint_proc`'s hard-kill teardown (surgical;
+     most likely).
+  2. Audit `process_messages` to add a timeout or
+     cancel-scope protection that catches the
+     orphaned-recv state.
+  3. Wrap subint IPC channel construction in a
+     sentinel that can force-close from the parent
+     side regardless of subint liveness.
+
+### `tests/discovery/test_registrar.py` (modified, +52/-1 LOC)
+
+> `git diff HEAD~1..HEAD -- tests/discovery/test_registrar.py`
+
+Wraps the `trio.run(main)` call at the bottom of
+`test_stale_entry_is_deleted` in
+`dump_on_hang(seconds=20, path=<per-method-tmp>)`.
+Adds a long inline comment that:
+- Enumerates variant-by-variant status
+  (`[trio]`/`[mp_*]` = clean; `[subint]` = hangs
+  + un-Ctrl-C-able)
+- Summarizes the `strace` evidence and root-cause
+  chain inline (so a future reader hitting this
+  test doesn't need to cross-ref the doc to
+  understand the hang shape)
+- Points at
+  `ai/conc-anal/subint_sigint_starvation_issue.md`
+  for full analysis
+- Cross-links to the *sibling*
+  `subint_cancel_delivery_hang_issue.md` so
+  readers can tell the two classes apart
+- Explains why it's kept un-`skip`ped: the dump
+  file is useful if the hang ever returns after
+  a refactor. pytest stderr capture would
+  otherwise eat `faulthandler` output, hence the
+  file path.
+
+### `tests/test_subint_cancellation.py` (modified, +26 LOC)
+
+> `git diff HEAD~1..HEAD -- tests/test_subint_cancellation.py`
+
+Extends the docstring of
+`test_subint_non_checkpointing_child` with a
+"KNOWN ISSUE (Ctrl-C-able hang)" block:
+- Describes the current hang: parent-side orphaned
+  IPC recv after hard-kill; distinct from the
+  SIGINT-starvation sibling class.
+- Cites `strace` distinguishing signal: wakeup-fd
+  `write() = 1` (not `EAGAIN`) — i.e. main loop
+  iterating.
+- Points at
+  `ai/conc-anal/subint_cancel_delivery_hang_issue.md`
+  for full analysis + candidate fix directions.
+- Clarifies that the *other* sibling doc
+  (SIGINT-starvation) is NOT what this test hits.
+
+## Non-code output
+
+### Classification reasoning (why two docs, not one)
+
+The user and I converged on the two-doc split after
+running the suites and noticing two *qualitatively
+different* hang symptoms:
+
+1. `test_stale_entry_is_deleted[subint]` — pytest
+   process un-Ctrl-C-able. Ctrl-C at the terminal
+   does nothing. Must kill-9 from another shell.
+2. `test_subint_non_checkpointing_child` — pytest
+   process Ctrl-C-able. One Ctrl-C at the prompt
+   unblocks cleanly and the test reports a hang
+   via pytest-timeout.
+
+From the user: "These cannot be the same bug.
+Different fix paths. Write them up separately or
+we'll keep conflating them."
+
+`strace` on the `[subint]` hang gave the decisive
+signal for the first class:
+
+```
+--- SIGINT {si_signo=SIGINT, si_code=SI_KERNEL} ---
+write(5, "\2", 1) = -1 EAGAIN (Resource temporarily unavailable)
+```
+
+fd 5 is Python's signal-wakeup-fd pipe. `EAGAIN`
+on a `write()` of 1 byte to a pipe means the pipe
+buffer is full → reader side (main Python thread
+inside `trio.run()`) isn't consuming. That's the
+GIL-hostage signature.
+
+The second class's `strace` showed `write(5, "\2",
+1) = 1` — clean drain — so the main trio loop was
+iterating and the hang had to be on the application
+side of things, not the kernel-↔-Python signal
+boundary.
+
+### Why the candidate fix for class 2 is "explicit parent-side channel abort"
+
+The second hang class has the trio loop alive. A
+parked `chan.recv()` that will never get bytes is
+fundamentally a tractor-side resource-lifetime bug
+— the IPC channel was torn down (subint destroyed)
+but no one explicitly raised
+`BrokenResourceError` at the parent-side receiver.
+The `subint_proc` hard-kill path is the natural
+place to add that notification, because it already
+knows the subint is unreachable at that point.
+
+Alternative fix paths (blanket timeouts on
+`process_messages`, sentinel-wrapped channels) are
+less surgical and risk masking unrelated bugs —
+hence the preference ordering in the doc.
+
+### Why we're not just patching the code now
+
+The user explicitly deferred the fix to a later
+commit: "Document both classes now, land the fix
+for class 2 separately so the diff reviews clean."
+This matches the incremental-commits preference
+from memory.

ai/prompt-io/claude/20260422T200723Z_797f57c_prompt_io.md

@@ -0,0 +1,155 @@
+---
+model: claude-opus-4-7[1m]
+service: claude
+session: subints-phase-b-hardening-and-fork-block
+timestamp: 2026-04-22T20:07:23Z
+git_ref: 797f57c
+scope: code
+substantive: true
+raw_file: 20260422T200723Z_797f57c_prompt_io.raw.md
+---
+
+## Prompt
+
+Session-spanning work on the Phase B `subint` spawn-backend.
+Three distinct sub-phases in one log:
+
+1. **Py3.13 gate tightening** — diagnose a reproducible hang
+   of subint spawn flow under py3.13 (works on py3.14), trace
+   to a private `_interpreters` module vintage issue, tighten
+   our feature gate from "`_interpreters` present" to "public
+   `concurrent.interpreters` present" (i.e. py3.14+).
+2. **Test-harness hardening** — add `pytest-timeout` dep, put
+   `@pytest.mark.timeout(30, method='thread')` on the
+   three known-hanging subint tests cataloged in
+   `ai/conc-anal/subint_sigint_starvation_issue.md`. Separately,
+   code-review the user's in-flight `skipon_spawn_backend`
+   marker implementation; find four bugs; refactor to use
+   `item.iter_markers()`.
+3. **`subint_fork` prototype → CPython-block finding** — draft
+   a WIP `subint_fork_proc` backend using a sub-interpreter as
+   a launchpad for `os.fork()` (to sidestep trio#1614). User
+   tests on py3.14, hits
+   `Fatal Python error: _PyInterpreterState_DeleteExceptMain:
+   not main interpreter`. Walk CPython sources (local clone at
+   `~/repos/cpython/`) to pinpoint the refusal
+   (`Modules/posixmodule.c:728` → `Python/pystate.c:1040`).
+   Revert implementation to a `NotImplementedError` stub in a
+   new `_subint_fork.py` submodule, document the finding in a
+   third `conc-anal/` doc with an upstream-report draft for
+   the CPython issue tracker. Finally, discuss user's proposed
+   workaround architecture (main-interp worker-thread
+   forkserver) and draft a standalone smoke-test script for
+   feasibility validation.
+
+## Response summary
+
+All three sub-phases landed concrete artifacts:
+
+**Sub-phase 1** — `_subint.py` + `_spawn.py` gates + error
+messages updated to require py3.14+ via the public
+`concurrent.interpreters` module presence check. Module
+docstring revised to explain the empirical reason
+(py3.13's private `_interpreters` vintage wedges under
+multi-trio-task usage even though minimal standalone
+reproducers work fine there). Test-module
+`pytest.importorskip` likewise switched.
+
+**Sub-phase 2** — `pytest-timeout>=2.3` added to `testing`
+dep group. `@pytest.mark.timeout(30, method='thread')`
+applied on:
+- `tests/discovery/test_registrar.py::test_stale_entry_is_deleted`
+- `tests/test_cancellation.py::test_cancel_while_childs_child_in_sync_sleep`
+- `tests/test_cancellation.py::test_multierror_fast_nursery`
+- `tests/test_subint_cancellation.py::test_subint_non_checkpointing_child`
+
+`method='thread'` documented inline as load-bearing — the
+GIL-starvation path that drops `SIGINT` would equally drop
+`SIGALRM`, so only a watchdog-thread timeout can reliably
+escape.
+
+`skipon_spawn_backend` plugin refactored into a single
+`iter_markers`-driven loop in `pytest_collection_modifyitems`
+(~30 LOC replacing ~30 LOC of nested conditionals). Four
+bugs dissolved: wrong `.get()` key, module-level `pytestmark`
+suppressing per-test marks, unhandled `pytestmark = [list]`
+form, `pytest.Makr` typo. Marker help text updated to
+document the variadic backend-list + `reason=` kwarg
+surface.
+
+**Sub-phase 3** — Prototype drafted (then reverted):
+
+- `tractor/spawn/_subint_fork.py` — new dedicated submodule
+  housing the `subint_fork_proc` stub. Module docstring +
+  fn docstring explain the attempt, the CPython-level
+  block, and the reason for keeping the stub in-tree
+  (documentation of the attempt + starting point if CPython
+  ever lifts the restriction).
+- `tractor/spawn/_spawn.py` — `'subint_fork'` registered as a
+  `SpawnMethodKey` literal + in `_methods`, so
+  `--spawn-backend=subint_fork` routes to a clean
+  `NotImplementedError` pointing at the analysis doc rather
+  than an "invalid backend" error.
+- `ai/conc-anal/subint_fork_blocked_by_cpython_post_fork_issue.md` —
+  third sibling conc-anal doc. Full annotated CPython
+  source walkthrough from user-visible
+  `Fatal Python error` → `Modules/posixmodule.c:728
+  PyOS_AfterFork_Child()` → `Python/pystate.c:1040
+  _PyInterpreterState_DeleteExceptMain()` gate. Includes a
+  copy-paste-ready upstream-report draft for the CPython
+  issue tracker with a two-tier ask (ideally "make it work",
+  minimally "cleaner error than `Fatal Python error`
+  aborting the child").
+- `ai/conc-anal/subint_fork_from_main_thread_smoketest.py` —
+  standalone zero-tractor-import CPython-level smoke test
+  for the user's proposed workaround architecture
+  (forkserver on a main-interp worker thread). Four
+  argparse-driven scenarios: `control_subint_thread_fork`
+  (reproduces the known-broken case as a test-harness
+  sanity),  `main_thread_fork` (baseline), `worker_thread_fork`
+  (architectural assertion), `full_architecture`
+  (end-to-end trio-in-subint in forked child). User will
+  run on py3.14 next.
+
+## Files changed
+
+See `git log 26fb820..HEAD --stat` for the canonical list.
+New files this session:
+- `tractor/spawn/_subint_fork.py`
+- `ai/conc-anal/subint_fork_blocked_by_cpython_post_fork_issue.md`
+- `ai/conc-anal/subint_fork_from_main_thread_smoketest.py`
+
+Modified (diff pointers in raw log):
+- `tractor/spawn/_subint.py` (py3.14 gate)
+- `tractor/spawn/_spawn.py` (`subint_fork` registration)
+- `tractor/_testing/pytest.py` (`skipon_spawn_backend` refactor)
+- `pyproject.toml` (`pytest-timeout` dep)
+- `tests/discovery/test_registrar.py`,
+  `tests/test_cancellation.py`,
+  `tests/test_subint_cancellation.py` (timeout marks,
+  cross-refs to conc-anal docs)
+
+## Human edits
+
+Several back-and-forth iterations with user-driven
+adjustments during the session:
+
+- User corrected my initial mis-classification of
+  `test_cancel_while_childs_child_in_sync_sleep[subint-False]`
+  as Ctrl-C-able — second strace showed `EAGAIN`, putting
+  it squarely in class A (GIL-starvation). Re-analysis
+  preserved in the raw log.
+- User independently fixed the `.get(reason)` → `.get('reason', reason)`
+  bug in the marker plugin before my review; preserved their
+  fix.
+- User suggested moving the `subint_fork_proc` stub from
+  the bottom of `_subint.py` into its own
+  `_subint_fork.py` submodule — applied.
+- User asked to keep the forkserver-architecture
+  discussion as background for the smoke-test rather than
+  committing to a tractor-side refactor until the smoke
+  test validates the CPython-level assumptions.
+
+Commit messages in this range (b025c982 … 797f57c) were
+drafted via `/commit-msg` + `rewrap.py --width 67`; user
+landed them with the usual review.

ai/prompt-io/claude/20260422T200723Z_797f57c_prompt_io.raw.md

@@ -0,0 +1,343 @@
+---
+model: claude-opus-4-7[1m]
+service: claude
+timestamp: 2026-04-22T20:07:23Z
+git_ref: 797f57c
+diff_cmd: git log 26fb820..HEAD  # all session commits since the destroy-race fix log
+---
+
+Session-spanning conversation covering the Phase B hardening
+of the `subint` spawn-backend and an investigation into a
+proposed `subint_fork` follow-up which turned out to be
+blocked at the CPython level. This log is a narrative capture
+of the substantive turns (not every message) and references
+the concrete code + docs the session produced. Per diff-ref
+mode the actual code diffs are pointed at via `git log` on
+each ref rather than duplicated inline.
+
+## Narrative of the substantive turns
+
+### Py3.13 hang / gate tightening
+
+Diagnosed a reproducible hang of the `subint` backend under
+py3.13 (test_spawning tests wedge after root-actor bringup).
+Root cause: py3.13's vintage of the private `_interpreters` C
+module has a latent thread/subint-interaction issue that
+`_interpreters.exec()` silently fails to progress under
+tractor's multi-trio usage pattern — even though a minimal
+standalone `threading.Thread` + `_interpreters.exec()`
+reproducer works fine on the same Python. Empirically
+py3.14 fixes it.
+
+Fix (from this session): tighten the `_has_subints` gate in
+`tractor.spawn._subint` from "private module importable" to
+"public `concurrent.interpreters` present" — which is 3.14+
+only. This leaves `subint_proc()` unchanged in behavior (we
+still call the *private* `_interpreters.create('legacy')`
+etc. under the hood) but refuses to engage on 3.13.
+
+Also tightened the matching gate in
+`tractor.spawn._spawn.try_set_start_method('subint')` and
+rev'd the corresponding error messages from "3.13+" to
+"3.14+" with a sentence explaining why. Test-module
+`pytest.importorskip` switched from `_interpreters` →
+`concurrent.interpreters` to match.
+
+### `pytest-timeout` dep + `skipon_spawn_backend` marker plumbing
+
+Added `pytest-timeout>=2.3` to the `testing` dep group with
+an inline comment pointing at the `ai/conc-anal/*.md` docs.
+Applied `@pytest.mark.timeout(30, method='thread')` (the
+`method='thread'` is load-bearing — `signal`-method
+`SIGALRM` suffers the same GIL-starvation path that drops
+`SIGINT` in the class-A hang pattern) to the three known-
+hanging subint tests cataloged in
+`subint_sigint_starvation_issue.md`.
+
+Separately code-reviewed the user's newly-staged
+`skipon_spawn_backend` pytest marker implementation in
+`tractor/_testing/pytest.py`. Found four bugs:
+
+1. `modmark.kwargs.get(reason)` called `.get()` with the
+   *variable* `reason` as the dict key instead of the string
+   `'reason'` — user-supplied `reason=` was never picked up.
+   (User had already fixed this locally via `.get('reason',
+   reason)` by the time my review happened — preserved that
+   fix.)
+2. The module-level `pytestmark` branch suppressed per-test
+   marker handling (the `else:` was an `else:` rather than
+   independent iteration).
+3. `mod_pytestmark.mark` assumed a single
+   `MarkDecorator` — broke on the valid-pytest `pytestmark =
+   [mark, mark]` list form.
+4. Typo: `pytest.Makr` → `pytest.Mark`.
+
+Refactored the hook to use `item.iter_markers(name=...)`
+which walks function + class + module scopes uniformly and
+handles both `pytestmark` forms natively. ~30 LOC replaced
+the original ~30 LOC of nested conditionals, all four bugs
+dissolved. Also updated the marker help string to reflect
+the variadic `*start_methods` + `reason=` surface.
+
+### `subint_fork_proc` prototype attempt
+
+User's hypothesis: the known trio+`fork()` issues
+(python-trio/trio#1614) could be sidestepped by using a
+sub-interpreter purely as a launchpad — `os.fork()` from a
+subint that has never imported trio → child is in a
+trio-free context. In the child `execv()` back into
+`python -m tractor._child` and the downstream handshake
+matches `trio_proc()` identically.
+
+Drafted the prototype at `tractor/spawn/_subint.py`'s bottom
+(originally — later moved to its own submod, see below):
+launchpad-subint creation, bootstrap code-string with
+`os.fork()` + `execv()`, driver-thread orchestration,
+parent-side `ipc_server.wait_for_peer()` dance. Registered
+`'subint_fork'` as a new `SpawnMethodKey` literal, added
+`case 'subint' | 'subint_fork':` feature-gate arm in
+`try_set_start_method()`, added entry in `_methods` dict.
+
+### CPython-level block discovered
+
+User tested on py3.14 and saw:
+
+```
+Fatal Python error: _PyInterpreterState_DeleteExceptMain: not main interpreter
+Python runtime state: initialized
+
+Current thread 0x00007f6b71a456c0 [subint-fork-lau] (most recent call first):
+  File "<script>", line 2 in <module>
+<script>:2: DeprecationWarning: This process (pid=802985) is multi-threaded, use of fork() may lead to deadlocks in the child.
+```
+
+Walked CPython sources (local clone at `~/repos/cpython/`):
+
+- **`Modules/posixmodule.c:728` `PyOS_AfterFork_Child()`** —
+  post-fork child-side cleanup. Calls
+  `_PyInterpreterState_DeleteExceptMain(runtime)` with
+  `goto fatal_error` on non-zero status. Has the
+  `// Ideally we could guarantee tstate is running main.`
+  self-acknowledging-fragile comment directly above.
+
+- **`Python/pystate.c:1040`
+  `_PyInterpreterState_DeleteExceptMain()`** — the
+  refusal. Hard `PyStatus_ERR("not main interpreter")` gate
+  when `tstate->interp != interpreters->main`. Docstring
+  formally declares the precondition ("If there is a
+  current interpreter state, it *must* be the main
+  interpreter"). `XXX` comments acknowledge further latent
+  issues within.
+
+Definitive answer to "Open Question 1" of the prototype
+docstring: **no, CPython does not support `os.fork()` from
+a non-main sub-interpreter**. Not because the fork syscall
+is blocked (it isn't — the parent returns a valid pid),
+but because the child cannot survive CPython's post-fork
+initialization. This is an enforced invariant, not an
+incidental limitation.
+
+### Revert: move to stub submod + doc the finding
+
+Per user request:
+
+1. Reverted the working `subint_fork_proc` body to a
+   `NotImplementedError` stub, MOVED to its own submod
+   `tractor/spawn/_subint_fork.py` (keeps `_subint.py`
+   focused on the working `subint_proc` backend).
+2. Updated `_spawn.py` to import the stub from the new
+   submod path; kept `'subint_fork'` in `SpawnMethodKey` +
+   `_methods` so `--spawn-backend=subint_fork` routes to a
+   clean `NotImplementedError` with pointer to the analysis
+   doc rather than an "invalid backend" error.
+3. Wrote
+   `ai/conc-anal/subint_fork_blocked_by_cpython_post_fork_issue.md`
+   with the full annotated CPython walkthrough + an
+   upstream-report draft for the CPython issue tracker.
+   Draft has a two-tier ask: ideally "make it work"
+   (pre-fork tstate-swap hook or `DeleteExceptFor(interp)`
+   variant), minimally "give us a clean `RuntimeError` in
+   the parent instead of a `Fatal Python error` aborting
+   the child silently".
+
+### Design discussion — main-interp-thread forkserver workaround
+
+User proposed: set up a "subint forking server" that fork()s
+on behalf of subint callers. Core insight: the CPython gate
+is on `tstate->interp`, not thread identity, so **any thread
+whose tstate is main-interp** can fork cleanly. A worker
+thread attached to main-interp (never entering a subint)
+satisfies the precondition.
+
+Structurally this is `mp.forkserver` (which tractor already
+has as `mp_forkserver`) but **in-process**: instead of a
+separate Python subproc as the fork server, we'd put the
+forkserver on a thread in the tractor parent process. Pros:
+faster spawn (no IPC marshalling to external server + no
+separate Python startup), inherits already-imported modules
+for free. Cons: less crash isolation (forkserver failure
+takes the whole process).
+
+Required tractor-side refactor: move the root actor's
+`trio.run()` off main-interp-main-thread (so main-thread can
+run the forkserver loop). Nontrivial; approximately the same
+magnitude as "Phase C".
+
+The design would also not fully resolve the class-A
+GIL-starvation issue because child actors' trio still runs
+inside subints (legacy config, msgspec PEP 684 pending).
+Would mitigate SIGINT-starvation specifically if signal
+handling moves to the forkserver thread.
+
+Recommended pre-commitment: a standalone CPython-only smoke
+test validating the four assumptions the arch rests on,
+before any tractor-side work.
+
+### Smoke-test script drafted
+
+Wrote `ai/conc-anal/subint_fork_from_main_thread_smoketest.py`:
+argparse-driven, four scenarios (`control_subint_thread_fork`
+reproducing the known-broken case, `main_thread_fork`
+baseline, `worker_thread_fork` the architectural assertion,
+`full_architecture` end-to-end with trio in a subint in the
+forked child). No `tractor` imports; pure CPython + `_interpreters`
++ `trio`. Bails cleanly on py<3.14. Pass/fail banners per
+scenario.
+
+User will validate on their py3.14 env next.
+
+## Per-code-artifact provenance
+
+### `tractor/spawn/_subint_fork.py` (new submod)
+
+> `git show 797f57c -- tractor/spawn/_subint_fork.py`
+
+NotImplementedError stub for the subint-fork backend. Module
+docstring + fn docstring explain the attempt, the CPython
+block, and why the stub is kept in-tree. No runtime behavior
+beyond raising with a pointer at the conc-anal doc.
+
+### `tractor/spawn/_spawn.py` (modified)
+
+> `git log 26fb820..HEAD -- tractor/spawn/_spawn.py`
+
+- Added `'subint_fork'` to `SpawnMethodKey` literal with a
+  block comment explaining the CPython-level block.
+- Generalized the `case 'subint':` arm to `case 'subint' |
+  'subint_fork':` since both use the same py3.14+ gate.
+- Registered `subint_fork_proc` in `_methods` with a
+  pointer-comment at the analysis doc.
+
+### `tractor/spawn/_subint.py` (modified across session)
+
+> `git log 26fb820..HEAD -- tractor/spawn/_subint.py`
+
+- Tightened `_has_subints` gate: dual-requires public
+  `concurrent.interpreters` + private `_interpreters`
+  (tests for py3.14-or-newer on the public-API presence,
+  then uses the private one for legacy-config subints
+  because `msgspec` still blocks the public isolated mode
+  per jcrist/msgspec#563).
+- Updated module docstring, `subint_proc()` docstring, and
+  gate-error messages to reflect the 3.14+ requirement and
+  the reason (py3.13 wedges under multi-trio usage even
+  though the private module exists there).
+
+### `tractor/_testing/pytest.py` (modified)
+
+> `git log 26fb820..HEAD -- tractor/_testing/pytest.py`
+
+- New `skipon_spawn_backend(*start_methods, reason=...)`
+  pytest marker expanded into `pytest.mark.skip(reason=...)`
+  at collection time via
+  `pytest_collection_modifyitems()`.
+- Implementation uses `item.iter_markers(name=...)` which
+  walks function + class + module scopes uniformly and
+  handles both `pytestmark = <single Mark>` and
+  `pytestmark = [mark, ...]` forms natively. ~30-LOC
+  single-loop refactor replacing a prior nested
+  conditional that had four bugs (see "Review" narrative
+  above).
+- Added `pytest.Config` / `pytest.Function` /
+  `pytest.FixtureRequest` type annotations on fixture
+  signatures while touching the file.
+
+### `pyproject.toml` (modified)
+
+> `git log 26fb820..HEAD -- pyproject.toml`
+
+Added `pytest-timeout>=2.3` to `testing` dep group with
+comment pointing at the `ai/conc-anal/` docs.
+
+### `tests/discovery/test_registrar.py`,
+`tests/test_subint_cancellation.py`,
+`tests/test_cancellation.py` (modified)
+
+> `git log 26fb820..HEAD -- tests/`
+
+Applied `@pytest.mark.timeout(30, method='thread')` on
+known-hanging subint tests. Extended comments to cross-
+reference the `ai/conc-anal/*.md` docs. `method='thread'`
+is documented inline as load-bearing (`signal`-method
+SIGALRM suffers the same GIL-starvation path that drops
+SIGINT).
+
+### `ai/conc-anal/subint_fork_blocked_by_cpython_post_fork_issue.md` (new)
+
+> `git show 797f57c -- ai/conc-anal/subint_fork_blocked_by_cpython_post_fork_issue.md`
+
+Third sibling doc under `conc-anal/`. Structure: TL;DR,
+context ("what we tried"), symptom (the user's exact
+`Fatal Python error` output), CPython source walkthrough
+with excerpted snippets from `posixmodule.c` +
+`pystate.c`, chain summary, definitive answer to Open
+Question 1, `## Upstream-report draft (for CPython issue
+tracker)` section with a two-tier ask, references.
+
+### `ai/conc-anal/subint_fork_from_main_thread_smoketest.py` (new, THIS turn)
+
+Zero-tractor-import smoke test for the proposed workaround
+architecture. Four argparse-driven scenarios covering the
+control case + baseline + arch-critical case + end-to-end.
+Pass/fail banners per scenario; clean `--help` output;
+py3.13 early-exit.
+
+## Non-code output (verbatim)
+
+### The `strace` signature that kicked off the CPython
+walkthrough
+
+```
+--- SIGINT {si_signo=SIGINT, si_code=SI_KERNEL} ---
+write(16, "\2", 1)                      = -1 EAGAIN (Resource temporarily unavailable)
+rt_sigreturn({mask=[WINCH]})            = 139801964688928
+```
+
+### Key user quotes framing the direction
+
+> ok actually we get this [fatal error] ... see if you can
+> take a look at what's going on, in particular wrt to
+> cpython's sources. pretty sure there's a local copy at
+> ~/repos/cpython/
+
+(Drove the CPython walkthrough that produced the
+definitive refusal chain.)
+
+> is there any reason we can't just sidestep this "must fork
+> from main thread in main subint" issue by simply ensuring
+> a "subint forking server" is always setup prior to
+> invoking trio in a non-main-thread subint ...
+
+(Drove the main-interp-thread-forkserver architectural
+discussion + smoke-test script design.)
+
+### CPython source tags for quick jump-back
+
+```
+Modules/posixmodule.c:728   PyOS_AfterFork_Child()
+Modules/posixmodule.c:753   // Ideally we could guarantee tstate is running main.
+Modules/posixmodule.c:778   status = _PyInterpreterState_DeleteExceptMain(runtime);
+
+Python/pystate.c:1040       _PyInterpreterState_DeleteExceptMain()
+Python/pystate.c:1044-1047  tstate->interp != main → PyStatus_ERR("not main interpreter")
+```

ai/prompt-io/claude/20260601T231429Z_0e3e008b_prompt_io.md

@@ -0,0 +1,146 @@
+---
+model: claude-opus-4-7[1m]
+service: claude
+session: trio-0.33-subproc-supervisor-retroactive
+timestamp: 2026-06-01T23:14:29Z
+git_ref: 0e3e008b
+scope: code
+substantive: true
+raw_file: 20260601T231429Z_0e3e008b_prompt_io.raw.md
+---
+
+## Prompt
+
+**RETROACTIVE LOG** — original session prompts not
+preserved; reconstructed from the staged work product.
+
+The work designs a `trio.Nursery.start()`-style wrapper
+around `trio.run_process()` for SC-friendly subprocess
+supervision. From the resulting code shape, the
+prompting intent was:
+
+1. Surface rc!=0 `CalledProcessError` DETERMINISTICALLY,
+   without the nursery-eg-wrapping that complicates
+   `collapse_eg()` usage and races the relay reader on
+   trio's `check=True`-driven cancel cascade.
+2. ALWAYS isolate the parent controlling-tty so a
+   spawned child can't emit terminal control-seqs onto
+   the launching tty (clobbering scrollback). Default
+   `stdin=DEVNULL`; default `stdout=DEVNULL` unless
+   explicitly relayed/overridden; distinguish "caller
+   passed nothing" from "caller passed `None` for
+   inherit".
+3. Optional live per-line relay of child std-streams to
+   the `tractor` log — STREAMED (not
+   buffered-until-exit) so long-lived daemon output is
+   visible during the run. Pick a custom log level that
+   shows at usual `info`/`devx` console levels but is
+   separately filterable.
+4. Concurrent pipe-drain reader MANDATORY when piping
+   without `capture_*` — without it the child blocks on
+   `write()` once the OS pipe buffer fills (~64KiB),
+   causing deadlocks on output bursts.
+5. Non-blocking `tn.start()` semantics: hand the live
+   `trio.Process` to the parent immediately;
+   supervise/relay run to completion in the supervisor
+   coro.
+6. Hermetic `trio`-only unit tests (no actor-runtime)
+   covering each of: per-line relay, tty isolation,
+   no-deadlock on >64KiB unnewlined output, CPE
+   rebuild w/ stderr relay, CPE rebuild on the silent
+   drain+capture path.
+
+## Response summary
+
+Adds `tractor/trionics/_subproc.py` (296 LOC) +
+`tests/trionics/test_subproc.py` (230 LOC) + a
+re-export in `tractor/trionics/__init__.py`.
+
+**`supervise_run_process()`** (public, re-exported)
+- `check=False` is forced to `trio.run_process`; the
+  rc-check runs in the supervisor coro AFTER `own_tn`
+  unwinds (both the child AND the relay readers have
+  hit EOF + fully drained). A BARE
+  `subprocess.CalledProcessError` is rebuilt + raised
+  from there, with `.stderr` bytes passed in the
+  constructor AND attached as an `add_note()`'d
+  `|_.stderr:` block for legible teardown logs.
+- `stdin=DEVNULL` always. `stdout` default chosen via a
+  `_UNSET` sentinel: `relay_stdout=True` → PIPE,
+  explicit `stdout=...` → as given, else `DEVNULL`.
+  `stderr` defaults to PIPE whenever we relay OR need
+  the CPE note (when `check=True`), else `DEVNULL`.
+- `relay_level='io'` (custom level 21; sorts just
+  above stdlib `INFO`=20 so it shows at usual
+  `info`/`devx` levels and stays separately
+  filterable). `runtime`=15 would silently filter at
+  default levels, so it's rejected as a default.
+- `task_status.started(trio_proc)` delivers the live
+  process immediately. The internal `own_tn`
+  supervises `trio.run_process` + any relay readers to
+  completion.
+- `**run_process_kwargs` forward verbatim;
+  `stdin/stdout/stderr/check` are MANAGED keys
+  (override on conflict).
+- Crash-handling deliberately NOT baked in — compose
+  `maybe_open_crash_handler()` on top at the call-site.
+
+**`_relay_stream_lines()`** (internal helper)
+- Three modes (combinable): `emit`-only (live per-line
+  relay), `accum`-only (silent drain+capture for a CPE
+  note), or both (live relay AND capture).
+- Per-line split handles cross-chunk residuals via a
+  rolling `residual` bytes buffer; flushes any trailing
+  un-newline-term'd line at EOF.
+- `async with stream:` ensures aclose at EOF/cancel
+  (mirrors trio's internal `_subprocess` drain idiom).
+
+**`_add_stderr_note()`** (internal helper)
+- `add_note()`s a `textwrap.indent(...)`'d
+  `|_.stderr:` block onto a `CalledProcessError` for
+  teardown logs.
+
+**Tests** (5 hermetic, trio-only) — `_capture_relay`
+fixture monkeypatches `_subproc.log.<level>` to a list:
+- `test_stdout_relayed_per_line`: per-line stdout
+  relay carries each `line=N` to the records.
+- `test_parent_tty_isolated`: `readlink /proc/self/fd/0`
+  and `fd/1` from the child show `pipe:` (fd1) +
+  `/dev/null` (fd0); NO `/dev/pts/*`.
+- `test_no_deadlock_on_big_unnewlined_output`: 200KiB
+  of `x` with no newlines completes inside
+  `fail_after(2)` — exercises the concurrent drain.
+- `test_stderr_relay_and_cpe_rebuild`: rc=3 with
+  `relay_stderr=True` raises bare CPE
+  (via `collapse_eg()`) with `b'boom' in cpe.stderr`,
+  the note attached, AND per-line live relay.
+- `test_nonrelay_cpe_note`: rc=7 with no relay still
+  produces CPE with `.stderr` + note via the silent
+  drain+capture path.
+
+## Files changed
+
+- `tractor/trionics/_subproc.py` — NEW. Public
+  `supervise_run_process()` + helpers
+  `_relay_stream_lines()` / `_add_stderr_note()` + the
+  `_UNSET` sentinel.
+- `tests/trionics/test_subproc.py` — NEW. 5 hermetic
+  trio-only tests + `_capture_relay` monkeypatch
+  fixture.
+- `tractor/trionics/__init__.py` — re-export
+  `supervise_run_process`.
+
+## Human edits
+
+**RETROACTIVE**: this log is being written from the
+staged diff, not from a live session. The code as
+staged is the canonical artifact; any human edits the
+user made during the originating design session are
+already integrated and cannot be separated post-hoc.
+The `.raw.md` sibling is a diff-pointer placeholder,
+NOT a pre-edit transcript.
+
+Future prompt-io entries for in-flight work should be
+written DURING the design session per the skill
+contract so the pre-edit `.raw.md` captures the
+unedited model output for genuine provenance.

ai/prompt-io/claude/20260601T231429Z_0e3e008b_prompt_io.raw.md

@@ -0,0 +1,106 @@
+---
+model: claude-opus-4-7[1m]
+service: claude
+timestamp: 2026-06-01T23:14:29Z
+git_ref: 0e3e008b
+diff_cmd: git diff HEAD~1..HEAD
+---
+
+# RETROACTIVE — original model output not preserved
+
+This `.raw.md` would normally contain the verbatim
+pre-human-edit response from the design session that
+produced the staged `_subproc.py` module + tests. That
+session's transcript is not available, so this file
+serves as a diff-pointer placeholder + transparency
+note.
+
+## Authoritative artifact
+
+The committed code IS the artifact of record. Once the
+companion commit lands, the unified diff is:
+
+> `git diff HEAD~1..HEAD -- tractor/trionics/_subproc.py`
+> `git diff HEAD~1..HEAD -- tests/trionics/test_subproc.py`
+> `git diff HEAD~1..HEAD -- tractor/trionics/__init__.py`
+
+Before committing, substitute `--cached` for the
+pre-commit form.
+
+## What is NOT here
+
+Because this is retroactive:
+- No verbatim chain-of-thought / discussion prose from
+  the design session.
+- No rejected alternatives the model considered before
+  arriving at the final shape (e.g. whether the
+  rc-check should live inside `own_tn` vs after it; the
+  `_UNSET` sentinel vs a `None`-means-DEVNULL
+  convention; `io` vs `info` as the default relay
+  level).
+- No pre-edit code blocks as the model first emitted
+  them, separable from any user cleanup applied before
+  the diff was staged.
+
+## Inferred design choices visible in the final code
+
+(Documented here because they're the kind of decision
+detail an unedited raw transcript would have captured.)
+
+1. **Post-drain rc-check in the supervisor coro body,
+   AFTER `own_tn.__aexit__`.** Placing the
+   `CalledProcessError` raise here (not inside
+   `own_tn`) means the EG-unwrap happens at the OUTER
+   `tn.start()` boundary — callers do `collapse_eg()`
+   if they want bare. Doing the raise INSIDE `own_tn`
+   would cancel the still-draining relay reader
+   mid-flight and lose stderr lines.
+
+2. **`_UNSET` sentinel for `stdout`.** A plain default
+   of `None` couldn't distinguish "use the safe
+   `DEVNULL` default" from "caller explicitly passed
+   `None` (inherit, presumably knowingly)". The
+   sentinel keeps the SAFE default while letting power
+   users opt into inherit.
+
+3. **`relay_level='io'` (custom level 21).** Chosen to
+   sort just above stdlib `INFO`=20 so a default
+   `--ll info` shows the relay, but it remains a
+   distinct level so users can filter
+   `tractor.trionics:io` separately. Picking
+   `runtime`=15 would have made the relay invisible at
+   default verbosity (a footgun for daemon supervisors
+   whose whole point is "I want to see this output").
+
+4. **Reader is MANDATORY, not opt-in cosmetic.** With
+   `stdout=PIPE` / `stderr=PIPE` we OWN the drain
+   responsibility — there's no `trio.capture_*` running
+   under the hood here. The ~64KiB OS pipe buffer
+   means a child writing more than that without us
+   reading hangs at `write()` — a deadlock that won't
+   show up in small-output tests, which is why the
+   200KiB-no-newline test is in the suite.
+
+5. **`task_status.started(trio_proc)` BEFORE the
+   `own_tn` exits.** Without this, `tn.start()` would
+   block until the child exits — losing the "start a
+   long-lived daemon and continue with parent work"
+   use case. With it, the parent gets the live process
+   handle immediately and the supervise+relay tasks
+   run in the supervisor coro until the child exits.
+
+6. **`__notes__` via `add_note()` for the CPE
+   `.stderr`.** The `.stderr` attribute is what
+   `subprocess` callers expect; the `add_note()` is
+   what trio's exception-rendering shows. Both wired so
+   programmatic AND human consumers see the stderr at
+   teardown.
+
+## Honesty statement
+
+This file's content is RECONSTRUCTED from the staged
+code, not extracted from a verbatim model transcript.
+The prompt-io skill's intent is for the `.raw.md` to
+be a pre-edit fossil; that's not possible here. Future
+work should write the prompt-io entry DURING the
+design session.

pyproject.toml

@@ -104,7 +104,7 @@ testing = [
 repl = [
   "pyperclip>=1.9.0",
   "prompt-toolkit>=3.0.50",
-  "xonsh>=0.23.0",
+  "xonsh>=0.23.8",
   "psutil>=7.0.0",
 ]
 lint = [

tests/devx/conftest.py

@@ -5,6 +5,7 @@
 from __future__ import annotations
 import platform
 import os
+import re
 import signal
 import time
 from typing import (
@@ -294,6 +295,26 @@ def expect(
 PROMPT = r"\(Pdb\+\)"
 
 
+# Strip terminal color / ANSI-VT100 escape sequences so
+# substring matching against REPL + traceback output stays
+# robust to color leakage — Python 3.13's colored tracebacks,
+# `pdbp`'s pygments highlighting, etc. — even when
+# `PYTHON_COLORS=0` (set in the `spawn` fixture) isn't honored
+# by every renderer in the spawned subproc.
+# Regex per https://stackoverflow.com/a/14693789
+_ansi_re: re.Pattern = re.compile(
+    r'\x1B(?:[@-Z\\-_]|\[[0-?]*[ -/]*[@-~])'
+)
+
+
+def ansi_strip(text: str) -> str:
+    '''
+    Remove ANSI/VT100 escape sequences from `text`.
+
+    '''
+    return _ansi_re.sub('', text)
+
+
 def in_prompt_msg(
     child: SpawnBase,
     parts: list[str],
@@ -313,7 +334,7 @@ def in_prompt_msg(
     '''
     __tracebackhide__: bool = False
 
-    before: str = str(child.before.decode())
+    before: str = ansi_strip(str(child.before.decode()))
     for part in parts:
         if part not in before:
             if pause_on_false:
@@ -333,9 +354,9 @@ def in_prompt_msg(
     return True
 
 
-# TODO: todo support terminal color-chars stripping so we can match
+# NB: color-char stripping (so we can match against call-stack
-# against call stack frame output from the the 'll' command the like!
+# frame output from the `ll` command and the like) is handled by
-# -[ ] SO answer for stipping ANSI codes: https://stackoverflow.com/a/14693789
+# `ansi_strip()` applied inside `in_prompt_msg()` + below.
 def assert_before(
     child: SpawnBase,
     patts: list[str],
@@ -356,7 +377,7 @@ def assert_before(
         err_on_false=True,
         **kwargs
     )
-    before: str = str(child.before.decode())
+    before: str = ansi_strip(str(child.before.decode()))
     return before
 
 

tests/devx/test_debugger.py

@@ -1186,7 +1186,12 @@ def test_shield_pause(
             "('cancelled_before_pause'",  # actor name
             _repl_fail_msg,
             "trio.Cancelled",
-            "raise Cancelled._create()",
+            # trio >=0.30 raises via a multi-line
+            # `raise Cancelled._create(source=.., reason=..,
+            #  source_task=..)` (cancel-reason metadata), so
+            # match the open-paren form only, NOT the legacy
+            # bare `()`.
+            "raise Cancelled._create(",
 
             # we should be handling a taskc inside
             # the first `.port_mortem()` sin-shield!
@@ -1204,7 +1209,12 @@ def test_shield_pause(
             "('root'",  # actor name
             _repl_fail_msg,
             "trio.Cancelled",
-            "raise Cancelled._create()",
+            # trio >=0.30 raises via a multi-line
+            # `raise Cancelled._create(source=.., reason=..,
+            #  source_task=..)` (cancel-reason metadata), so
+            # match the open-paren form only, NOT the legacy
+            # bare `()`.
+            "raise Cancelled._create(",
 
             # handling a taskc inside the first unshielded
             # `.port_mortem()`.

tests/devx/test_proctitle.py

@@ -8,9 +8,9 @@ after `Actor` construction, so any spawned sub-actor process
 should:
 
   - have `argv[0]` (== `/proc/<pid>/cmdline`) start with
-    `tractor[<aid.reprol()>]`
+    `<_def_prefix>[<aid.reprol()>]` (currently `_subactor[…]`)
-  - have `/proc/<pid>/comm` start with `tractor[` (kernel
+  - have `/proc/<pid>/comm` start with `<_def_prefix>[`
-    truncates to ~15 bytes)
+    (kernel truncates to ~15 bytes)
   - be detected as a tractor sub-actor by
     `_is_tractor_subactor(pid)` via the cmdline marker.
 
@@ -27,7 +27,10 @@ import trio
 import tractor
 
 from tractor.runtime._runtime import Actor
-from tractor.devx._proctitle import set_actor_proctitle
+from tractor.devx._proctitle import (
+    set_actor_proctitle,
+    _def_prefix,
+)
 from tractor._testing._reap import (
     _is_tractor_subactor,
     _read_cmdline,
@@ -41,8 +44,9 @@ _non_linux: bool = platform.system() != 'Linux'
 def test_set_actor_proctitle_format():
     '''
     `set_actor_proctitle()` returns the canonical
-    `tractor[<aid.reprol()>]` form and actually mutates
+    `<_def_prefix>[<aid.reprol()>]` form (currently
-    the running proc's title.
+    `_subactor[…]`) and actually mutates the running
+    proc's title.
 
     '''
     pytest.importorskip(
@@ -60,12 +64,14 @@ def test_set_actor_proctitle_format():
         )
         title: str = set_actor_proctitle(actor)
 
-        # canonical wrapping: `tractor[<aid.reprol()>]`. We
+        # canonical wrapping: `<_def_prefix>[<aid.reprol()>]`.
-        # compare against the runtime-computed `reprol()`
+        # We source BOTH the prefix (`_def_prefix`) and the
-        # rather than a hard-coded value so the test stays
+        # runtime-computed `reprol()` rather than hard-coding,
-        # decoupled from `Aid.reprol()`'s internal format
+        # so the test stays decoupled from the prefix shape
-        # (currently `<name>@<pid>`, but could evolve).
+        # (flipped to `_subactor` in `3a45dbd5`) AND from
-        expected: str = f'tractor[{actor.aid.reprol()}]'
+        # `Aid.reprol()`'s internal format (currently
+        # `<name>@<pid>`, but could evolve).
+        expected: str = f'{_def_prefix}[{actor.aid.reprol()}]'
         assert title == expected
         # sanity: the actor's name must be in the title
         # somewhere (so a future `reprol()` change that
@@ -140,15 +146,17 @@ def test_subactor_proctitle_visible_via_proc():
     )
 
     pid, info = matched[0]
-    # canonical proctitle prefix in cmdline (full form)
+    # canonical proctitle prefix in cmdline (full form);
-    assert info['cmdline'].startswith('tractor[proctitle_boi@'), (
+    # prefix sourced from `_def_prefix` so it tracks the
-        f'cmdline missing `tractor[proctitle_boi@…]` prefix: '
+    # `3a45dbd5` flip (`tractor[` -> `_subactor[`).
+    assert info['cmdline'].startswith(f'{_def_prefix}[proctitle_boi@'), (
+        f'cmdline missing `{_def_prefix}[proctitle_boi@…]` prefix: '
         f'{info["cmdline"]!r}'
     )
     # comm is kernel-truncated to ~15 bytes — just check the
-    # `tractor[` prefix made it.
+    # `<_def_prefix>[` prefix made it.
-    assert info['comm'].startswith('tractor['), (
+    assert info['comm'].startswith(f'{_def_prefix}['), (
-        f'comm missing `tractor[` prefix: {info["comm"]!r}'
+        f'comm missing `{_def_prefix}[` prefix: {info["comm"]!r}'
     )
     # intrinsic-signal detector should match.
     assert info['is_tractor'] is True

tests/spawn/test_main_thread_forkserver.py

@@ -0,0 +1,652 @@
+'''
+Integration exercises for the `tractor.spawn._main_thread_forkserver`
+submodule at three tiers:
+
+1. the low-level primitives
+   (`fork_from_worker_thread()` from `_main_thread_forkserver`
+   + `run_subint_in_worker_thread()` from
+   `_subint_forkserver`) driven from inside a real
+   `trio.run()` in the parent process,
+
+2. the full `main_thread_forkserver_proc` spawn backend wired
+   through tractor's normal actor-nursery + portal-RPC
+   machinery — i.e. `open_root_actor` + `open_nursery` +
+   `run_in_actor` against a subactor spawned via fork from a
+   main-interp worker thread.
+
+Background
+----------
+`ai/conc-anal/subint_fork_blocked_by_cpython_post_fork_issue.md`
+establishes that `os.fork()` from a non-main sub-interpreter
+aborts the child at the CPython level. The sibling
+`subint_fork_from_main_thread_smoketest.py` proves the escape
+hatch: fork from a main-interp *worker thread* (one that has
+never entered a subint) works, and the forked child can then
+host its own `trio.run()` inside a fresh subint.
+
+Those smoke-test scenarios are standalone — no trio runtime
+in the *parent*. Tiers (1)+(2) here cover the primitives
+driven from inside `trio.run()` in the parent, and tier (3)
+(the `*_spawn_basic` test) drives the registered
+`main_thread_forkserver` spawn backend end-to-end against
+the tractor runtime.
+
+Gating
+------
+- py3.14+ (via `concurrent.interpreters` presence)
+- no `--spawn-backend` restriction — the backend-level test
+  flips `tractor.spawn._spawn._spawn_method` programmatically
+  (via `try_set_start_method('main_thread_forkserver')`) and
+  restores it on teardown, so these tests are independent of
+  the session-level CLI backend choice.
+
+'''
+from __future__ import annotations
+from functools import partial
+import os
+from pathlib import Path
+import platform
+import select
+import signal
+import subprocess
+import sys
+import time
+
+import pytest
+import trio
+
+import tractor
+from tractor.devx import dump_on_hang
+
+
+# Gate: subint forkserver primitives require py3.14+. Check
+# the public stdlib wrapper's presence (added in 3.14) rather
+# than `_interpreters` directly — see
+# `tractor.spawn._subint` for why.
+pytest.importorskip('concurrent.interpreters')
+
+from tractor.spawn._main_thread_forkserver import (  # noqa: E402
+    fork_from_worker_thread,
+    wait_child,
+)
+from tractor.spawn._subint_forkserver import (  # noqa: E402
+    run_subint_in_worker_thread,
+)
+from tractor.spawn import _spawn as _spawn_mod  # noqa: E402
+from tractor.spawn._spawn import try_set_start_method  # noqa: E402
+
+
+# ----------------------------------------------------------------
+# child-side callables (passed via `child_target=` across fork)
+# ----------------------------------------------------------------
+
+
+_CHILD_TRIO_BOOTSTRAP: str = (
+    'import trio\n'
+    'async def _main():\n'
+    '    await trio.sleep(0.05)\n'
+    '    return 42\n'
+    'result = trio.run(_main)\n'
+    'assert result == 42, f"trio.run returned {result}"\n'
+)
+
+
+def _child_trio_in_subint() -> int:
+    '''
+    `child_target` for the trio-in-child scenario: drive a
+    trivial `trio.run()` inside a fresh legacy-config subint
+    on a worker thread.
+
+    Returns an exit code suitable for `os._exit()`:
+    - 0: subint-hosted `trio.run()` succeeded
+    - 3: driver thread hang (timeout inside `run_subint_in_worker_thread`)
+    - 4: subint bootstrap raised some other exception
+
+    '''
+    try:
+        run_subint_in_worker_thread(
+            _CHILD_TRIO_BOOTSTRAP,
+            thread_name='child-subint-trio-thread',
+        )
+    except RuntimeError:
+        # timeout / thread-never-returned
+        return 3
+    except BaseException:
+        return 4
+    return 0
+
+
+# ----------------------------------------------------------------
+# parent-side harnesses (run inside `trio.run()`)
+# ----------------------------------------------------------------
+
+
+async def run_fork_in_non_trio_thread(
+    deadline: float,
+    *,
+    child_target=None,
+) -> int:
+    '''
+    From inside a parent `trio.run()`, off-load the
+    forkserver primitive to a main-interp worker thread via
+    `trio.to_thread.run_sync()` and return the forked child's
+    pid.
+
+    Then `wait_child()` on that pid (also off-loaded so we
+    don't block trio's event loop on `waitpid()`) and assert
+    the child exited cleanly.
+
+    '''
+    with trio.fail_after(deadline):
+        # NOTE: `fork_from_worker_thread` internally spawns its
+        # own dedicated `threading.Thread` (not from trio's
+        # cache) and joins it before returning — so we can
+        # safely off-load via `to_thread.run_sync` without
+        # worrying about the trio-thread-cache recycling the
+        # runner. Pass `abandon_on_cancel=False` for the
+        # same "bounded + clean" rationale we use in
+        # `_subint.subint_proc`.
+        pid: int = await trio.to_thread.run_sync(
+            partial(
+                fork_from_worker_thread,
+                child_target,
+                thread_name='test-subint-forkserver',
+            ),
+            abandon_on_cancel=False,
+        )
+        assert pid > 0
+
+        ok, status_str = await trio.to_thread.run_sync(
+            partial(
+                wait_child,
+                pid,
+                expect_exit_ok=True,
+            ),
+            abandon_on_cancel=False,
+        )
+        assert ok, (
+            f'forked child did not exit cleanly: '
+            f'{status_str}'
+        )
+        return pid
+
+
+# ----------------------------------------------------------------
+# tests
+# ----------------------------------------------------------------
+
+
+# Bounded wall-clock via `pytest-timeout` (`method='thread'`)
+# for the usual GIL-hostage safety reason documented in the
+# sibling `test_subint_cancellation.py` / the class-A
+# `subint_sigint_starvation_issue.md`. Each test also has an
+# inner `trio.fail_after()` so assertion failures fire fast
+# under normal conditions.
+@pytest.mark.timeout(30, method='thread')
+def test_fork_from_worker_thread_via_trio(
+) -> None:
+    '''
+    Baseline: inside `trio.run()`, call
+    `fork_from_worker_thread()` via `trio.to_thread.run_sync()`,
+    get a child pid back, reap the child cleanly.
+
+    No trio-in-child. If this regresses we know the parent-
+    side trio↔worker-thread plumbing is broken independent
+    of any child-side subint machinery.
+
+    '''
+    deadline: float = 10.0
+    with dump_on_hang(
+        seconds=deadline,
+        path='/tmp/main_thread_forkserver_baseline.dump',
+    ):
+        pid: int = trio.run(
+            partial(run_fork_in_non_trio_thread, deadline),
+        )
+    # parent-side sanity — we got a real pid back.
+    assert isinstance(pid, int) and pid > 0
+    # by now the child has been waited on; it shouldn't be
+    # reap-able again.
+    with pytest.raises((ChildProcessError, OSError)):
+        os.waitpid(pid, os.WNOHANG)
+
+
+@pytest.mark.timeout(30, method='thread')
+def test_fork_and_run_trio_in_child() -> None:
+    '''
+    End-to-end: inside the parent's `trio.run()`, off-load
+    `fork_from_worker_thread()` to a worker thread, have the
+    forked child then create a fresh subint and run
+    `trio.run()` inside it on yet another worker thread.
+
+    This is the full "forkserver + trio-in-subint-in-child"
+    pattern the proposed `main_thread_forkserver` spawn backend
+    would rest on.
+
+    '''
+    deadline: float = 15.0
+    with dump_on_hang(
+        seconds=deadline,
+        path='/tmp/main_thread_forkserver_trio_in_child.dump',
+    ):
+        pid: int = trio.run(
+            partial(
+                run_fork_in_non_trio_thread,
+                deadline,
+                child_target=_child_trio_in_subint,
+            ),
+        )
+    assert isinstance(pid, int) and pid > 0
+
+
+# ----------------------------------------------------------------
+# tier-3 backend test: drive the registered `main_thread_forkserver`
+# spawn backend end-to-end through tractor's actor-nursery +
+# portal-RPC machinery.
+# ----------------------------------------------------------------
+
+
+async def _trivial_rpc() -> str:
+    '''
+    Minimal subactor-side RPC body: just return a sentinel
+    string the parent can assert on.
+
+    '''
+    return 'hello from subint-forkserver child'
+
+
+async def _happy_path_forkserver(
+    reg_addr: tuple[str, int | str],
+    deadline: float,
+) -> None:
+    '''
+    Parent-side harness: stand up a root actor, open an actor
+    nursery, spawn one subactor via the currently-selected
+    spawn backend (which this test will have flipped to
+    `main_thread_forkserver`), run a trivial RPC through its
+    portal, assert the round-trip result.
+
+    '''
+    with trio.fail_after(deadline):
+        async with (
+            tractor.open_root_actor(
+                registry_addrs=[reg_addr],
+            ),
+            tractor.open_nursery() as an,
+        ):
+            portal: tractor.Portal = await an.run_in_actor(
+                _trivial_rpc,
+                name='subint-forkserver-child',
+            )
+            result: str = await portal.wait_for_result()
+            assert result == 'hello from subint-forkserver child'
+
+
+@pytest.fixture
+def forkserver_spawn_method():
+    '''
+    Flip `tractor.spawn._spawn._spawn_method` to
+    `'main_thread_forkserver'` for the duration of a test,
+    then restore whatever was in place before (usually the
+    session-level CLI choice, typically `'trio'`).
+
+    Without this, other tests in the same session would
+    observe the global flip and start spawning via fork —
+    which is almost certainly NOT what their assertions were
+    written against.
+
+    '''
+    prev_method: str = _spawn_mod._spawn_method
+    prev_ctx = _spawn_mod._ctx
+    try_set_start_method('main_thread_forkserver')
+    try:
+        yield
+    finally:
+        _spawn_mod._spawn_method = prev_method
+        _spawn_mod._ctx = prev_ctx
+
+
+@pytest.mark.timeout(60, method='thread')
+def test_main_thread_forkserver_spawn_basic(
+    reg_addr: tuple[str, int | str],
+    forkserver_spawn_method,
+) -> None:
+    '''
+    Happy-path: spawn ONE subactor via the
+    `main_thread_forkserver` backend (parent-side fork from a
+    main-interp worker thread), do a trivial portal-RPC
+    round-trip, tear the nursery down cleanly.
+
+    If this passes, the "forkserver + tractor runtime" arch
+    is proven end-to-end: the registered
+    `main_thread_forkserver_proc` spawn target successfully
+    forks a child, the child runs `_actor_child_main()` +
+    completes IPC handshake + serves an RPC, and the parent
+    reaps via `_ForkedProc.wait()` without regressing any of
+    the normal nursery teardown invariants.
+
+    '''
+    deadline: float = 20.0
+    with dump_on_hang(
+        seconds=deadline,
+        path='/tmp/main_thread_forkserver_spawn_basic.dump',
+    ):
+        trio.run(
+            partial(
+                _happy_path_forkserver,
+                reg_addr,
+                deadline,
+            ),
+        )
+
+
+# ----------------------------------------------------------------
+# tier-4 DRAFT: orphaned-subactor SIGINT survivability
+#
+# Motivating question: with `main_thread_forkserver`, the child's
+# `trio.run()` lives on the fork-inherited worker thread which
+# is NOT `threading.main_thread()` — so trio cannot install its
+# `signal.set_wakeup_fd`-based SIGINT handler. If the parent
+# goes away via `SIGKILL` (no IPC `Portal.cancel_actor()`
+# possible), does SIGINT on the orphan child cleanly tear it
+# down via CPython's default `KeyboardInterrupt` delivery, or
+# does it hang?
+#
+# Working hypothesis (unverified pre-this-test): post-fork the
+# child is effectively single-threaded (only the fork-worker
+# tstate survived), so SIGINT → default handler → raises
+# `KeyboardInterrupt` on the only thread — which happens to be
+# the one driving trio's event loop — so trio observes it at
+# the next checkpoint. If so, we're "fine" on this backend
+# despite the missing trio SIGINT handler.
+#
+# Cross-backend generalization (decide after this passes):
+# - applicable to any backend whose subactors are separate OS
+#   processes: `trio`, `mp_spawn`, `mp_forkserver`,
+#   `main_thread_forkserver`.
+# - NOT applicable to plain `subint` (subactors are in-process
+#   subinterpreters, no orphan child process to SIGINT).
+# - move path: lift the harness script into
+#   `tests/_orphan_harness.py`, parametrize on the session's
+#   `_spawn_method`, add `skipif _spawn_method == 'subint'`.
+# ----------------------------------------------------------------
+
+
+_ORPHAN_HARNESS_SCRIPT: str = '''
+import os
+import sys
+import trio
+import tractor
+from tractor.spawn._spawn import try_set_start_method
+
+async def _sleep_forever() -> None:
+    print(f"CHILD_PID={os.getpid()}", flush=True)
+    await trio.sleep_forever()
+
+async def _main(reg_addr):
+    async with (
+        tractor.open_root_actor(registry_addrs=[reg_addr]),
+        tractor.open_nursery() as an,
+    ):
+        portal = await an.run_in_actor(
+            _sleep_forever,
+            name="orphan-test-child",
+        )
+        print(f"PARENT_READY={os.getpid()}", flush=True)
+        await trio.sleep_forever()
+
+if __name__ == "__main__":
+    backend = sys.argv[1]
+    host = sys.argv[2]
+    port = int(sys.argv[3])
+    try_set_start_method(backend)
+    trio.run(_main, (host, port))
+'''
+
+
+def _read_marker(
+    proc: subprocess.Popen,
+    marker: str,
+    timeout: float,
+    _buf: dict,
+) -> str:
+    '''
+    Block until `<marker>=<value>\\n` appears on `proc.stdout`
+    and return `<value>`. Uses a per-proc byte buffer (`_buf`)
+    to carry partial lines across calls.
+
+    '''
+    deadline: float = time.monotonic() + timeout
+    remainder: bytes = _buf.get('remainder', b'')
+    prefix: bytes = f'{marker}='.encode()
+    while time.monotonic() < deadline:
+        # drain any complete lines already buffered
+        while b'\n' in remainder:
+            line, remainder = remainder.split(b'\n', 1)
+            if line.startswith(prefix):
+                _buf['remainder'] = remainder
+                return line[len(prefix):].decode().strip()
+        ready, _, _ = select.select([proc.stdout], [], [], 0.2)
+        if not ready:
+            continue
+        chunk: bytes = os.read(proc.stdout.fileno(), 4096)
+        if not chunk:
+            break
+        remainder += chunk
+    _buf['remainder'] = remainder
+    raise TimeoutError(
+        f'Never observed marker {marker!r} on harness stdout '
+        f'within {timeout}s'
+    )
+
+
+def _process_alive(pid: int) -> bool:
+    '''Liveness probe for a pid we do NOT parent (post-orphan).'''
+    try:
+        os.kill(pid, 0)
+        return True
+    except ProcessLookupError:
+        return False
+
+
+# Known-gap test — `main_thread_forkserver` orphan-SIGINT
+# handling. See
+# `ai/conc-anal/subint_forkserver_orphan_sigint_hang_issue.md`.
+# `strict=True` so if a future fix closes the gap the
+# XPASS surfaces as a FAIL and forces us to drop the
+# mark intentionally.
+@pytest.mark.xfail(
+    strict=True,
+    reason=(
+        'Orphan subactor SIGINT delivery: trio event loop '
+        'on non-main thread post-fork doesn\'t see the '
+        'external SIGINT → KBI path. See tracker doc.\n'
+        'ai/conc-anal/subint_forkserver_orphan_sigint_hang_issue.md'
+    ),
+)
+@pytest.mark.timeout(
+    30,
+    method='thread',
+)
+def test_orphaned_subactor_sigint_cleanup_DRAFT(
+    reg_addr: tuple[str, int | str],
+    tmp_path: Path,
+) -> None:
+    '''
+    DRAFT — orphaned-subactor SIGINT survivability under the
+    `main_thread_forkserver` backend.
+
+    Sequence:
+      1. Spawn a harness subprocess that brings up a root
+         actor + one `sleep_forever` subactor via
+         `main_thread_forkserver`.
+      2. Read the harness's stdout for `PARENT_READY=<pid>`
+         and `CHILD_PID=<pid>` markers (confirms the
+         parent→child IPC handshake completed).
+      3. `SIGKILL` the parent (no IPC cancel possible — the
+         whole point of this test).
+      4. `SIGINT` the orphan child.
+      5. Poll `os.kill(child_pid, 0)` for up to 10s — assert
+         the child exits.
+
+    Empirical result (2026-04, py3.14): currently **FAILS** —
+    SIGINT on the orphan child doesn't unwind the trio loop,
+    despite trio's `KIManager` handler being correctly
+    installed in the subactor (the post-fork thread IS
+    `threading.main_thread()` on py3.14). `faulthandler` dump
+    shows the subactor wedged in `trio/_core/_io_epoll.py::
+    get_events` — the signal's supposed wakeup of the event
+    loop isn't firing. Full analysis + diagnostic evidence
+    in `ai/conc-anal/
+    subint_forkserver_orphan_sigint_hang_issue.md`.
+
+    The runtime's *intentional* "KBI-as-OS-cancel" path at
+    `tractor/spawn/_entry.py::_trio_main:164` is therefore
+    unreachable under this backend+config. Closing the gap is
+    aligned with existing design intent (make the already-
+    designed behavior actually fire), not a new feature.
+    Marked `xfail(strict=True)` so the
+    mark flips to XPASS→fail once the gap is closed and we'll
+    know to drop the mark.
+
+    '''
+    if platform.system() != 'Linux':
+        pytest.skip(
+            'orphan-reparenting semantics only exercised on Linux'
+        )
+
+    script_path = tmp_path / '_orphan_harness.py'
+    script_path.write_text(_ORPHAN_HARNESS_SCRIPT)
+
+    # Offset the port so we don't race the session reg_addr with
+    # any concurrently-running backend test's listener.
+    host: str = reg_addr[0]
+    port: int = int(reg_addr[1]) + 17
+
+    proc: subprocess.Popen = subprocess.Popen(
+        [
+            sys.executable,
+            str(script_path),
+            'main_thread_forkserver',
+            host,
+            str(port),
+        ],
+        stdout=subprocess.PIPE,
+        stderr=subprocess.STDOUT,
+    )
+    parent_pid: int | None = None
+    child_pid: int | None = None
+    buf: dict = {}
+    try:
+        child_pid = int(_read_marker(proc, 'CHILD_PID', 15.0, buf))
+        parent_pid = int(_read_marker(proc, 'PARENT_READY', 15.0, buf))
+
+        # sanity: both alive before we start killing stuff
+        assert _process_alive(parent_pid), (
+            f'harness parent pid={parent_pid} gone before '
+            f'SIGKILL — test premise broken'
+        )
+        assert _process_alive(child_pid), (
+            f'orphan-candidate child pid={child_pid} gone '
+            f'before test started'
+        )
+
+        # step 3: kill parent — no IPC cancel arrives at child.
+        # `proc.wait()` reaps the zombie so it truly disappears
+        # from the process table (otherwise `os.kill(pid, 0)`
+        # keeps reporting it as alive).
+        os.kill(parent_pid, signal.SIGKILL)
+        try:
+            proc.wait(timeout=3.0)
+        except subprocess.TimeoutExpired:
+            pytest.fail(
+                f'harness parent pid={parent_pid} did not die '
+                f'after SIGKILL — test premise broken'
+            )
+        assert _process_alive(child_pid), (
+            f'child pid={child_pid} died along with parent — '
+            f'did the parent reap it before SIGKILL took? '
+            f'test premise requires an orphan.'
+        )
+
+        # step 4+5: SIGINT the orphan, poll for exit.
+        os.kill(child_pid, signal.SIGINT)
+        timeout: float = 6.0
+        cleanup_deadline: float = time.monotonic() + timeout
+        while time.monotonic() < cleanup_deadline:
+            if not _process_alive(child_pid):
+                return  # <- success path
+            time.sleep(0.1)
+
+        pytest.fail(
+            f'Orphan subactor (pid={child_pid}) did NOT exit '
+            f'within 10s of SIGINT under `main_thread_forkserver` '
+            f'→ trio on non-main thread did not observe the '
+            f'default CPython KeyboardInterrupt; backend needs '
+            f'explicit SIGINT plumbing.'
+        )
+    finally:
+        # best-effort cleanup to avoid leaking orphans across
+        # the test session regardless of outcome.
+        for pid in (parent_pid, child_pid):
+            if pid is None:
+                continue
+            try:
+                os.kill(pid, signal.SIGKILL)
+            except ProcessLookupError:
+                pass
+        try:
+            proc.kill()
+        except OSError:
+            pass
+        try:
+            proc.wait(timeout=2.0)
+        except subprocess.TimeoutExpired:
+            pass
+
+
+# ----------------------------------------------------------------
+# regression guard: variant-2 (`subint_forkserver`) placeholder
+# MUST raise `NotImplementedError` today — guards against future
+# commits accidentally re-aliasing the key to the variant-1
+# coroutine (which was a transient state during the rename).
+# ----------------------------------------------------------------
+def test_subint_forkserver_key_errors_cleanly() -> None:
+    '''
+    `--spawn-backend=subint_forkserver` is reserved for the
+    eventual variant-2 (subint-isolated child runtime)
+    backend, gated on jcrist/msgspec#1026 unblocking PEP 684
+    isolated-mode subints upstream.
+
+    Until that lands, the dispatch entry MUST raise
+    `NotImplementedError` immediately rather than silently
+    aliasing to `main_thread_forkserver_proc`. Verify the
+    error message also surfaces both the working-backend
+    pointer and the upstream-blocker ref so an operator
+    arriving at the error has somewhere to go.
+
+    '''
+    import asyncio
+    from tractor.spawn._spawn import _methods
+
+    proc = _methods['subint_forkserver']
+    with pytest.raises(NotImplementedError) as ei:
+        # signature args match `main_thread_forkserver_proc`'s
+        # — the stub raises before touching them so dummy
+        # values are fine.
+        asyncio.run(
+            proc(
+                'x', None, None, {}, [],
+                ('127.0.0.1', 0), {},
+            )
+        )
+
+    msg: str = str(ei.value)
+    assert 'main_thread_forkserver' in msg, (
+        f'stub error msg should redirect to the working '
+        f'variant-1 backend; got: {msg!r}'
+    )
+    assert 'msgspec#1026' in msg or '1026' in msg, (
+        f'stub error msg should reference the upstream '
+        f'blocker (jcrist/msgspec#1026); got: {msg!r}'
+    )

tests/spawn/test_subint_cancellation.py

@@ -0,0 +1,237 @@
+'''
+Cancellation + hard-kill semantics audit for the `subint` spawn
+backend.
+
+Exercises the escape-hatch machinery added to
+`tractor.spawn._subint` (module-level `_HARD_KILL_TIMEOUT`,
+bounded shields around the soft-kill / thread-join sites, daemon
+driver-thread abandonment) so that future stdlib regressions or
+our own refactors don't silently re-introduce the hangs first
+diagnosed during the Phase B.2/B.3 bringup (issue #379).
+
+Every test in this module:
+- is wrapped in `trio.fail_after()` for a deterministic per-test
+  wall-clock ceiling (the whole point of these tests is to fail
+  fast when our escape hatches regress; an unbounded test would
+  defeat itself),
+- arms `tractor.devx.dump_on_hang()` to capture a stack dump on
+  failure — without it, a hang here is opaque because pytest's
+  stderr capture swallows `faulthandler` output by default
+  (hard-won lesson from the original diagnosis),
+- skips on py<3.13 (no `_interpreters`) and on any
+  `--spawn-backend` other than `'subint'` (these tests are
+  subint-specific by design — they'd be nonsense under `trio` or
+  `mp_*`).
+
+'''
+from __future__ import annotations
+from functools import partial
+
+import pytest
+import trio
+import tractor
+from tractor.devx import dump_on_hang
+
+
+# Gate: the `subint` backend requires py3.14+. Check the
+# public stdlib wrapper's presence (added in 3.14) rather than
+# the private `_interpreters` module (which exists on 3.13 but
+# wedges under tractor's usage — see `tractor.spawn._subint`).
+pytest.importorskip('concurrent.interpreters')
+
+# Subint-only: read the spawn method that `pytest_configure`
+# committed via `try_set_start_method()`. By the time this module
+# imports, the CLI backend choice has been applied.
+from tractor.spawn._spawn import _spawn_method  # noqa: E402
+
+if _spawn_method != 'subint':
+    pytestmark = pytest.mark.skip(
+        reason=(
+            "subint-specific cancellation audit — "
+            "pass `--spawn-backend=subint` to run."
+        ),
+    )
+
+
+# ----------------------------------------------------------------
+# child-side task bodies (run inside the spawned subint)
+# ----------------------------------------------------------------
+
+
+async def _trivial_rpc() -> str:
+    '''
+    Minimal RPC body for the baseline happy-teardown test.
+    '''
+    return 'hello from subint'
+
+
+async def _spin_without_trio_checkpoints() -> None:
+    '''
+    Block the main task with NO trio-visible checkpoints so any
+    `Portal.cancel_actor()` arriving over IPC has nothing to hand
+    off to.
+
+    `threading.Event.wait(timeout)` releases the GIL (so other
+    threads — including trio's IO/RPC tasks — can progress) but
+    does NOT insert a trio checkpoint, so the subactor's main
+    task never notices cancellation.
+
+    This is the exact "stuck subint" scenario the hard-kill
+    shields exist to survive.
+    '''
+    import threading
+    never_set = threading.Event()
+    while not never_set.is_set():
+        # 1s re-check granularity; low enough not to waste CPU,
+        # high enough that even a pathologically slow
+        # `_HARD_KILL_TIMEOUT` won't accidentally align with a
+        # wake.
+        never_set.wait(timeout=1.0)
+
+
+# ----------------------------------------------------------------
+# parent-side harnesses (driven inside `trio.run(...)`)
+# ----------------------------------------------------------------
+
+
+async def _happy_path(
+    reg_addr: tuple[str, int|str],
+    deadline: float,
+) -> None:
+    with trio.fail_after(deadline):
+        async with (
+            tractor.open_root_actor(
+                registry_addrs=[reg_addr],
+            ),
+            tractor.open_nursery() as an,
+        ):
+            portal: tractor.Portal = await an.run_in_actor(
+                _trivial_rpc,
+                name='subint-happy',
+            )
+            result: str = await portal.wait_for_result()
+            assert result == 'hello from subint'
+
+
+async def _spawn_stuck_then_cancel(
+    reg_addr: tuple[str, int|str],
+    deadline: float,
+) -> None:
+    with trio.fail_after(deadline):
+        async with (
+            tractor.open_root_actor(
+                registry_addrs=[reg_addr],
+            ),
+            tractor.open_nursery() as an,
+        ):
+            await an.run_in_actor(
+                _spin_without_trio_checkpoints,
+                name='subint-stuck',
+            )
+            # Give the child time to reach its non-checkpointing
+            # loop before we cancel; the precise value doesn't
+            # matter as long as it's a handful of trio schedule
+            # ticks.
+            await trio.sleep(0.5)
+            an.cancel_scope.cancel()
+
+
+# ----------------------------------------------------------------
+# tests
+# ----------------------------------------------------------------
+
+
+def test_subint_happy_teardown(
+    reg_addr: tuple[str, int|str],
+) -> None:
+    '''
+    Baseline: spawn a subactor, do one portal RPC, close nursery
+    cleanly. No cancel, no faults.
+
+    If this regresses we know something's wrong at the
+    spawn/teardown layer unrelated to the hard-kill escape
+    hatches.
+
+    '''
+    deadline: float = 10.0
+    with dump_on_hang(
+        seconds=deadline,
+        path='/tmp/subint_cancellation_happy.dump',
+    ):
+        trio.run(partial(_happy_path, reg_addr, deadline))
+
+
+# Wall-clock bound via `pytest-timeout` (`method='thread'`)
+# as defense-in-depth over the inner `trio.fail_after(15)`.
+# Under the orphaned-channel hang class described in
+# `ai/conc-anal/subint_cancel_delivery_hang_issue.md`, SIGINT
+# is still deliverable and this test *should* be unwedgeable
+# by the inner trio timeout — but sibling subint-backend
+# tests in this repo have also exhibited the
+# `subint_sigint_starvation_issue.md` GIL-starvation flavor,
+# so `method='thread'` keeps us safe in case ordering or
+# load shifts the failure mode.
+@pytest.mark.timeout(
+    3,  # NOTE never passes pre-3.14+ subints support.
+    method='thread',
+)
+def test_subint_non_checkpointing_child(
+    reg_addr: tuple[str, int|str],
+) -> None:
+    '''
+    Cancel a subactor whose main task is stuck in a non-
+    checkpointing Python loop.
+
+    `Portal.cancel_actor()` may be delivered over IPC but the
+    main task never checkpoints to observe the Cancelled —
+    so the subint's `trio.run()` can't exit gracefully.
+
+    The parent `subint_proc` bounded-shield + daemon-driver-
+    thread combo should abandon the thread after
+    `_HARD_KILL_TIMEOUT` and let the parent return cleanly.
+
+    Wall-clock budget:
+    - ~0.5s: settle time for child to enter the stuck loop
+    - ~3s: `_HARD_KILL_TIMEOUT` (soft-kill wait)
+    - ~3s: `_HARD_KILL_TIMEOUT` (thread-join wait)
+    - margin
+
+    KNOWN ISSUE (Ctrl-C-able hang):
+    -------------------------------
+    This test currently hangs past the hard-kill timeout for
+    reasons unrelated to the subint teardown itself — after
+    the subint is destroyed, a parent-side trio task appears
+    to park on an orphaned IPC channel (no clean EOF
+    delivered to a waiting receive). Unlike the
+    SIGINT-starvation sibling case in
+    `test_stale_entry_is_deleted`, this hang IS Ctrl-C-able
+    (`strace` shows SIGINT wakeup-fd `write() = 1`, not
+    `EAGAIN`) — i.e. the main trio loop is still iterating
+    normally. That makes this *our* bug to fix, not a
+    CPython-level limitation.
+
+    See `ai/conc-anal/subint_cancel_delivery_hang_issue.md`
+    for the full analysis + candidate fix directions
+    (explicit parent-side channel abort in `subint_proc`
+    teardown being the most likely surgical fix).
+
+    The sibling `ai/conc-anal/subint_sigint_starvation_issue.md`
+    documents the *other* hang class (abandoned-legacy-subint
+    thread + shared-GIL starvation → signal-wakeup-fd pipe
+    fills → SIGINT silently dropped) — that one is
+    structurally blocked on msgspec PEP 684 adoption and is
+    NOT what this test is hitting.
+
+    '''
+    deadline: float = 15.0
+    with dump_on_hang(
+        seconds=deadline,
+        path='/tmp/subint_cancellation_stuck.dump',
+    ):
+        trio.run(
+            partial(
+                _spawn_stuck_then_cancel,
+                reg_addr,
+                deadline,
+            ),
+        )

tests/test_context_stream_semantics.py

@@ -263,9 +263,18 @@ def test_simple_context(
         except error_parent:
             pass
         except BaseExceptionGroup as beg:
-            # XXX: on windows it seems we may have to expect the group error
+            # XXX: on windows — and under `trio>=0.33`'s strict
+            # exception-groups, where a lone `error_parent` now
+            # arrives wrapped in a 1-exc `ExceptionGroup` rather than
+            # collapsed to the bare exc — accept either the
+            # `error_parent` nested in the group OR a cancel-only
+            # (multi-cancelled) group.
             from tractor.trionics import is_multi_cancelled
-            assert is_multi_cancelled(beg)
+            assert (
+                is_multi_cancelled(beg)
+                or
+                beg.subgroup(error_parent) is not None
+            )
     else:
         trio.run(main)
 

tests/test_log_sys.py

@@ -162,6 +162,66 @@ def test_implicit_mod_name_applied_for_child(
     assert submod.log.logger in sub_logs
 
 
+def test_io_custom_level_registered():
+    '''
+    The `IO`(21) level (registered via `add_log_level()` at
+    import, for `tractor.trionics._subproc`'s std-stream relay)
+    is fully wired and SHOWN BY DEFAULT at `info`-level consoles
+    since `21 >= INFO(20)`.
+
+    '''
+    import logging
+    assert log.CUSTOM_LEVELS.get('IO') == 21
+    assert logging.getLevelName(21) == 'IO'
+    assert log.STD_PALETTE.get('IO')
+    assert log.BOLD_PALETTE['bold'].get('IO')
+
+    iolog = log.get_logger('io_lvl_test')
+    assert callable(getattr(iolog, 'io', None))
+    # emit must not raise
+    iolog.io('hello from the IO level')
+
+    # 21 >= INFO(20) -> shown when console set to `info`
+    assert 21 >= logging.INFO
+
+
+def test_add_log_level_pluggable():
+    '''
+    `add_log_level()` is the single pluggable entry-point: one
+    call wires `CUSTOM_LEVELS` + `addLevelName` + both palettes +
+    a same-named `StackLevelAdapter` emit method (so
+    `get_logger()`'s per-level audit passes).
+
+    '''
+    import logging
+    name: str = 'XLVL'
+    val: int = 19
+    try:
+        log.add_log_level(name, val, 'cyan')
+
+        assert log.CUSTOM_LEVELS[name] == val
+        assert logging.getLevelName(val) == name
+        assert log.STD_PALETTE[name] == 'cyan'
+        assert log.BOLD_PALETTE['bold'][name] == 'bold_cyan'
+
+        # the audit in `get_logger()` (asserts a method per
+        # `CUSTOM_LEVELS` entry) must still pass.
+        xlog = log.get_logger('xlvl_test')
+        emit = getattr(xlog, name.lower(), None)
+        assert callable(emit)
+        emit('hello from a plugged-in level')
+
+    finally:
+        # best-effort cleanup of our module-global mutations so
+        # later `get_logger()` audits don't see a half-removed
+        # level.
+        log.CUSTOM_LEVELS.pop(name, None)
+        log.STD_PALETTE.pop(name, None)
+        log.BOLD_PALETTE['bold'].pop(name, None)
+        if hasattr(log.StackLevelAdapter, name.lower()):
+            delattr(log.StackLevelAdapter, name.lower())
+
+
 # TODO, moar tests against existing feats:
 # ------ - ------
 # - [ ] color settings?

tests/test_spawning.py

@@ -150,7 +150,12 @@ async def test_most_beautiful_word(
     The main ``tractor`` routine.
 
     '''
-    with trio.fail_after(1):
+    # actor spawn + IPC round-trip is comfortably sub-second on a
+    # warm box, but slow/noisy CI runners (esp. macOS) blow a flat
+    # 1s deadline. Scale for CI/CPU-throttle headroom — `== 1s`
+    # locally where `cpu_scaling_factor()` is `1.0`.
+    from .conftest import cpu_scaling_factor
+    with trio.fail_after(1 * cpu_scaling_factor()):
         async with tractor.open_nursery(
             debug_mode=debug_mode,
         ) as an:

tests/trionics/test_subproc.py

@@ -0,0 +1,235 @@
+'''
+Unit tests for `tractor.trionics.supervise_run_process` (in
+`tractor.trionics._subproc`) and its per-line std-stream relay.
+
+Hermetic `trio`-only coverage (no actor-runtime needed):
+
+- per-line stdout relay -> `log.io`
+- parent controlling-tty isolation (child fd1 is a pipe, fd0
+  `/dev/null` — never the parent `/dev/pts/*`)
+- mandatory concurrent pipe-drain (no deadlock on >64KiB
+  no-newline output)
+- live stderr relay + `CalledProcessError` rebuild (rc!=0 note)
+- legacy capture-stderr CPE note path
+
+'''
+from functools import partial
+import platform
+import subprocess
+
+import pytest
+import trio
+
+from tractor.trionics import (
+    _subproc,
+    collapse_eg,
+    supervise_run_process,
+)
+
+
+def _capture_relay(monkeypatch, level: str = 'io') -> list[str]:
+    '''
+    Redirect `_subproc.log.<level>` (the relay's emit method —
+    `io` by default, see `supervise_run_process(relay_level=...)`)
+    into a list so tests can assert on the relayed lines.
+
+    '''
+    records: list[str] = []
+    monkeypatch.setattr(
+        _subproc.log,
+        level,
+        lambda msg, *a, **k: records.append(msg),
+    )
+    return records
+
+
+def test_stdout_relayed_per_line(monkeypatch):
+    records = _capture_relay(monkeypatch)
+
+    cmd = [
+        'sh', '-c',
+        'for i in 1 2 3; do echo line=$i; done',
+    ]
+
+    async def main():
+        async with trio.open_nursery() as tn:
+            await tn.start(
+                partial(
+                    supervise_run_process,
+                    cmd,
+                    label='t-out',
+                    relay_stdout=True,
+                )
+            )
+
+    trio.run(main)
+
+    out_lines = [r for r in records if '[t-out:out]' in r]
+    assert any('line=1' in r for r in out_lines)
+    assert any('line=2' in r for r in out_lines)
+    assert any('line=3' in r for r in out_lines)
+
+
+@pytest.mark.skipif(
+    platform.system() != 'Linux',
+    reason='reads `/proc/self/fd` — Linux-only',
+)
+def test_parent_tty_isolated(monkeypatch):
+    records = _capture_relay(monkeypatch)
+
+    cmd = [
+        'sh', '-c',
+        'readlink /proc/self/fd/0; readlink /proc/self/fd/1',
+    ]
+
+    async def main():
+        async with trio.open_nursery() as tn:
+            await tn.start(
+                partial(
+                    supervise_run_process,
+                    cmd,
+                    label='t-tty',
+                    relay_stdout=True,
+                )
+            )
+
+    trio.run(main)
+
+    relayed = '\n'.join(records)
+    # fd1 (stdout) must be OUR pipe, never a controlling tty.
+    assert 'pipe:' in relayed
+    assert '/dev/pts/' not in relayed
+    # fd0 (stdin) is pinned to DEVNULL.
+    assert '/dev/null' in relayed
+
+
+def test_no_deadlock_on_big_unnewlined_output(monkeypatch):
+    '''
+    >64KiB of output with NO newline: only completes because the
+    relay reader concurrently drains the pipe (else the child
+    blocks on `write()` when the OS pipe buffer fills).
+
+    '''
+    records = _capture_relay(monkeypatch)
+
+    cmd = [
+        'sh', '-c',
+        'head -c 200000 /dev/zero | tr "\\0" x',
+    ]
+
+    async def main():
+        # generous vs the ~ms real runtime, but bounded so a
+        # genuine pipe-fill deadlock fails fast.
+        with trio.fail_after(2):
+            async with trio.open_nursery() as tn:
+                await tn.start(
+                    partial(
+                        supervise_run_process,
+                        cmd,
+                        label='t-big',
+                        relay_stdout=True,
+                    )
+                )
+
+    trio.run(main)
+
+    big = ''.join(
+        r.split('] ', 1)[-1]
+        for r in records
+        if '[t-big:out]' in r
+    )
+    assert len(big) == 200_000
+
+
+def test_stderr_relay_and_cpe_rebuild(monkeypatch):
+    '''
+    `relay_stderr=True` PIPEs stderr ourselves (mutually
+    exclusive with trio's `capture_stderr`), so on rc!=0 the
+    wrapper rebuilds a `CalledProcessError` from the live
+    accumulator and `.add_note()`s its `.stderr` — AND the
+    stderr is relayed per-line live.
+
+    '''
+    records = _capture_relay(monkeypatch)
+
+    cmd = [
+        'sh', '-c',
+        'echo boom 1>&2; exit 3',
+    ]
+
+    async def main():
+        # `collapse_eg()` unwraps the parent-nursery's single-exc
+        # eg so the bare CPE bubbles straight out (mirrors real
+        # caller usage).
+        async with (
+            collapse_eg(),
+            trio.open_nursery() as tn,
+        ):
+            await tn.start(
+                partial(
+                    supervise_run_process,
+                    cmd,
+                    label='t-err',
+                    relay_stderr=True,
+                    check=True,
+                )
+            )
+
+    with pytest.raises(subprocess.CalledProcessError) as ei:
+        trio.run(main)
+
+    cpe = ei.value
+    assert cpe.returncode == 3
+    # rebuilt `.stderr` (trio did NOT capture since we PIPE'd it).
+    assert b'boom' in (cpe.stderr or b'')
+    # note attached for legible teardown reporting.
+    assert any(
+        'boom' in n
+        for n in getattr(cpe, '__notes__', [])
+    )
+    # AND it was relayed live per-line.
+    assert any(
+        '[t-err:err]' in r and 'boom' in r
+        for r in records
+    )
+
+
+def test_nonrelay_cpe_note(monkeypatch):
+    '''
+    No live relay: stderr is silently drained + captured (NOT
+    emitted), and on rc!=0 the wrapper rebuilds the
+    `CalledProcessError` from that accumulator with a `.stderr`
+    note — same deterministic post-drain path as the relay case.
+
+    '''
+    cmd = [
+        'sh', '-c',
+        'echo nope 1>&2; exit 7',
+    ]
+
+    async def main():
+        async with (
+            collapse_eg(),
+            trio.open_nursery() as tn,
+        ):
+            await tn.start(
+                partial(
+                    supervise_run_process,
+                    cmd,
+                    label='t-legacy',
+                    check=True,
+                    # relay_* default False -> silent
+                    # drain+capture for the CPE note.
+                )
+            )
+
+    with pytest.raises(subprocess.CalledProcessError) as ei:
+        trio.run(main)
+
+    cpe = ei.value
+    assert cpe.returncode == 7
+    assert b'nope' in (cpe.stderr or b'')
+    assert any(
+        'nope' in n
+        for n in getattr(cpe, '__notes__', [])
+    )

tractor/_root.py

@@ -155,7 +155,6 @@ async def maybe_block_bp(
             os.environ.pop('PYTHONBREAKPOINT', None)
 
 
-
 @acm
 async def open_root_actor(
     *,
@@ -186,6 +185,7 @@ async def open_root_actor(
     # enables the multi-process debugger support
     debug_mode: bool = False,
     maybe_enable_greenback: bool = False,  # `.pause_from_sync()/breakpoint()` support
+
     # ^XXX NOTE^ the perf implications of use,
     # https://greenback.readthedocs.io/en/latest/principle.html#performance
     enable_stack_on_sig: bool = False,

tractor/_testing/_reap.py

@@ -90,7 +90,6 @@ keys are caller-defined).
 
 '''
 from __future__ import annotations
-
 import os
 import pathlib
 import re
@@ -99,6 +98,9 @@ import stat
 import sys
 import time
 
+
+from tractor.devx import _proctitle
+
 # `/dev/shm` is the POSIX-shm filesystem on Linux + FreeBSD.
 # macOS uses `shm_open` syscalls without a fs-visible path,
 # so the shm helpers refuse to run there.
@@ -230,9 +232,9 @@ def _read_comm(pid: int) -> str:
 #   while `cmdline` for zombies often reads as empty.
 _TRACTOR_PROC_CMDLINE_MARKERS: tuple[str, ...] = (
     'tractor._child',
-    'tractor[',
+    _proctitle._def_prefix,
 )
-_TRACTOR_PROC_COMM_MARKER: str = 'tractor['
+_TRACTOR_PROC_COMM_MARKER: str = _proctitle._def_prefix
 
 
 def _is_tractor_subactor(pid: int) -> bool:

tractor/devx/_proctitle.py

@@ -24,7 +24,10 @@ which" at a glance without needing to read full
 `/proc/<pid>/cmdline`.
 
 Format:
-  ``tractor[<aid.reprol()>]``    e.g. ``tractor[doggy@1027301b]``
+  ``<_def_prefix>[<aid.reprol()>]``  e.g. ``_subactor[doggy@1027301b]``
+(prefix from the `_def_prefix` const, flipped `tractor` ->
+`_subactor` so sub-actor procs are visually distinct from the
+root in `ps`/`htop` and the reap-recognition markers.)
 
 Uses the canonical `Aid.reprol()` form
 (``<name>@<uuid_short>``) so the proc-title matches the
@@ -52,7 +55,13 @@ except ImportError:
     _stp = None
 
 
-def set_actor_proctitle(actor: 'Actor') -> str | None:
+_def_prefix: str = '_subactor'
+
+
+def set_actor_proctitle(
+    actor: 'Actor',
+    prefix: str = _def_prefix,
+) -> str | None:
     '''
     Set the calling process's proc-title to identify it as a
     tractor sub-actor.
@@ -69,6 +78,6 @@ def set_actor_proctitle(actor: 'Actor') -> str | None:
     if _stp is None:
         return None
 
-    title: str = f'tractor[{actor.aid.reprol()}]'
+    title: str = f'{prefix}[{actor.aid.reprol()}]'
     _stp.setproctitle(title)
     return title

tractor/ipc/_server.py

@@ -398,7 +398,7 @@ async def handle_stream_from_peer(
         uid,
         None,
     )
-    if event:
+    if event is not None:
         con_status_steps += (
             ' -> Waking subactor spawn waiters: '
             f'{event.statistics().tasks_waiting}\n'

tractor/log.py

@@ -262,6 +262,63 @@ class StackLevelAdapter(LoggerAdapter):
         )
 
 
+def add_log_level(
+    name: str,
+    value: int,
+    color: str = 'white',
+) -> None:
+    '''
+    Register a new custom log level with `tractor`'s logging
+    machinery in ONE call — the single pluggable entry-point that
+    keeps the (otherwise hand-synced) pieces consistent:
+
+    - `CUSTOM_LEVELS[name]` (drives the `stacklevel` bump in
+      `StackLevelAdapter.log()` + the `get_logger()` audit).
+    - `logging.addLevelName()` registration.
+    - `STD_PALETTE`/`BOLD_PALETTE` color entries (consumed when
+      `get_console_log()` builds its `ColoredFormatter`).
+    - a same-named (lowercase) emit method bound on
+      `StackLevelAdapter` so `log.<name>('msg')` works (and so
+      `get_logger()`'s per-level method audit passes).
+
+    Idempotent: re-registering an existing name is a no-op-ish
+    refresh (won't clobber an already-bound method).
+
+    '''
+    name_up: str = name.upper()
+    name_lo: str = name.lower()
+
+    CUSTOM_LEVELS[name_up] = value
+    logging.addLevelName(value, name_up)
+    STD_PALETTE[name_up] = color
+    BOLD_PALETTE['bold'][name_up] = f'bold_{color}'
+
+    if not hasattr(StackLevelAdapter, name_lo):
+        # bind via default-arg so `value` is captured (not
+        # late-bound); delegates to `.log()` exactly like the
+        # hand-written level methods above.
+        def _emit(
+            self,
+            msg: str,
+            *,
+            _level: int = value,
+        ) -> None:
+            return self.log(_level, msg)
+
+        _emit.__name__ = name_lo
+        _emit.__qualname__ = f'StackLevelAdapter.{name_lo}'
+        setattr(StackLevelAdapter, name_lo, _emit)
+
+
+# `IO`: child-subproc std-stream relay (see
+# `tractor.trionics._subproc`). Value 21 sits just ABOVE
+# `INFO`(20) so it's SHOWN BY DEFAULT at the usual `info`/`devx`
+# console levels (a `runtime`(15) relay would be silently
+# filtered) yet still distinctly labelled/colored + separately
+# filterable.
+add_log_level('IO', 21, 'purple')
+
+
 # TODO IDEAs:
 # -[ ] move to `.devx.pformat`?
 # -[ ] do per task-name and actor-name color coding

tractor/runtime/_runtime.py

@@ -873,9 +873,12 @@ class Actor:
             if self._spawn_method in (
                 'trio',
                 'subint',
-                # `subint_forkserver` parent-side sends a
+                # `main_thread_forkserver` (and the future
-                # `SpawnSpec` over IPC just like the other two
+                # variant-2 `subint_forkserver`) parent-side
-                # — fork child-side runtime is trio-native.
+                # sends a `SpawnSpec` over IPC just like the
+                # other two — fork child-side runtime is
+                # trio-native.
+                'main_thread_forkserver',
                 'subint_forkserver',
             ):
 

tractor/runtime/_state.py

@@ -139,7 +139,7 @@ _RUNTIME_VARS_DEFAULTS: dict[str, Any] = {
     # `debug_mode: bool` settings
     '_debug_mode': False,  # bool
     'repl_fixture': False,  # |AbstractContextManager[bool]
-    
+
     'use_greenback': False,  # `.pause_from_sync()`/`breakpoint()`
     'use_stackscope': False,  # trio-task-stack dumps on SIGUSR1
 

tractor/spawn/_spawn.py

@@ -22,6 +22,7 @@ over multiple backends.
 from __future__ import annotations
 import multiprocessing as mp
 import platform
+import sys
 from typing import (
     Any,
     Awaitable,
@@ -42,6 +43,7 @@ from tractor.log import get_logger
 from tractor.discovery._addr import (
     UnwrappedAddress,
 )
+from .._exceptions import ActorFailure
 from ._reap import unlink_uds_bind_addrs
 from tractor.runtime._portal import Portal
 from tractor.runtime._runtime import Actor
@@ -64,6 +66,34 @@ SpawnMethodKey = Literal[
     'trio',  # supported on all platforms
     'mp_spawn',
     'mp_forkserver',  # posix only
+    'subint',  # py3.14+ via `concurrent.interpreters` (PEP 734)
+    # EXPERIMENTAL — blocked at the CPython level. The
+    # design goal was a `trio+fork`-safe subproc spawn via
+    # `os.fork()` from a trio-free launchpad sub-interpreter,
+    # but CPython's `PyOS_AfterFork_Child` → `_PyInterpreterState_DeleteExceptMain`
+    # requires fork come from the main interp. See
+    # `tractor.spawn._subint_fork` +
+    # `ai/conc-anal/subint_fork_blocked_by_cpython_post_fork_issue.md`
+    # + issue #379 for the full analysis.
+    'subint_fork',
+    # EXPERIMENTAL — the `subint_fork` workaround. `os.fork()`
+    # from a non-trio worker thread (never entered a subint)
+    # is CPython-legal and works cleanly; forked child runs
+    # `tractor._child._actor_child_main()` against a trio
+    # runtime, exactly like `trio_proc` but via fork instead
+    # of subproc-exec. See `tractor.spawn._main_thread_forkserver`.
+    'main_thread_forkserver',
+    # Variant-2: same fork machinery as `main_thread_forkserver`
+    # but the child enters a sub-interpreter to host its
+    # `trio.run()`. Gated on jcrist/msgspec#1026 unblocking
+    # PEP 684 isolated-mode subints upstream — until then
+    # `subint_forkserver_proc` is a clean `NotImplementedError`
+    # stub pointing at variant-1 (`main_thread_forkserver`) +
+    # the upstream blocker. The key is reserved here (not just
+    # aliased to variant-1) so once upstream lands the impl can
+    # flip in-place without API churn. See
+    # `tractor.spawn._subint_forkserver`.
+    'subint_forkserver',
 ]
 _spawn_method: SpawnMethodKey = 'trio'
 
@@ -81,6 +111,71 @@ else:
         await trio.lowlevel.wait_readable(proc.sentinel)
 
 
+async def wait_for_peer_or_proc_death(
+    ipc_server,
+    uid: tuple[str, str],
+    # TODO? not not types?
+    proc_wait: 'Callable[[], Awaitable]',
+    proc_repr: str = '',
+
+) -> 'tuple[trio.Event, Channel]':
+    '''
+    Race `IPCServer.wait_for_peer(uid)` against the sub-proc's
+    own `.wait()` coroutine. Whichever completes first cancels
+    the other.
+
+    Used by every spawn-backend to detect a sub-actor that
+    *dies during boot* before completing the parent-handshake-
+    callback (e.g. crashed on import, exec'd-out, kernel-killed
+    pre-`_actor_child_main`). Without this race, the
+    handshake-wait — backed by an unsignalled `trio.Event` —
+    parks the spawning task forever and leaves the dead child
+    as a zombie since nobody calls `proc.wait()` to reap.
+
+    On normal handshake-complete: returns `(event, chan)`
+    identical to a bare `wait_for_peer`.
+
+    On proc-death-first: raises `ActorFailure` carrying the
+    proc's exit code, allowing the supervisor to surface a
+    clean error rather than hanging indefinitely.
+
+    `proc_wait` is a 0-arg async callable returning the proc's
+    exit-status — kept generic so each backend can pass its
+    own (`trio.Process.wait`, `_ForkedProc.wait`,
+    `proc_waiter(mp.Process)`, etc.).
+
+    `proc_repr` is an optional string used in the
+    `ActorFailure` message for diag.
+
+    '''
+    result: dict = {}
+
+    async def _await_handshake():
+        event, chan = await ipc_server.wait_for_peer(uid)
+        result['handshake'] = (event, chan)
+        boot_n.cancel_scope.cancel()
+
+    async def _await_death():
+        rc = await proc_wait()
+        result['died'] = rc
+        boot_n.cancel_scope.cancel()
+
+    async with trio.open_nursery() as boot_n:
+        boot_n.start_soon(_await_handshake)
+        boot_n.start_soon(_await_death)
+
+    if 'handshake' in result:
+        return result['handshake']
+
+    # only reached if proc-death won the race
+    raise ActorFailure(
+        f'Sub-actor {uid!r} died during boot '
+        f'(rc={result.get("died")!r}) before completing '
+        f'parent-handshake.\n'
+        f'  proc: {proc_repr}'
+    )
+
+
 def try_set_start_method(
     key: SpawnMethodKey
 
@@ -113,7 +208,38 @@ def try_set_start_method(
         case 'mp_spawn':
             _ctx = mp.get_context('spawn')
 
-        case 'trio':
+        case (
+            'trio'
+            | 'main_thread_forkserver'
+        ):
+            _ctx = None
+
+        case (
+            'subint'
+            | 'subint_fork'
+            | 'subint_forkserver'
+        ):
+            # All subint-family backends need no `mp.context`;
+            # all four feature-gate on the py3.14 public
+            # `concurrent.interpreters` wrapper (PEP 734). See
+            # `tractor.spawn._subint` for the detailed
+            # reasoning. `subint_fork` is blocked at the
+            # CPython level (raises `NotImplementedError`);
+            # `main_thread_forkserver` is the working
+            # variant-1 backend; `subint_forkserver` aliases
+            # to it today, reserved for the future variant-2
+            # subint-isolated-child runtime once upstream
+            # msgspec#1026 unblocks.
+            from ._subint import _has_subints
+            if not _has_subints:
+                raise RuntimeError(
+                    f'Spawn method {key!r} requires Python 3.14+.\n'
+                    f'(On py3.13 the private `_interpreters` C '
+                    f'module exists but tractor\'s spawn flow '
+                    f'wedges — see `tractor.spawn._subint` '
+                    f'docstring for details.)\n'
+                    f'Current runtime: {sys.version}'
+                )
             _ctx = None
 
         case _:
@@ -465,6 +591,10 @@ async def new_proc(
 # `hard_kill`/`proc_waiter` from this module.
 from ._trio import trio_proc
 from ._mp import mp_proc
+from ._subint import subint_proc
+from ._subint_fork import subint_fork_proc
+from ._main_thread_forkserver import main_thread_forkserver_proc
+from ._subint_forkserver import subint_forkserver_proc
 
 
 # proc spawning backend target map
@@ -472,4 +602,25 @@ _methods: dict[SpawnMethodKey, Callable] = {
     'trio': trio_proc,
     'mp_spawn': mp_proc,
     'mp_forkserver': mp_proc,
+    'subint': subint_proc,
+    # blocked at CPython level — see `_subint_fork.py` +
+    # `ai/conc-anal/subint_fork_blocked_by_cpython_post_fork_issue.md`.
+    # Kept here so `--spawn-backend=subint_fork` routes to a
+    # clean `NotImplementedError` with pointer to the analysis,
+    # rather than an "invalid backend" error.
+    'subint_fork': subint_fork_proc,
+    # Variant-1 (working today): fork from a regular main-interp
+    # worker thread, child runs trio on its own main interp.
+    # Validated by
+    # `ai/conc-anal/subint_fork_from_main_thread_smoketest.py`.
+    # See `tractor.spawn._main_thread_forkserver`.
+    'main_thread_forkserver': main_thread_forkserver_proc,
+    # Variant-2 (future, reserved): same fork machinery but
+    # child enters a sub-interpreter to host its `trio.run()`
+    # — gated on jcrist/msgspec#1026 unblocking PEP 684
+    # isolated-mode subints. Today the stub raises
+    # `NotImplementedError` pointing at the variant-1 backend
+    # + upstream blocker. See
+    # `tractor.spawn._subint_forkserver`.
+    'subint_forkserver': subint_forkserver_proc,
 }

tractor/spawn/_subint.py

@@ -0,0 +1,473 @@
+# tractor: structured concurrent "actors".
+# Copyright 2018-eternity Tyler Goodlet.
+
+# This program is free software: you can redistribute it and/or modify
+# it under the terms of the GNU Affero General Public License as published by
+# the Free Software Foundation, either version 3 of the License, or
+# (at your option) any later version.
+
+# This program is distributed in the hope that it will be useful,
+# but WITHOUT ANY WARRANTY; without even the implied warranty of
+# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+# GNU Affero General Public License for more details.
+
+# You should have received a copy of the GNU Affero General Public License
+# along with this program.  If not, see <https://www.gnu.org/licenses/>.
+
+'''
+Sub-interpreter (`subint`) actor spawning backend.
+
+Spawns each sub-actor as a CPython PEP 734 sub-interpreter
+(`concurrent.interpreters.Interpreter`) driven on its own OS
+thread — same-process state isolation with faster start-up
+than an OS subproc, while preserving tractor's existing
+IPC-based actor boundary.
+
+Availability
+------------
+Requires Python **3.14+**. The private `_interpreters` C
+module we actually call into has shipped since 3.13, but
+that vintage has a latent bug in its thread/subint
+interaction which wedges tractor's spawn flow after
+`_interpreters.create()` — the driver `threading.Thread`
+silently never makes progress inside `_interpreters.exec()`.
+(Minimal standalone reproductions with threading +
+`_interpreters.exec()` work fine on 3.13; only our
+multi-trio-task usage triggers the hang. 3.14 fixes it.)
+On older runtimes the module still imports (so the registry
+stays introspectable) but `subint_proc()` raises.
+
+'''
+from __future__ import annotations
+import sys
+import threading
+from typing import (
+    Any,
+    TYPE_CHECKING,
+)
+
+import trio
+from trio import TaskStatus
+
+
+# NOTE: we reach into the *private* `_interpreters` C module
+# for the actual subint create/exec/destroy calls rather than
+# `concurrent.interpreters`' public API because the public API
+# only exposes PEP 734's `'isolated'` config (per-interp GIL).
+# Under `'isolated'`, any C extension missing the
+# `Py_mod_multiple_interpreters` slot (PEP 684) refuses to
+# import; in our stack that's `msgspec` — which tractor uses
+# pervasively in the IPC layer — so isolated-mode subints
+# can't finish booting the sub-actor's `trio.run()`. msgspec
+# PEP 684 support is open upstream at jcrist/msgspec#563.
+#
+# Dropping to the `'legacy'` config keeps the main GIL + lets
+# existing C extensions load normally while preserving the
+# state isolation we actually care about for the actor model
+# (separate `sys.modules` / `__main__` / globals).
+#
+# But — we feature-gate on the **public** `concurrent.interpreters`
+# module (3.14+) even though we only call into the private
+# `_interpreters` module. Reason: the private module has
+# shipped since 3.13, but the thread/subint interactions
+# tractor relies on (`threading.Thread` driving
+# `_interpreters.exec(..., legacy)` while a trio loop runs in
+# the parent + another inside the subint + IPC between them)
+# hang silently on 3.13 and only work cleanly on 3.14. See
+# docstring above for the empirical details. Using the public
+# module's existence as the gate keeps this check honest.
+#
+# Migration path: when msgspec (jcrist/msgspec#563) and any
+# other PEP 684-holdout C deps opt-in, we can switch to the
+# public `concurrent.interpreters.create()` API (isolated
+# mode) and pick up per-interp-GIL parallelism for free.
+#
+# References:
+# - PEP 734 (`concurrent.interpreters` public API):
+#   https://peps.python.org/pep-0734/
+# - PEP 684 (per-interpreter GIL / `Py_mod_multiple_interpreters`):
+#   https://peps.python.org/pep-0684/
+# - stdlib docs (3.14+):
+#   https://docs.python.org/3.14/library/concurrent.interpreters.html
+# - CPython public wrapper source (`Lib/concurrent/interpreters/`):
+#   https://github.com/python/cpython/tree/main/Lib/concurrent/interpreters
+# - CPython private C ext source
+#   (`Modules/_interpretersmodule.c`):
+#   https://github.com/python/cpython/blob/main/Modules/_interpretersmodule.c
+# - msgspec PEP 684 upstream tracker:
+#   https://github.com/jcrist/msgspec/issues/563
+try:
+    # gate: presence of the public 3.14 stdlib wrapper (we
+    # don't actually use it below, see NOTE above).
+    from concurrent import interpreters as _public_interpreters  # noqa: F401  # type: ignore
+    # actual driver: the private C module (also present on
+    # 3.13 but we refuse that version — see gate above).
+    import _interpreters  # type: ignore
+    _has_subints: bool = True
+except ImportError:
+    _interpreters = None  # type: ignore
+    _has_subints: bool = False
+
+
+from tractor.log import get_logger
+from tractor.msg import (
+    types as msgtypes,
+    pretty_struct,
+)
+from tractor.runtime._state import current_actor
+from tractor.runtime._portal import Portal
+from ._spawn import cancel_on_completion
+
+
+if TYPE_CHECKING:
+    from tractor.discovery._addr import UnwrappedAddress
+    from tractor.ipc import (
+        _server,
+    )
+    from tractor.runtime._runtime import Actor
+    from tractor.runtime._supervise import ActorNursery
+
+
+log = get_logger('tractor')
+
+
+# How long we'll wait (in seconds) inside the shielded soft-kill
+# / teardown blocks before abandoning the sub-interpreter to its
+# fate. See the "hard-kill" comments at the two shield sites.
+#
+# Unbounded shields are a Bad Idea with subints: because CPython
+# doesn't deliver SIGINT into sub-interpreters and the legacy
+# config shares the main GIL, a stuck subint can otherwise lock
+# the parent trio loop (and the user's Ctrl-C) indefinitely.
+_HARD_KILL_TIMEOUT: float = 3.0
+
+
+async def subint_proc(
+    name: str,
+    actor_nursery: ActorNursery,
+    subactor: Actor,
+    errors: dict[tuple[str, str], Exception],
+
+    # passed through to actor main
+    bind_addrs: list[UnwrappedAddress],
+    parent_addr: UnwrappedAddress,
+    _runtime_vars: dict[str, Any],  # serialized and sent to _child
+    *,
+    infect_asyncio: bool = False,
+    task_status: TaskStatus[Portal] = trio.TASK_STATUS_IGNORED,
+    proc_kwargs: dict[str, any] = {}
+
+) -> None:
+    '''
+    Create a new sub-actor hosted inside a PEP 734
+    sub-interpreter running on a dedicated OS thread,
+    reusing tractor's existing UDS/TCP IPC handshake
+    for parent<->child channel setup.
+
+    Supervision model mirrors `trio_proc()`:
+    - parent awaits `ipc_server.wait_for_peer()` for the
+      child to connect back; on success yields a `Portal`
+      via `task_status.started()`
+    - on graceful shutdown we await the sub-interpreter's
+      `trio.run()` completing naturally (driven by the
+      child's actor runtime)
+    - on cancellation we send `Portal.cancel_actor()` and
+      then wait for the subint's trio loop to exit cleanly
+      — unblocking the worker thread so the `Interpreter`
+      can be closed
+
+    '''
+    if not _has_subints:
+        raise RuntimeError(
+            f'The {"subint"!r} spawn backend requires Python 3.14+.\n'
+            f'(On py3.13 the private `_interpreters` C module '
+            f'exists but tractor\'s spawn flow wedges — see '
+            f'`tractor.spawn._subint` docstring for details.)\n'
+            f'Current runtime: {sys.version}'
+        )
+
+    interp_id: int = _interpreters.create('legacy')
+    log.runtime(
+        f'Created sub-interpreter (legacy cfg) for sub-actor\n'
+        f'(>\n'
+        f' |_interp_id={interp_id}\n'
+    )
+
+    uid: tuple[str, str] = subactor.aid.uid
+    loglevel: str | None = subactor.loglevel
+
+    # Build a bootstrap code string driven via `_interpreters.exec()`.
+    # All of `uid` (`tuple[str, str]`), `loglevel` (`str|None`),
+    # `parent_addr` (`tuple[str, int|str]` — see `UnwrappedAddress`)
+    # and `infect_asyncio` (`bool`) `repr()` to valid Python
+    # literals, so we can embed them directly.
+    #
+    # ?TODO, future SpawnSpec enrichment: if we ever want to pass
+    # non-`repr()`-roundtrippable values (a pre-built `SpawnSpec`
+    # struct, a credential token, a callable) we should switch to
+    # `_interpreters.set___main___attrs(interp_id, {...})` — the
+    # API anyio uses in `to_interpreter._Worker.call()`. Pattern:
+    # https://github.com/agronholm/anyio/blob/master/src/anyio/to_interpreter.py
+    # (the `set___main___attrs` site). See also tracking issue
+    # `#379`.
+    bootstrap: str = (
+        'from tractor._child import _actor_child_main\n'
+        '_actor_child_main(\n'
+        f'    uid={uid!r},\n'
+        f'    loglevel={loglevel!r},\n'
+        f'    parent_addr={parent_addr!r},\n'
+        f'    infect_asyncio={infect_asyncio!r},\n'
+        f'    spawn_method={"subint"!r},\n'
+        ')\n'
+    )
+
+    cancelled_during_spawn: bool = False
+    subint_exited = trio.Event()
+    ipc_server: _server.Server = actor_nursery._actor.ipc_server
+
+    # Capture a trio token so the driver thread can signal
+    # `subint_exited.set()` back into the parent trio loop.
+    trio_token = trio.lowlevel.current_trio_token()
+
+    def _subint_target() -> None:
+        '''
+        Dedicated OS-thread target: runs `_interpreters.exec()`
+        once and exits.
+
+        We intentionally use a plain `threading.Thread` here
+        rather than `trio.to_thread.run_sync()` because trio's
+        thread cache would *recycle* the same OS thread for
+        subsequent jobs — leaving CPython's subinterpreter
+        tstate attached to that cached worker and blocking
+        `_interpreters.destroy()` in the teardown block below.
+        A dedicated thread truly exits after `exec()` returns,
+        releasing the tstate so destroy can proceed.
+
+        '''
+        try:
+            _interpreters.exec(interp_id, bootstrap)
+        # XXX without this catch, a hard subint-bootstrap
+        # failure (e.g. `ImportError` because the actor module
+        # isn't importable inside the subint, or a syntax error
+        # in the bootstrap str) goes only to Python's default
+        # thread-excepthook and is INVISIBLE to the parent
+        # task. At minimum, log via `log.exception` so the
+        # operator sees what failed.
+        # ?TODO, surface the captured exc to the parent task
+        # via a `nonlocal err` slot inspected after
+        # `subint_exited.wait()` — see anyio's
+        # `to_interpreter._interp_call` `(retval, is_exception)`
+        # tuple pattern + `_subint_forkserver.py:480-494`'s
+        # equivalent which already does this. Skipped here for
+        # now: re-raise from the parent must coordinate with
+        # the existing `trio.Cancelled` paths around the
+        # `subint_exited.wait()` calls (lines 327, 362).
+        # NOTE: this whole dedicated-thread machinery may go
+        # away under #450 (PEP 684 isolated mode), in which
+        # case `trio.to_thread.run_sync(Interpreter.exec, ...)`
+        # would handle exception propagation natively.
+        except BaseException:
+            log.exception(
+                f'subint bootstrap raised — interp_id={interp_id}'
+            )
+        finally:
+            try:
+                trio.from_thread.run_sync(
+                    subint_exited.set,
+                    trio_token=trio_token,
+                )
+            except trio.RunFinishedError:
+                # parent trio loop has already exited (proc
+                # teardown); nothing to signal.
+                pass
+
+    # NOTE: `daemon=True` so a stuck subint can never block
+    # process exit — if our `_HARD_KILL_TIMEOUT` paths below
+    # have to abandon this thread, Python's interpreter
+    # shutdown won't wait for it forever. Tradeoff: any
+    # subint state still live at abandon-time may leak.
+    driver_thread = threading.Thread(
+        target=_subint_target,
+        name=f'subint-driver[{interp_id}]',
+        daemon=True,
+    )
+
+    try:
+        try:
+            driver_thread.start()
+
+            try:
+                event, chan = await ipc_server.wait_for_peer(uid)
+            except trio.Cancelled:
+                cancelled_during_spawn = True
+                raise
+
+            portal = Portal(chan)
+            actor_nursery._children[uid] = (
+                subactor,
+                interp_id,  # proxy for the normal `proc` slot
+                portal,
+            )
+
+            sspec = msgtypes.SpawnSpec(
+                _parent_main_data=subactor._parent_main_data,
+                enable_modules=subactor.enable_modules,
+                reg_addrs=subactor.reg_addrs,
+                bind_addrs=bind_addrs,
+                _runtime_vars=_runtime_vars,
+            )
+            log.runtime(
+                f'Sending spawn spec to subint child\n'
+                f'{{}}=> {chan.aid.reprol()!r}\n'
+                f'\n'
+                f'{pretty_struct.pformat(sspec)}\n'
+            )
+            await chan.send(sspec)
+
+            curr_actor: Actor = current_actor()
+            curr_actor._actoruid2nursery[uid] = actor_nursery
+
+            task_status.started(portal)
+
+            with trio.CancelScope(shield=True):
+                await actor_nursery._join_procs.wait()
+
+            async with trio.open_nursery() as lifecycle_n:
+                if portal in actor_nursery._cancel_after_result_on_exit:
+                    lifecycle_n.start_soon(
+                        cancel_on_completion,
+                        portal,
+                        subactor,
+                        errors,
+                    )
+
+                # Soft-kill analog: wait for the subint to exit
+                # naturally; on cancel, send a graceful cancel
+                # via the IPC portal and then wait for the
+                # driver thread to finish so `_interpreters.destroy()`
+                # won't race with a running interpreter.
+                try:
+                    await subint_exited.wait()
+                except trio.Cancelled:
+                    # Bounded shield: we want to ATTEMPT a
+                    # graceful cancel via the portal, but we
+                    # MUST NOT let the shield trap user
+                    # Ctrl-C / parent teardown forever if the
+                    # subint is already unreachable (e.g., the
+                    # IPC channel was broken — which is exactly
+                    # what `test_ipc_channel_break_during_stream`
+                    # exercises). After `_HARD_KILL_TIMEOUT` we
+                    # drop the shield and let `Cancelled`
+                    # propagate; the outer teardown will force
+                    # things along.
+                    with (
+                        trio.CancelScope(shield=True),
+                        trio.move_on_after(
+                            _HARD_KILL_TIMEOUT,
+                        ) as cs,
+                    ):
+                        log.cancel(
+                            f'Soft-killing subint sub-actor\n'
+                            f'c)=> {chan.aid.reprol()}\n'
+                            f'   |_interp_id={interp_id}\n'
+                        )
+                        try:
+                            await portal.cancel_actor()
+                        except (
+                            trio.BrokenResourceError,
+                            trio.ClosedResourceError,
+                        ):
+                            # channel already down — subint will
+                            # exit on its own timeline (or won't,
+                            # in which case the timeout below
+                            # is our escape).
+                            pass
+                        await subint_exited.wait()
+                    if cs.cancelled_caught:
+                        log.warning(
+                            f'Soft-kill of subint sub-actor timed '
+                            f'out after {_HARD_KILL_TIMEOUT}s — '
+                            f'subint may still be running; '
+                            f'escalating to thread-abandon.\n'
+                            f'   |_interp_id={interp_id}\n'
+                            f'   |_aid={chan.aid.reprol()}\n'
+                        )
+                    raise
+                finally:
+                    lifecycle_n.cancel_scope.cancel()
+
+        finally:
+            # Ensure the driver thread is *fully* joined before
+            # destroying the subint. `subint_exited.set()` fires
+            # from inside the thread but returns to trio before
+            # the thread's bootstrap cleanup finishes; calling
+            # `destroy()` too eagerly can race with tstate
+            # teardown. Off-load the blocking `.join()` to a
+            # cache thread (which carries no subint tstate of
+            # its own, so no cache conflict).
+            #
+            # Bounded shield: if the driver thread never exits
+            # (soft-kill failed above, subint stuck in
+            # non-checkpointing Python, etc.) we MUST abandon
+            # it rather than wedge the parent forever. The
+            # thread is `daemon=True` so proc-exit won't block
+            # on it either. Subsequent `_interpreters.destroy()`
+            # on a still-running subint raises `InterpreterError`
+            # which we log and swallow — the abandoned subint
+            # will be torn down by process exit.
+            with (
+                trio.CancelScope(shield=True),
+                trio.move_on_after(_HARD_KILL_TIMEOUT) as cs,
+            ):
+                if driver_thread.is_alive():
+                    # XXX `abandon_on_cancel=True` is load-bearing:
+                    # the default (False) makes `to_thread.run_sync`
+                    # ignore the enclosing `move_on_after` and
+                    # block until `driver_thread.join()` returns —
+                    # which is exactly what we can't wait for here.
+                    await trio.to_thread.run_sync(
+                        driver_thread.join,
+                        abandon_on_cancel=True,
+                    )
+            if cs.cancelled_caught:
+                # Disambiguate "thread leaked but subint already
+                # done" from "thread alive because subint is
+                # genuinely wedged" — pattern borrowed from
+                # trio-parallel's `_sint.SintWorker.is_alive()`.
+                still_running: bool = _interpreters.is_running(
+                    interp_id,
+                )
+                log.warning(
+                    f'Subint driver thread did not exit within '
+                    f'{_HARD_KILL_TIMEOUT}s — abandoning.\n'
+                    f'   |_interp_id={interp_id}\n'
+                    f'   |_thread={driver_thread.name}\n'
+                    f'   |_subint_still_running={still_running}\n'
+                    f'(This usually means portal-cancel could '
+                    f'not be delivered — e.g., IPC channel was '
+                    f'already broken. The subint will continue '
+                    f'running until process exit terminates the '
+                    f'daemon thread.)'
+                )
+
+            with trio.CancelScope(shield=True):
+                try:
+                    _interpreters.destroy(interp_id)
+                    log.runtime(
+                        f'Destroyed sub-interpreter\n'
+                        f')>\n'
+                        f' |_interp_id={interp_id}\n'
+                    )
+                except _interpreters.InterpreterError as e:
+                    log.warning(
+                        f'Could not destroy sub-interpreter '
+                        f'{interp_id}: {e}\n'
+                        f'(expected if the driver thread was '
+                        f'abandoned above; the subint is still '
+                        f'running and will be reaped at process '
+                        f'exit.)'
+                    )
+
+    finally:
+        if not cancelled_during_spawn:
+            actor_nursery._children.pop(uid, None)

tractor/spawn/_subint_fork.py

@@ -0,0 +1,153 @@
+# tractor: structured concurrent "actors".
+# Copyright 2018-eternity Tyler Goodlet.
+
+# This program is free software: you can redistribute it and/or modify
+# it under the terms of the GNU Affero General Public License as published by
+# the Free Software Foundation, either version 3 of the License, or
+# (at your option) any later version.
+
+# This program is distributed in the hope that it will be useful,
+# but WITHOUT ANY WARRANTY; without even the implied warranty of
+# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+# GNU Affero General Public License for more details.
+
+# You should have received a copy of the GNU Affero General Public License
+# along with this program.  If not, see <https://www.gnu.org/licenses/>.
+
+'''
+`subint_fork` spawn backend — BLOCKED at CPython level.
+
+The idea was to use a sub-interpreter purely as a launchpad
+from which to call `os.fork()`, sidestepping the well-known
+trio+fork issues (python-trio/trio#1614 etc.) by guaranteeing
+the forking interp had never imported `trio`.
+
+**IT DOES NOT WORK ON CURRENT CPYTHON.** The fork syscall
+itself succeeds (in the parent), but the forked CHILD
+process aborts immediately during CPython's post-fork
+cleanup — `PyOS_AfterFork_Child()` calls
+`_PyInterpreterState_DeleteExceptMain()` which refuses to
+operate when the current tstate belongs to a non-main
+sub-interpreter.
+
+Full annotated walkthrough from the user-visible error
+(`Fatal Python error: _PyInterpreterState_DeleteExceptMain:
+not main interpreter`) down to the specific CPython source
+lines that enforce this is in
+`ai/conc-anal/subint_fork_blocked_by_cpython_post_fork_issue.md`.
+
+We keep this submodule as a dedicated documentation of the
+attempt. If CPython ever lifts the restriction (e.g., via a
+force-destroy primitive or a hook that swaps tstate to main
+pre-fork), the structural sketch preserved in this file's
+git history is a concrete starting point for a working impl.
+
+See also: issue #379's "Our own thoughts, ideas for
+`fork()`-workaround/hacks..." section.
+
+'''
+from __future__ import annotations
+import sys
+from typing import (
+    Any,
+    TYPE_CHECKING,
+)
+
+import trio
+from trio import TaskStatus
+
+from tractor.runtime._portal import Portal
+from ._subint import _has_subints
+
+
+if TYPE_CHECKING:
+    from tractor.discovery._addr import UnwrappedAddress
+    from tractor.runtime._runtime import Actor
+    from tractor.runtime._supervise import ActorNursery
+
+
+async def subint_fork_proc(
+    name: str,
+    actor_nursery: ActorNursery,
+    subactor: Actor,
+    errors: dict[tuple[str, str], Exception],
+
+    bind_addrs: list[UnwrappedAddress],
+    parent_addr: UnwrappedAddress,
+    _runtime_vars: dict[str, Any],
+    *,
+    infect_asyncio: bool = False,
+    task_status: TaskStatus[Portal] = trio.TASK_STATUS_IGNORED,
+    proc_kwargs: dict[str, any] = {},
+
+) -> None:
+    '''
+    EXPERIMENTAL — currently blocked by a CPython invariant.
+
+    Attempted design
+    ----------------
+    1. Parent creates a fresh legacy-config subint.
+    2. A worker OS-thread drives the subint through a
+       bootstrap that calls `os.fork()`.
+    3. In the forked CHILD, `os.execv()` back into
+       `python -m tractor._child` (fresh process).
+    4. In the fork-PARENT, the launchpad subint is destroyed;
+       parent-side trio task proceeds identically to
+       `trio_proc()` (wait for child connect-back, send
+       `SpawnSpec`, yield `Portal`, etc.).
+
+    Why it doesn't work
+    -------------------
+    CPython's `PyOS_AfterFork_Child()` (in
+    `Modules/posixmodule.c`) calls
+    `_PyInterpreterState_DeleteExceptMain()` (in
+    `Python/pystate.c`) as part of post-fork cleanup. That
+    function requires the current `PyThreadState` belong to
+    the **main** interpreter. When `os.fork()` is called
+    from within a sub-interpreter, the child wakes up with
+    its tstate still pointing at the (now-stale) subint, and
+    this check fails with `PyStatus_ERR("not main
+    interpreter")`, triggering a `fatal_error` goto and
+    aborting the child process.
+
+    CPython devs acknowledge the fragility with a
+    `// Ideally we could guarantee tstate is running main.`
+    comment right above the call site.
+
+    See
+    `ai/conc-anal/subint_fork_blocked_by_cpython_post_fork_issue.md`
+    for the full annotated walkthrough + upstream-report
+    draft.
+
+    Why we keep this stub
+    ---------------------
+    - Documents the attempt in-tree so the next person who
+      has this idea finds the reason it doesn't work rather
+      than rediscovering the same CPython-level dead end.
+    - If CPython ever lifts the restriction (e.g., via a
+      force-destroy primitive or a hook that swaps tstate
+      to main pre-fork), this submodule's git history holds
+      the structural sketch of what a working impl would
+      look like.
+
+    '''
+    if not _has_subints:
+        raise RuntimeError(
+            f'The {"subint_fork"!r} spawn backend requires '
+            f'Python 3.14+.\n'
+            f'Current runtime: {sys.version}'
+        )
+
+    raise NotImplementedError(
+        'The `subint_fork` spawn backend is blocked at the '
+        'CPython level — `os.fork()` from a non-main '
+        'sub-interpreter is refused by '
+        '`PyOS_AfterFork_Child()` → '
+        '`_PyInterpreterState_DeleteExceptMain()`, which '
+        'aborts the child with '
+        '`Fatal Python error: not main interpreter`.\n'
+        '\n'
+        'See '
+        '`ai/conc-anal/subint_fork_blocked_by_cpython_post_fork_issue.md` '
+        'for the full analysis + upstream-report draft.'
+    )

tractor/spawn/_subint_forkserver.py

@@ -0,0 +1,313 @@
+# tractor: structured concurrent "actors".
+# Copyright 2018-eternity Tyler Goodlet.
+
+# This program is free software: you can redistribute it and/or modify
+# it under the terms of the GNU Affero General Public License as published by
+# the Free Software Foundation, either version 3 of the License, or
+# (at your option) any later version.
+
+# This program is distributed in the hope that it will be useful,
+# but WITHOUT ANY WARRANTY; without even the implied warranty of
+# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+# GNU Affero General Public License for more details.
+
+# You should have received a copy of the GNU Affero General Public License
+# along with this program.  If not, see <https://www.gnu.org/licenses/>.
+
+'''
+Variant-2 (future) "subint forkserver" placeholder — reserved
+for the eventual subint-isolated-child runtime variant.
+
+> **Status:** reserved key, stub impl. Today
+> `--spawn-backend=subint_forkserver` raises a clean
+> `NotImplementedError` from `subint_forkserver_proc()`
+> below, pointing at variant-1
+> (`--spawn-backend=main_thread_forkserver`, see
+> `tractor.spawn._main_thread_forkserver`) and the upstream
+> blocker
+> ([jcrist/msgspec#1026](https://github.com/jcrist/msgspec/issues/1026)).
+> The key is reserved here (not aliased to variant-1) so the
+> literal subint-hosted-child impl can flip in-place once
+> msgspec#1026 unblocks PEP 684 isolated-mode subints
+> upstream — no API churn at the call site.
+
+Future arch — what subints would buy us
+---------------------------------------
+
+When msgspec#1026 unblocks isolated-mode subints (PEP 684
+per-interp GIL), three concrete wins land — these are the
+reason the `'subint_forkserver'` key is reserved as a
+distinct backend rather than just folded into
+`'main_thread_forkserver'`:
+
+**(1) Cheaper forks (smaller main-interp COW image)**
+
+Today (variant 1) the parent's main interp carries the full
+tractor stack: trio runtime, msgspec codecs, IPC layer,
+every user module the actor imported. When the forkserver
+worker calls `os.fork()` the child inherits ALL of that as
+COW memory — even though most gets overwritten when the
+child boots its own `trio.run()`.
+
+Variant 2 moves the parent's `trio.run()` into a subint (its
+own `sys.modules` / `__main__` / globals). The main interp
+**stays minimal** — just the forkserver-thread plumbing +
+bare CPython. The main interp becomes the *literal*
+forkserver: an intentionally-empty execution context whose
+only job is to call `os.fork()` cleanly. Inherited COW image
+shrinks proportionally.
+
+**(2) True parallelism between forkserver and trio
+(per-interp GIL)**
+
+Variant-1 today: the forkserver worker and the trio.run()
+thread share the main GIL — when one runs the other waits.
+Spawn requests briefly stall trio while the worker takes
+the GIL to call `os.fork()`. PEP 684 isolated-mode gives
+each subint its own GIL: forkserver thread on main + trio
+on subint actually run in parallel. Spawn latency drops,
+trio loop doesn't notice the fork happening.
+
+**(3) Multi-actor-per-process (the architectural prize)**
+
+The bigger payoff and the reason `_subint.py` (the in-thread
+`subint` backend) exists in parallel with this module. With
+per-interp-GIL subints, one process can host:
+
+- main interp: forkserver thread + bookkeeping
+- subint A: actor 1's `trio.run()`
+- subint B: actor 2's `trio.run()`
+- subint C: ...
+
+`os.fork()` becomes the **last-resort** spawn — used only
+when a new OS process is actually required (cgroups,
+namespaces, security boundary, multi-host distribution).
+Within a single process, subint-per-actor is radically
+cheaper: no fork, no COW, no inherited-fd cleanup — just
+`_interpreters.create()` + `_interpreters.exec()`.
+
+The three backends converge on a coherent story:
+
+- `subint` → in-process spawn (cheap, GIL-isolated),
+- `main_thread_forkserver` → cross-process spawn today
+  (variant 1, working),
+- `subint_forkserver` → cross-process spawn with
+  isolated-subint child (variant 2, this module, future).
+
+What lives here today
+---------------------
+
+- `run_subint_in_worker_thread()` — companion primitive to
+  `_main_thread_forkserver.fork_from_worker_thread()`. Creates
+  a fresh `legacy`-config sub-interpreter and drives a given
+  bootstrap code string through `_interpreters.exec()` on a
+  dedicated worker thread; destroys the subint after the
+  thread joins. Used today by the
+  `subint_fork_from_main_thread_smoketest.py` feasibility
+  check; will be wired into the variant-2
+  `subint_forkserver_proc` spawn-coroutine when it lands.
+- (legacy re-exports of fork primitives kept for backward-
+  compatible imports until external consumers migrate to
+  `_main_thread_forkserver`)
+
+What will live here when variant 2 ships
+----------------------------------------
+
+- `subint_forkserver_proc()` — the variant-2 spawn-backend
+  coroutine. Same fork machinery as variant 1, but the
+  fork-child enters a fresh subint (via
+  `run_subint_in_worker_thread`) before booting its
+  `trio.run()`. Net effect: child runtime is GIL-isolated
+  from the parent + any sibling actors in the same process.
+- A stub `subint_forkserver_proc` is added in a follow-up
+  commit that raises `NotImplementedError(...)` pointing at
+  this docstring + jcrist/msgspec#1026 + tractor #379, so
+  `--spawn-backend=subint_forkserver` errors cleanly today
+  rather than silently aliasing variant 1.
+
+See also
+--------
+
+- `tractor.spawn._main_thread_forkserver` — variant 1,
+  working today; for the full design rationale, fork-
+  semantics analysis, and trio×fork hazard breakdown.
+- `tractor.spawn._subint` — the in-thread `subint` backend
+  (one process, one actor per subint, no fork).
+- `tractor.spawn._subint_fork` — RFC stub for the
+  fork-from-non-main-subint strategy that is blocked at the
+  CPython level.
+- [#379](https://github.com/goodboy/tractor/issues/379)
+  — subint backend umbrella tracking issue.
+- [jcrist/msgspec#1026](https://github.com/jcrist/msgspec/issues/1026)
+  — upstream blocker for PEP 684 isolated-mode subints.
+- [#450](https://github.com/goodboy/tractor/issues/450) —
+  thread-constraints audit follow-up tied to msgspec#1026.
+
+'''
+from __future__ import annotations
+import threading
+from typing import (
+    Any,
+    TYPE_CHECKING,
+)
+
+import trio
+from trio import TaskStatus
+
+from tractor.log import get_logger
+from ._subint import _has_subints
+
+if TYPE_CHECKING:
+    from tractor.discovery._addr import UnwrappedAddress
+    from tractor.runtime._portal import Portal
+    from tractor.runtime._runtime import Actor
+    from tractor.runtime._supervise import ActorNursery
+
+# Private CPython subint API — used by `run_subint_in_worker_thread`
+# below. Imported only when 3.14+ is detected (via `_has_subints`
+# from `_subint`); on older runtimes the symbol is `None` and
+# the function raises a clean `RuntimeError` on entry.
+if _has_subints:
+    import _interpreters  # type: ignore
+else:
+    _interpreters = None  # type: ignore
+
+
+log = get_logger('tractor')
+
+
+def run_subint_in_worker_thread(
+    bootstrap: str,
+    *,
+    thread_name: str = 'subint-trio',
+    join_timeout: float = 10.0,
+
+) -> None:
+    '''
+    Create a fresh legacy-config sub-interpreter and drive
+    the given `bootstrap` code string through
+    `_interpreters.exec()` on a dedicated worker thread.
+
+    Naming mirrors `fork_from_worker_thread()`:
+    "<action>_in_worker_thread" — the action here is "run a
+    subint", not "run trio" per se. Typical `bootstrap`
+    content does import `trio` + call `trio.run()`, but
+    nothing about this primitive requires trio; it's a
+    generic "host a subint on a worker thread" helper.
+    Intended mainly for use inside a fork-child (see
+    `tractor.spawn._subint_forkserver` module docstring) but
+    works anywhere.
+
+    See `tractor.spawn._subint.subint_proc` for the matching
+    pattern tractor uses at the sub-actor level.
+
+    Destroys the subint after the thread joins.
+
+    '''
+    if not _has_subints:
+        raise RuntimeError(
+            'subint-forkserver primitives require Python '
+            '3.14+.'
+        )
+
+    interp_id: int = _interpreters.create('legacy')
+    log.runtime(
+        f'Created child-side subint for trio.run()\n'
+        f'(>\n'
+        f' |_interp_id={interp_id}\n'
+    )
+
+    err: BaseException | None = None
+
+    def _drive() -> None:
+        nonlocal err
+        try:
+            _interpreters.exec(interp_id, bootstrap)
+        except BaseException as e:
+            err = e
+            log.exception(
+                f'Failed to .exec() in subint ??\n'
+                f'_interpreters.exec(\n'
+                f'    interp_id={interp_id!r},\n'
+                f'    bootstrap={bootstrap!r},\n'
+                f') => {err!r}\n'
+            )
+
+    worker: threading.Thread = threading.Thread(
+        target=_drive,
+        name=thread_name,
+        daemon=False,
+    )
+    worker.start()
+    worker.join(timeout=join_timeout)
+
+    try:
+        _interpreters.destroy(interp_id)
+    except _interpreters.InterpreterError as e:
+        log.warning(
+            f'Could not destroy child-side subint '
+            f'{interp_id}: {e}'
+        )
+
+    if worker.is_alive():
+        raise RuntimeError(
+            f'child-side subint trio-driver thread '
+            f'{thread_name!r} did not return within '
+            f'{join_timeout}s.'
+        )
+    if err is not None:
+        raise err
+
+
+async def subint_forkserver_proc(
+    name: str,
+    actor_nursery: ActorNursery,
+    subactor: Actor,
+    errors: dict[tuple[str, str], Exception],
+
+    bind_addrs: list[UnwrappedAddress],
+    parent_addr: UnwrappedAddress,
+    _runtime_vars: dict[str, Any],
+    *,
+    infect_asyncio: bool = False,
+    task_status: TaskStatus[Portal] = trio.TASK_STATUS_IGNORED,
+    proc_kwargs: dict[str, any] = {},
+
+) -> None:
+    '''
+    PLACEHOLDER — variant-2 (subint-isolated child runtime)
+    spawn-backend coroutine. Reserved for the eventual impl
+    that uses `run_subint_in_worker_thread()` in the fork-child
+    to host the child's `trio.run()` inside a fresh subint.
+
+    Today this stub raises immediately so
+    `--spawn-backend=subint_forkserver` errors cleanly with a
+    pointer to the working variant-1 backend
+    (`main_thread_forkserver`) and the upstream blocker
+    ([jcrist/msgspec#1026](https://github.com/jcrist/msgspec/issues/1026)).
+
+    See this module's top-level docstring for the future-arch
+    design + what lives here when the variant-2 impl lands.
+
+    '''
+    raise NotImplementedError(
+        f'`{ "subint_forkserver"!r}` spawn backend is reserved '
+        f'for the future variant-2 (subint-isolated child '
+        f'runtime) — gated on jcrist/msgspec#1026 unblocking '
+        f'PEP 684 isolated-mode subints upstream.\n'
+        f'\n'
+        f'For the working fork-based backend today, use '
+        f'`--spawn-backend=main_thread_forkserver` (variant '
+        f'1: fork from a regular main-interp worker thread, '
+        f'child runs trio on its own main interp).\n'
+        f'\n'
+        f'See:\n'
+        f'  - tractor.spawn._main_thread_forkserver — the '
+        f'working variant-1 impl + design rationale\n'
+        f'  - tractor.spawn._subint_forkserver — this '
+        f'module\'s docstring for the variant-2 future-arch\n'
+        f'  - https://github.com/goodboy/tractor/issues/379 '
+        f'(subint umbrella)\n'
+        f'  - https://github.com/jcrist/msgspec/issues/1026 '
+        f'(upstream PEP 684 blocker)'
+    )

tractor/trionics/__init__.py

@@ -38,3 +38,6 @@ from ._taskc import (
     maybe_raise_from_masking_exc as maybe_raise_from_masking_exc,
     start_or_cancel as start_or_cancel,
 )
+from ._subproc import (
+    supervise_run_process as supervise_run_process,
+)

tractor/trionics/_subproc.py

@@ -0,0 +1,296 @@
+# tractor: distributed structured concurrency.
+# Copyright 2018-eternity Tyler Goodlet.
+
+# This program is free software: you can redistribute it and/or modify
+# it under the terms of the GNU Affero General Public License as published by
+# the Free Software Foundation, either version 3 of the License, or
+# (at your option) any later version.
+
+# This program is distributed in the hope that it will be useful,
+# but WITHOUT ANY WARRANTY; without even the implied warranty of
+# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+# GNU Affero General Public License for more details.
+
+# You should have received a copy of the GNU Affero General Public License
+# along with this program.  If not, see <https://www.gnu.org/licenses/>.
+
+'''
+SC-friendly `trio.run_process()` supervision: a `tn.start()`
+style wrapper which surfaces rc!=0 errors deterministically and
+(optionally) live-relays the child's std-streams to the `tractor`
+log.
+
+'''
+from __future__ import annotations
+from functools import partial
+import subprocess
+import textwrap
+from typing import (
+    Callable,
+)
+
+import trio
+
+from ..log import get_logger
+
+log = get_logger()
+
+
+# sentinel so `supervise_run_process(stdout=...)` can tell
+# "caller passed nothing" (-> tty-safe `DEVNULL` default) from
+# an explicit `stdout=None` (inherit) override.
+_UNSET = object()
+
+
+def _add_stderr_note(
+    cpe: subprocess.CalledProcessError,
+    stderr_bytes: bytes,
+) -> None:
+    '''
+    Attach an indented `|_.stderr:` note to a
+    `CalledProcessError` for legible rc!=0 reporting at
+    teardown.
+
+    '''
+    stderr_str: str = stderr_bytes.decode(errors='replace')
+    cpe.add_note(
+        f'|_.stderr:\n'
+        f'{textwrap.indent(stderr_str, prefix=" "*3)}'
+    )
+
+
+async def _relay_stream_lines(
+    stream: trio.abc.ReceiveStream,
+    *,
+    emit: Callable[[str], None]|None = None,
+    tag: str = '',
+    accum: bytearray|None = None,
+) -> None:
+    '''
+    Concurrently drain a child subproc's `stdout`/`stderr`
+    PIPE; relay each COMPLETE line to `emit` (a bound
+    `log.<level>` method) prefixed with `tag` (e.g.
+    `f'{label}:out'`) and/or append raw bytes to `accum`.
+
+    This reader is MANDATORY whenever a bare
+    `stdout=`/`stderr=PIPE` is used WITHOUT `trio`'s
+    `capture_*` (which would spawn trio's own internal drain
+    task): nothing else drains the OS pipe, so once its kernel
+    buffer (~64KiB) fills the child blocks on `write()` ->
+    deadlock.
+
+    Modes (combine freely):
+    - `emit`-only:  live per-line relay (e.g. `relay_stdout`).
+    - `accum`-only: silent drain + capture (e.g. stderr kept
+      for a `CalledProcessError` note WITHOUT relaying it).
+    - both: relay AND capture (e.g. `relay_stderr` with `check=True`).
+
+    '''
+    # NOTE, mirrors `trio._subprocess`'s internal
+    # `async with stream: async for ...` drain idiom — except
+    # here we EMIT per-line (and/or accumulate) instead of
+    # only accumulating.
+    residual: bytes = b''
+    async with stream:                   # aclose at EOF/cancel
+        async for chunk in stream:       # ends at child-exit EOF
+            if accum is not None:
+                accum += chunk
+            if emit is None:
+                continue                 # drain(+accum)-only
+            buf: bytes = residual + chunk
+            *lines, residual = buf.split(b'\n')
+            for raw in lines:
+                line: str = raw.decode(
+                    errors='replace',
+                ).rstrip('\r')
+                emit(f'[{tag}] {line}')
+
+    # flush any trailing partial (un-newline-term'd) line @ EOF
+    if (
+        emit is not None
+        and
+        residual
+    ):
+        line: str = residual.decode(
+            errors='replace',
+        ).rstrip('\r')
+        emit(f'[{tag}] {line}')
+
+
+async def supervise_run_process(
+    cmd: list[str]|str,
+    *,
+    check: bool = True,
+    label: str|None = None,
+
+    # per-line `log.*` relay of the child's std-streams
+    # (tty-safe, capture-safe, STREAMED — not
+    # buffered-until-exit, so it suits long-lived daemons).
+    relay_stdout: bool = False,
+    relay_stderr: bool = False,
+
+    # default `io` (our custom level, value 21): the relay
+    # exists to make windowless-spawn output VISIBLE, and
+    # `IO`(21) sorts just ABOVE `INFO`(20) so it shows at the
+    # usual `info`/`devx` console levels (a `runtime`(15) relay
+    # would be silently filtered) while staying distinctly
+    # labelled + separately filterable.
+    relay_level: str = 'io',
+
+    # non-relay `stdout` override; defaults (via `_UNSET`) to
+    # `DEVNULL` so we NEVER inherit (+ thus can't clobber) the
+    # parent controlling-tty.
+    stdout: int = _UNSET,
+
+    task_status: trio.TaskStatus[
+        trio.Process
+    ] = trio.TASK_STATUS_IGNORED,
+
+    # any other `trio.run_process()` kwarg (env, shell, cwd,
+    # start_new_session, executable, ...) forwarded verbatim;
+    # our MANAGED keys (stdin/stdout/stderr/check) are set
+    # below and WIN on conflict.
+    **run_process_kwargs,
+) -> None:
+    '''
+    A `trio.Nursery.start()`-style `trio.run_process()`
+    wrapper which,
+
+    - surfaces a rc!=0 `subprocess.CalledProcessError`
+      DETERMINISTICALLY: we pass `check=False` to `trio` and
+      do our OWN post-drain rc-check, (re)building + raising a
+      BARE CPE (with a `.stderr` note) from this coro's body
+      AFTER the child exits — so there's no nursery-eg-wrapped
+      CPE to catch/`collapse_eg`, and the relay reader is never
+      race-cancelled mid-drain.
+
+    - ALWAYS isolates the parent controlling-tty
+      (`stdin=DEVNULL`, and `stdout=DEVNULL` unless
+      relayed/overridden) so a spawned program can't emit
+      terminal control-seqs onto the launching tty (which
+      would clobber its scrollback).
+
+    - optionally live-relays `stdout`/`stderr` per-line to
+      `log.<relay_level>` via concurrent reader tasks (see
+      `_relay_stream_lines`).
+
+    Delivers the live `trio.Process` via
+    `task_status.started()` then SUPERVISES it (the
+    `run_process` bg task + any relay readers) to completion
+    in this coro — i.e. the parent `tn.start()` returns
+    immediately/non-blocking.
+
+    NOTE: any crash-handling / `repl_fixture` layer is
+    intentionally NOT baked in here — compose it ON TOP at the
+    call-site, e.g.
+
+        async with maybe_open_crash_handler():
+            await tn.start(
+                partial(supervise_run_process, cmd, ...),
+            )
+
+    '''
+    emit: Callable[[str], None] = getattr(log, relay_level)
+    tag: str = (
+        label
+        or
+        (cmd if isinstance(cmd, str) else ' '.join(cmd))
+    )
+
+    # forward any extra `trio.run_process` kwargs verbatim;
+    # MANAGED keys below override on conflict.
+    rp_kwargs: dict = dict(run_process_kwargs)
+
+    # XXX ALWAYS isolate the controlling-tty's stdin.
+    rp_kwargs['stdin'] = subprocess.DEVNULL
+
+    # stdout: relay -> our own PIPE (drained by the reader
+    # below); else an explicit override; else tty-safe
+    # `DEVNULL`.
+    if relay_stdout:
+        rp_kwargs['stdout'] = subprocess.PIPE
+    elif stdout is not _UNSET:
+        rp_kwargs['stdout'] = stdout
+    else:
+        rp_kwargs['stdout'] = subprocess.DEVNULL
+
+    # stderr: PIPE (+ our reader) when we either RELAY it OR
+    # need it captured for a rc!=0 CPE note; else tty-safe
+    # `DEVNULL`. We accumulate ONLY when `check` (the note is
+    # the only consumer).
+    #
+    # XXX we ALWAYS pass `check=False` to `trio` and do our
+    # OWN deterministic post-drain rc-check (below) so `trio`
+    # never raises a nursery-eg-wrapped CPE — no `collapse_eg`
+    # workaround, no reader race-cancel.
+    want_stderr_pipe: bool = relay_stderr or check
+    stderr_accum: bytearray|None = bytearray() if check else None
+    rp_kwargs['check'] = False
+    rp_kwargs['stderr'] = (
+        subprocess.PIPE if want_stderr_pipe
+        else subprocess.DEVNULL
+    )
+
+    async with trio.open_nursery() as own_tn:
+        trio_proc: trio.Process = await own_tn.start(
+            partial(
+                trio.run_process,
+                cmd,
+                **rp_kwargs,
+            )
+        )
+
+        # spin up the concurrent pipe-drain relay reader(s) —
+        # see `_relay_stream_lines` for why these are mandatory
+        # (not cosmetic) when piping without `capture_*`.
+        if relay_stdout:
+            own_tn.start_soon(
+                partial(
+                    _relay_stream_lines,
+                    trio_proc.stdout,
+                    emit=emit,
+                    tag=f'{tag}:out',
+                )
+            )
+        if want_stderr_pipe:
+            own_tn.start_soon(
+                partial(
+                    _relay_stream_lines,
+                    trio_proc.stderr,
+                    # relay live only if asked; else silent
+                    # drain+capture for the CPE note.
+                    emit=emit if relay_stderr else None,
+                    tag=f'{tag}:err',
+                    accum=stderr_accum,
+                )
+            )
+
+        # hand the live proc up to the parent WITHOUT blocking
+        # on the bg supervise/relay tasks (keeps non-blocking
+        # `tn.start()` semantics).
+        task_status.started(trio_proc)
+
+    # ===== deterministic post-drain rc-check (BOTH paths) =====
+    # `own_tn` only unwinds once `run_process` AND the relay
+    # reader(s) have hit EOF + FULLY drained — so `stderr_accum`
+    # is COMPLETE here (no race vs an early CPE-cancel). Rebuild
+    # + raise a BARE `CalledProcessError` (the parent `tn` will
+    # eg-wrap it like any task-raise; callers `collapse_eg()` if
+    # they want it bare).
+    if (
+        check
+        and
+        trio_proc.returncode
+    ):
+        stderr_bytes: bytes = (
+            bytes(stderr_accum)
+            if stderr_accum is not None
+            else b''
+        )
+        cpe = subprocess.CalledProcessError(
+            returncode=trio_proc.returncode,
+            cmd=trio_proc.args,
+            stderr=stderr_bytes,
+        )
+        _add_stderr_note(cpe, stderr_bytes)
+        raise cpe

tractor/trionics/_taskc.py

@@ -340,7 +340,16 @@ async def start_or_cancel(
         if (
             rte.args
             and
-            'started' in rte.args[0]
+            isinstance(rte.args[0], str)
+            and
+            # match trio's *exact* "child exited without calling
+            # task_status.started()" wording — a bare `'started'`
+            # substring would also match a child task's OWN
+            # `RuntimeError(...started...)` and (under cancellation)
+            # demote it to a `Cancelled`, losing the real error. The
+            # `isinstance` guard also avoids a `TypeError` when
+            # `rte.args[0]` isn't a `str`.
+            'child exited without calling' in rte.args[0]
         ):
             # re-raises the in-flight `trio.Cancelled` IFF we're
             # under effective cancellation; else a cheap no-op and

uv.lock

@@ -797,7 +797,7 @@ dev = [
     { name = "pytest-timeout", specifier = ">=2.3" },
     { name = "stackscope", specifier = ">=0.2.2,<0.3" },
     { name = "typing-extensions", specifier = ">=4.14.1" },
-    { name = "xonsh", specifier = ">=0.23.0" },
+    { name = "xonsh", specifier = ">=0.23.8" },
 ]
 devx = [
     { name = "stackscope", specifier = ">=0.2.2,<0.3" },
@@ -809,7 +809,7 @@ repl = [
     { name = "prompt-toolkit", specifier = ">=3.0.50" },
     { name = "psutil", specifier = ">=7.0.0" },
     { name = "pyperclip", specifier = ">=1.9.0" },
-    { name = "xonsh", specifier = ">=0.23.0" },
+    { name = "xonsh", specifier = ">=0.23.8" },
 ]
 subints = [{ name = "msgspec", marker = "python_full_version >= '3.14'", specifier = ">=0.21.0" }]
 sync-pause = [{ name = "greenback", marker = "python_full_version == '3.13.*'", specifier = ">=1.2.1,<2" }]

xontrib/tractor_diag.xsh

@@ -488,6 +488,7 @@ def _tractor_reap(args):
         reap,
         reap_shm,
         reap_uds,
+        _TRACTOR_PROC_CMDLINE_MARKERS,
     )
 
     rc: int = 0
@@ -500,9 +501,8 @@ def _tractor_reap(args):
         else:
             pids = find_orphans()
             mode = (
-                'orphans (PPid==1, intrinsic '
+                f'orphans (PPid==1, intrinsic '
-                'cmdline/comm match — `tractor[…]` or '
+                f'cmdline/comm match — {_TRACTOR_PROC_CMDLINE_MARKERS}'
-                '`tractor._child`)'
             )
 
         if not pids: