Retract the hand-rolled tunnel peeler from plan-03

§3.2 specced a pure fn `_peel_tunnel_segs(proto_names) ->
(bearer_names, tunnel_specs, overlay_names)` to split a maddr at
its tunnel seg. It should never be written: `py-multiaddr` ships
that whole surface already and the plan simply missed it, even
though gh #443's 2nd bullet links the README sections in
question.

Replaced w/ a ⚠️ CORRECTION carrying the verified API table
(`.decapsulate_code(P_WG)` for the bearer, `.split()`/`.join()`
for a seg tail, `.value_for_protocol()` to read a value,
`.encapsulate()` to recompose) plus *why* it works on an infix
`/wg/` seg: the cut is by proto-code, never by matching an addr
value, and the key seg has no addr of its own.

Also,
- adopt `bearer`/`overlay` as the role names throughout, and say
  plainly why not `inner`/`outer` — the call-stack reading of
  "inner" is the exact opposite of the encapsulation one.
- warn that `value_for_protocol('ip4')` on a full tunnelled
  maddr silently yields the *bearer's* host; only call it on a
  peeled sub-maddr.
- note nesting (wg-in-wg) falls out of `.decapsulate_code()`
  cutting at the *last* occurrence, so peel repeatedly rather
  than recursing through a bespoke splitter.
- `mk_maddr()` for `TunnelledAddress` is `.encapsulate()`
  composition, not `str` building.
- README: drop the "degrades to a plain segment split" para,
  since that path is gone — no codec now means one actionable
  raise.

(this patch was generated in some part by `claude-code` using `claude-opus-5` (`anthropic`))

.claude/skills/run-tests/SKILL.md

@@ -1,632 +0,0 @@
----
-name: run-tests
-description: >
-  Run tractor test suite (or subsets). Use when the user wants
-  to run tests, verify changes, or check for regressions.
-argument-hint: "[test-path-or-pattern] [--opts]"
-allowed-tools:
-  - Bash(python -m pytest *)
-  - Bash(python -c *)
-  - Bash(python --version *)
-  - Bash(UV_PROJECT_ENVIRONMENT=py* uv run python *)
-  - Bash(UV_PROJECT_ENVIRONMENT=py* uv run pytest *)
-  - Bash(UV_PROJECT_ENVIRONMENT=py* uv sync *)
-  - Bash(UV_PROJECT_ENVIRONMENT=py* uv pip show *)
-  - Bash(git rev-parse *)
-  - Bash(ls *)
-  - Bash(cat *)
-  - Bash(jq * .pytest_cache/*)
-  # process inspection + SIGINT-first cleanup ladder (see
-  # the zombie-actor pre-flight / teardown steps below).
-  - Bash(ss *)
-  - Bash(pgrep *)
-  - Bash(pkill *)
-  - Bash(sleep *)
-  - Bash(rm -f /tmp/registry@*.sock)
-  - Read
-  - Grep
-  - Glob
-  - Task
-  - AskUserQuestion
----
-
-Run the `tractor` test suite using `pytest`. Follow this
-process:
-
-## 1. Parse user intent
-
-From the user's message and any arguments, determine:
-
-- **scope**: full suite, specific file(s), specific
-  test(s), or a keyword pattern (`-k`).
-- **transport**: which IPC transport protocol to test
-  against (default: `tcp`, also: `uds`).
-- **options**: any extra pytest flags the user wants
-  (e.g. `--ll debug`, `--tpdb`, `-x`, `-v`).
-
-If the user provides a bare path or pattern as argument,
-treat it as the test target. Examples:
-
-- `/run-tests` → full suite
-- `/run-tests test_local.py` → single file
-- `/run-tests test_registrar -v` → file + verbose
-- `/run-tests -k cancel` → keyword filter
-- `/run-tests tests/ipc/ --tpt-proto uds` → subdir + UDS
-
-## 2. Construct the pytest command
-
-Base command:
-```
-python -m pytest
-```
-
-### Default flags (always include unless user overrides):
-- `-x` (stop on first failure)
-- `--tb=short` (concise tracebacks)
-- `--no-header` (reduce noise)
-
-### Path resolution:
-- If the user gives a bare filename like `test_local.py`,
-  resolve it under `tests/`.
-- If the user gives a subdirectory like `ipc/`, resolve
-  under `tests/ipc/`.
-- Glob if needed: `tests/**/test_*<pattern>*.py`
-
-### Key pytest options for this project:
-
-| Flag | Purpose |
-|---|---|
-| `--ll <level>` | Set tractor log level (e.g. `debug`, `info`, `runtime`) |
-| `--tpdb` / `--debug-mode` | Enable tractor's multi-proc debugger |
-| `--tpt-proto <key>` | IPC transport: `tcp` (default) or `uds` |
-| `--spawn-backend <be>` | Spawn method: `trio` (default), `mp_spawn`, `mp_forkserver` |
-| `-k <expr>` | pytest keyword filter |
-| `-v` / `-vv` | Verbosity |
-| `-s` | No output capture (useful with `--tpdb`) |
-
-### Common combos:
-```sh
-# quick smoke test of core modules
-python -m pytest tests/test_local.py tests/test_rpc.py -x --tb=short --no-header
-
-# full suite, stop on first failure
-python -m pytest tests/ -x --tb=short --no-header
-
-# specific test with debug
-python -m pytest tests/discovery/test_registrar.py::test_reg_then_unreg -x -s --tpdb --ll debug
-
-# run with UDS transport
-python -m pytest tests/ -x --tb=short --no-header --tpt-proto uds
-
-# keyword filter
-python -m pytest tests/ -x --tb=short --no-header -k "cancel and not slow"
-```
-
-## 3. Pre-flight: venv detection (MANDATORY)
-
-**Always verify a `uv` venv is active before running
-`python` or `pytest`.** This project uses
-`UV_PROJECT_ENVIRONMENT=py<MINOR>` naming (e.g.
-`py313`) — never `.venv`.
-
-### Step 1: detect active venv
-
-Run this check first:
-
-```sh
-python -c "
-import sys, os
-venv = os.environ.get('VIRTUAL_ENV', '')
-prefix = sys.prefix
-print(f'VIRTUAL_ENV={venv}')
-print(f'sys.prefix={prefix}')
-print(f'executable={sys.executable}')
-"
-```
-
-### Step 2: interpret results
-
-**Case A — venv is active** (`VIRTUAL_ENV` is set
-and points to a `py<MINOR>/` dir under the project
-root or worktree):
-
-Use bare `python` / `python -m pytest` for all
-commands. This is the normal, fast path.
-
-**Case B — no venv active** (`VIRTUAL_ENV` is empty
-or `sys.prefix` points to a system Python):
-
-Use `AskUserQuestion` to ask the user:
-
-> "No uv venv is active. Should I activate one
-> via `UV_PROJECT_ENVIRONMENT=py<MINOR> uv sync`,
-> or would you prefer to activate your shell venv
-> first?"
-
-Options:
-1. **"Create/sync venv"** — run
-   `UV_PROJECT_ENVIRONMENT=py<MINOR> uv sync` where
-   `<MINOR>` is detected from `python --version`
-   (e.g. `313` for 3.13). Then use
-   `py<MINOR>/bin/python` for all subsequent
-   commands in this session.
-2. **"I'll activate it myself"** — stop and let the
-   user `source py<MINOR>/bin/activate` or similar.
-
-**Case C — inside a git worktree** (`git rev-parse
---git-common-dir` differs from `--git-dir`):
-
-Verify Python resolves from the **worktree's own
-venv**, not the main repo's:
-
-```sh
-python -c "import tractor; print(tractor.__file__)"
-```
-
-If the path points outside the worktree, create a
-worktree-local venv:
-
-```sh
-UV_PROJECT_ENVIRONMENT=py<MINOR> uv sync
-```
-
-Then use `py<MINOR>/bin/python` for all commands.
-
-**Why this matters**: without the correct venv,
-subprocesses spawned by tractor resolve modules
-from the wrong editable install, causing spurious
-`AttributeError` / `ModuleNotFoundError`.
-
-### Fallback: `uv run`
-
-If the user can't or won't activate a venv, all
-`python` and `pytest` commands can be prefixed
-with `UV_PROJECT_ENVIRONMENT=py<MINOR> uv run`:
-
-```sh
-# instead of: python -m pytest tests/ -x
-UV_PROJECT_ENVIRONMENT=py313 uv run pytest tests/ -x
-
-# instead of: python -c 'import tractor'
-UV_PROJECT_ENVIRONMENT=py313 uv run python -c 'import tractor'
-```
-
-`uv run` auto-discovers the project and venv,
-but is slower than a pre-activated venv due to
-lock-file resolution on each invocation. Prefer
-activating the venv when possible.
-
-### Step 3: import + collection checks
-
-After venv is confirmed, always run these
-(especially after refactors or module moves):
-
-```sh
-# 1. package import smoke check
-python -c 'import tractor; print(tractor)'
-
-# 2. verify all tests collect (no import errors)
-python -m pytest tests/ -x -q --co 2>&1 | tail -5
-```
-
-If either fails, fix the import error before running
-any actual tests.
-
-### Step 4: zombie-actor / stale-registry check (MANDATORY)
-
-The tractor runtime's default registry address is
-**`127.0.0.1:1616`** (TCP) / `/tmp/registry@1616.sock`
-(UDS). Whenever any prior test run — especially one
-using a fork-based backend like `subint_forkserver` —
-leaks a child actor process, that zombie keeps the
-registry port bound and **every subsequent test
-session fails to bind**, often presenting as 50+
-unrelated failures ("all tests broken"!) across
-backends.
-
-**This has to be checked before the first run AND
-after any cancelled/SIGINT'd run** — signal failures
-in the middle of a test can leave orphan children.
-
-```sh
-# 1. TCP registry — any listener on :1616? (primary signal)
-ss -tlnp 2>/dev/null | grep ':1616' || echo 'TCP :1616 free'
-
-# 2. leftover actor/forkserver procs — scoped to THIS
-#    repo's python path, so we don't false-flag legit
-#    long-running tractor-using apps (e.g. `piker`,
-#    downstream projects that embed tractor).
-pgrep -af "$(pwd)/py[0-9]*/bin/python.*_actor_child_main|subint-forkserv" \
-  | grep -v 'grep\|pgrep' \
-  || echo 'no leaked actor procs from this repo'
-
-# 3. stale UDS registry sockets
-ls -la /tmp/registry@*.sock 2>/dev/null \
-  || echo 'no leaked UDS registry sockets'
-```
-
-**Interpretation:**
-
-- **TCP :1616 free AND no stale sockets** → clean,
-  proceed. The actor-procs probe is secondary — false
-  positives are common (piker, any other tractor-
-  embedding app); only cleanup if `:1616` is bound or
-  sockets linger.
-- **TCP :1616 bound OR stale sockets present** →
-  surface PIDs + cmdlines to the user, offer cleanup:
-
-  ```sh
-  # 1. GRACEFUL FIRST (tractor is structured concurrent — it
-  #    catches SIGINT as an OS-cancel in `_trio_main` and
-  #    cascades Portal.cancel_actor via IPC to every descendant.
-  #    So always try SIGINT first with a bounded timeout; only
-  #    escalate to SIGKILL if graceful cleanup doesn't complete).
-  pkill -INT -f "$(pwd)/py[0-9]*/bin/python.*_actor_child_main|subint-forkserv"
-
-  # 2. bounded wait for graceful teardown (usually sub-second).
-  #    Loop until the processes exit, or timeout. Keep the
-  #    bound tight — hung/abrupt-killed descendants usually
-  #    hang forever, so don't wait more than a few seconds.
-  for i in $(seq 1 10); do
-    pgrep -f "$(pwd)/py[0-9]*/bin/python.*_actor_child_main|subint-forkserv" >/dev/null || break
-    sleep 0.3
-  done
-
-  # 3. ESCALATE TO SIGKILL only if graceful didn't finish.
-  if pgrep -f "$(pwd)/py[0-9]*/bin/python.*_actor_child_main|subint-forkserv" >/dev/null; then
-    echo 'graceful teardown timed out — escalating to SIGKILL'
-    pkill -9 -f "$(pwd)/py[0-9]*/bin/python.*_actor_child_main|subint-forkserv"
-  fi
-
-  # 4. if a test zombie holds :1616 specifically and doesn't
-  #    match the above pattern, find its PID the hard way:
-  ss -tlnp 2>/dev/null | grep ':1616'   # prints `users:(("<name>",pid=NNNN,...))`
-  # then (same SIGINT-first ladder):
-  # kill -INT <NNNN>; sleep 1; kill -9 <NNNN> 2>/dev/null
-
-  # 5. remove stale UDS sockets
-  rm -f /tmp/registry@*.sock
-
-  # 6. re-verify
-  ss -tlnp 2>/dev/null | grep ':1616' || echo 'TCP :1616 now free'
-  ```
-
-**Never ignore stale registry state.** If you see the
-"all tests failing" pattern — especially
-`trio.TooSlowError` / connection refused / address in
-use on many unrelated tests — check registry **before**
-spelunking into test code. The failure signature will
-be identical across backends because they're all
-fighting for the same port.
-
-**False-positive warning for step 2:** a plain
-`pgrep -af '_actor_child_main'` will also match
-legit long-running tractor-embedding apps (e.g.
-`piker` at `~/repos/piker/py*/bin/python3 -m
-tractor._child ...`). Always scope to the current
-repo's python path, or only use step 1 (`:1616`) as
-the authoritative signal.
-
-## 4. Run and report
-
-- Run the constructed command.
-- Use a timeout of **600000ms** (10min) for full suite
-  runs, **120000ms** (2min) for single-file runs.
-- If the suite is large (full `tests/`), consider running
-  in the background and checking output when done.
-- Use `--lf` (last-failed) to re-run only previously
-  failing tests when iterating on a fix.
-
-### On failure:
-- Show the failing test name(s) and short traceback.
-- If the failure looks related to recent changes, point
-  out the likely cause and suggest a fix.
-- **Check the known-flaky list** (section 8) before
-  investigating — don't waste time on pre-existing
-  timeout issues.
-- **NEVER auto-commit fixes.** If you apply a code fix
-  during test iteration, leave it unstaged. Tell the
-  user what changed and suggest they review the
-  worktree state, stage files manually, and use
-  `/commit-msg` (inline or in a separate session) to
-  generate the commit message. The human drives all
-  `git add` and `git commit` operations.
-
-### On success:
-- Report the pass/fail/skip counts concisely.
-
-## 5. Test directory layout (reference)
-
-```
-tests/
-├── conftest.py          # root fixtures, daemon, signals
-├── devx/                # debugger/tooling tests
-├── ipc/                 # transport protocol tests
-├── msg/                 # messaging layer tests
-├── discovery/           # discovery subsystem tests
-│   ├── test_multiaddr.py  # multiaddr construction
-│   └── test_registrar.py  # registry/discovery protocol
-├── test_local.py        # registrar + local actor basics
-├── test_rpc.py          # RPC error handling
-├── test_spawning.py     # subprocess spawning
-├── test_multi_program.py  # multi-process tree tests
-├── test_cancellation.py # cancellation semantics
-├── test_context_stream_semantics.py  # ctx streaming
-├── test_inter_peer_cancellation.py   # peer cancel
-├── test_infected_asyncio.py  # trio-in-asyncio
-└── ...
-```
-
-## 6. Change-type → test mapping
-
-After modifying specific modules, run the corresponding
-test subset first for fast feedback:
-
-| Changed module(s) | Run these tests first |
-|---|---|
-| `runtime/_runtime.py`, `runtime/_state.py` | `test_local.py test_rpc.py test_spawning.py test_root_runtime.py` |
-| `discovery/` (`_registry`, `_discovery`, `_addr`) | `tests/discovery/ test_multi_program.py test_local.py` |
-| `_context.py`, `_streaming.py` | `test_context_stream_semantics.py test_advanced_streaming.py` |
-| `ipc/` (`_chan`, `_server`, `_transport`) | `tests/ipc/ test_2way.py` |
-| `runtime/_portal.py`, `runtime/_rpc.py` | `test_rpc.py test_cancellation.py` |
-| `spawn/` (`_spawn`, `_entry`) | `test_spawning.py test_multi_program.py` |
-| `devx/debug/` | `tests/devx/test_debugger.py` (slow!) |
-| `to_asyncio.py` | `test_infected_asyncio.py test_root_infect_asyncio.py` |
-| `msg/` | `tests/msg/` |
-| `_exceptions.py` | `test_remote_exc_relay.py test_inter_peer_cancellation.py` |
-| `runtime/_supervise.py` | `test_cancellation.py test_spawning.py` |
-
-## 7. Quick-check shortcuts
-
-### After refactors (fastest first-pass):
-```sh
-# import + collect check
-python -c 'import tractor' && python -m pytest tests/ -x -q --co 2>&1 | tail -3
-
-# core subset (~10s)
-python -m pytest tests/test_local.py tests/test_rpc.py tests/test_spawning.py tests/discovery/test_registrar.py -x --tb=short --no-header
-```
-
-### Inspect last failures (without re-running):
-
-When the user asks "what failed?", "show failures",
-or wants to check the last-failed set before
-re-running — read the pytest cache directly. This
-is instant and avoids test collection overhead.
-
-```sh
-python -c "
-import json, pathlib, sys
-p = pathlib.Path('.pytest_cache/v/cache/lastfailed')
-if not p.exists():
-    print('No lastfailed cache found.'); sys.exit()
-data = json.loads(p.read_text())
-# filter to real test node IDs (ignore junk
-# entries that can accumulate from system paths)
-tests = sorted(k for k in data if k.startswith('tests/'))
-if not tests:
-    print('No failures recorded.')
-else:
-    print(f'{len(tests)} last-failed test(s):')
-    for t in tests:
-        print(f'  {t}')
-"
-```
-
-**Why not `--cache-show` or `--co --lf`?**
-
-- `pytest --cache-show 'cache/lastfailed'` works
-  but dumps raw dict repr including junk entries
-  (stale system paths that leak into the cache).
-- `pytest --co --lf` actually *collects* tests which
-  triggers import resolution and is slow (~0.5s+).
-  Worse, when cached node IDs don't exactly match
-  current parametrize IDs (e.g. param names changed
-  between runs), pytest falls back to collecting
-  the *entire file*, giving false positives.
-- Reading the JSON directly is instant, filterable
-  to `tests/`-prefixed entries, and shows exactly
-  what pytest recorded — no interpretation.
-
-**After inspecting**, re-run the failures:
-```sh
-python -m pytest --lf -x --tb=short --no-header
-```
-
-### Full suite in background:
-When core tests pass and you want full coverage while
-continuing other work, run in background:
-```sh
-python -m pytest tests/ -x --tb=short --no-header -q
-```
-(use `run_in_background=true` on the Bash tool)
-
-## 8. Known flaky tests
-
-These tests have **pre-existing** timing/environment
-sensitivity. If they fail with `TooSlowError` or
-pexpect `TIMEOUT`, they are almost certainly NOT caused
-by your changes — note them and move on.
-
-| Test | Typical error | Notes |
-|---|---|---|
-| `devx/test_debugger.py::test_multi_nested_subactors_error_through_nurseries` | pexpect TIMEOUT | Debugger pexpect timing |
-| `test_cancellation.py::test_cancel_via_SIGINT_other_task` | TooSlowError | Signal handling race |
-| `test_inter_peer_cancellation.py::test_peer_spawns_and_cancels_service_subactor` | TooSlowError | Async timing (both param variants) |
-| `test_docs_examples.py::test_example[we_are_processes.py]` | `assert None == 0` | `__main__` missing `__file__` in subproc |
-
-**Rule of thumb**: if a test fails with `TooSlowError`,
-`trio.TooSlowError`, or `pexpect.TIMEOUT` and you didn't
-touch the relevant code path, it's flaky — skip it.
-
-## 9. The pytest-capture hang pattern (CHECK THIS FIRST)
-
-**Symptom:** a tractor test hangs indefinitely under
-default `pytest` but passes instantly when you add
-`-s` (`--capture=no`).
-
-**Cause:** tractor subactors (especially under fork-
-based backends) inherit pytest's stdout/stderr
-capture pipes via fds 1,2. Under high-volume error
-logging (e.g. multi-level cancel cascade, nested
-`run_in_actor` failures, anything triggering
-`RemoteActorError` + `ExceptionGroup` traceback
-spew), the **64KB Linux pipe buffer fills** faster
-than pytest drains it. Subactor writes block → can't
-finish exit → parent's `waitpid`/pidfd wait blocks →
-deadlock cascades up the tree.
-
-**Pre-existing guards in the tractor harness** that
-encode this same knowledge — grep these FIRST
-before spelunking:
-
-- `tests/conftest.py:258-260` (in the `daemon`
-  fixture): `# XXX: too much logging will lock up
-  the subproc (smh)` — downgrades `trace`/`debug`
-  loglevel to `info` to prevent the hang.
-- `tests/conftest.py:316`: `# can lock up on the
-  _io.BufferedReader and hang..` — noted on the
-  `proc.stderr.read()` post-SIGINT.
-
-**Debug recipe (in priority order):**
-
-1. **Try `-s` first.** If the hang disappears with
-   `pytest -s`, you've confirmed it's capture-pipe
-   fill. Skip spelunking.
-2. **Lower the loglevel.** Default `--ll=error` on
-   this project; if you've bumped it to `debug` /
-   `info`, try dropping back. Each log level
-   multiplies pipe-pressure under fault cascades.
-3. **If you MUST use default capture + high log
-   volume**, redirect subactor stdout/stderr in the
-   child prelude (e.g.
-   `tractor.spawn._subint_forkserver._child_target`
-   post-`_close_inherited_fds`) to `/dev/null` or a
-   file.
-
-**Signature tells you it's THIS bug (vs. a real
-code hang):**
-
-- Multi-actor test under fork-based backend
-  (`subint_forkserver`, eventually `trio_proc` too
-  under enough log volume).
-- Multiple `RemoteActorError` / `ExceptionGroup`
-  tracebacks in the error path.
-- Test passes with `-s` in the 5-10s range, hangs
-  past pytest-timeout (usually 30+ s) without `-s`.
-- Subactor processes visible via `pgrep -af
-  subint-forkserv` or similar after the hang —
-  they're alive but blocked on `write()` to an
-  inherited stdout fd.
-
-**Historical reference:** this deadlock cost a
-multi-session investigation (4 genuine cascade
-fixes landed along the way) that only surfaced the
-capture-pipe issue AFTER the deeper fixes let the
-tree actually tear down enough to produce pipe-
-filling log volume. Full post-mortem in
-`ai/conc-anal/subint_forkserver_test_cancellation_leak_issue.md`.
-Lesson codified here so future-me grep-finds the
-workaround before digging.
-
-## 10. Reaping zombie subactors (`tractor-reap`)
-
-**Symptom:** after a `pytest` run crashes, times out,
-or is `Ctrl+C`'d, subactor forks (esp. under
-`subint_forkserver`) can be reparented to `init`
-(PPid==1) and linger. They hold onto ports, inherit
-pytest's capture-pipe fds, and flakify later
-sessions.
-
-**Two layers of defense:**
-
-### a) Session-scoped auto-fixture (always on)
-
-`tractor/_testing/pytest.py::_reap_orphaned_subactors`
-runs at pytest session teardown. It walks `/proc` for
-direct descendants of the pytest pid, SIGINTs them,
-waits up to 3s, then SIGKILLs survivors. SC-polite:
-gives the subactor runtime a chance to run its trio
-cancel shield + IPC teardown before escalation.
-
-This is *autouse* and session-scoped — you don't need
-to do anything. It just runs.
-
-### b) `scripts/tractor-reap` CLI (manual reap)
-
-For the **pytest-died-mid-session** case (Ctrl+C, OOM
-kill, hung process you had to `kill -9`), the fixture
-never ran. Reach for the CLI:
-
-```sh
-# default: orphans (PPid==1, cwd==repo, cmd contains python)
-scripts/tractor-reap
-
-# descendant-mode: from a still-live supervisor
-scripts/tractor-reap --parent <pytest-pid>
-
-# see what would be reaped, don't signal
-scripts/tractor-reap -n
-
-# tune the SIGINT → SIGKILL grace window
-scripts/tractor-reap --grace 5
-```
-
-Exit code: `0` if everyone exited on SIGINT, `1` if
-SIGKILL had to escalate — so you can chain it in CI
-health-checks (`scripts/tractor-reap || <alert>`).
-
-**What it matches** (orphan-mode):
-- `PPid == 1` (reparented to init → definitely
-  orphaned, not just a currently-running child)
-- `cwd == <repo-root>` (keeps the sweep scoped; won't
-  touch unrelated init-children elsewhere)
-- `python` in cmdline
-
-**What it does not do:** kill anything whose PPid is
-still a live tractor parent. If the parent is alive
-it's not an orphan; use `--parent <pid>` if you need
-to force-reap under a still-live supervisor.
-
-**When NOT to run it:** while a pytest session is
-active in another terminal. It's safe (won't touch
-that session's live children in orphan-mode) but can
-race if the target session is mid-teardown.
-
-### c) `--shm` / `--shm-only`: orphan-segment sweep
-
-Because `tractor.ipc._mp_bs.disable_mantracker()`
-turns off `mp.resource_tracker` (see
-`ai/conc-anal/subint_forkserver_mp_shared_memory_issue.md`),
-a hard-crashing actor can leave `/dev/shm/<key>`
-segments behind that nothing else GCs.
-
-```sh
-# process reap THEN shm sweep
-scripts/tractor-reap --shm
-
-# shm sweep only (skip process phase)
-scripts/tractor-reap --shm-only
-
-# dry-run: list candidates, don't unlink
-scripts/tractor-reap --shm -n
-```
-
-**Match criteria** (very conservative — this is a
-shared-system path, can't be wrong):
-- segment is a regular file under `/dev/shm`,
-- owned by the **current uid** (`stat.st_uid`),
-- AND **no live process holds it open** —
-  enumerated by walking every readable
-  `/proc/<pid>/maps` (post-mmap mappings) AND
-  `/proc/<pid>/fd/*` (pre-mmap shm-opened fds).
-
-The "nobody has it open" check is the
-kernel-canonical "is this leaked?" test — same
-answer `lsof /dev/shm/<key>` would give. No
-reliance on tractor-specific naming, so it works
-for any tractor app. Critically, it WILL NOT touch
-segments held by other apps you have running
-(e.g. `piker`, `lttng-ust-*`, `aja-shm-*` —
-verified locally with 81 in-use segments correctly
-preserved).

.claude/skills/run-tests/test-harness-reference.md

@@ -0,0 +1,255 @@
+# Tractor Test Harness Reference
+
+This repository-local file supplements the canonical [`/run-tests` skill][1]
+from the [`ai.skillz` repository][2]. Its deployer links the shared `SKILL.md`
+into both
+`.claude/skills/run-tests/` and `.opencode/skills/run-tests/` while preserving
+this project-owned reference:
+
+```text
+bash /path/to/ai.skillz/scripts/deploy.sh run-tests /path/to/tractor --provider all --method symlink
+```
+
+Keep shared environment permission, process-signal safety, target selection,
+failure inspection, and result reporting policy in the deployed `SKILL.md`.
+
+[1]: https://github.com/baudco/ai.skillz/blob/2d4896ca7e38fe2cb3090cdefc7245be4241a6d2/skills/run-tests/SKILL.md
+[2]: https://github.com/baudco/ai.skillz
+
+## Project And Environment
+
+- Project/import: `tractor`
+- Test root: `tests/`
+- Supported Python: `>=3.13,<3.15`
+- Runner: pytest `>=9.0.3`
+- Test dependencies: the `dev` group includes the `testing` group
+- CI uses uv's default `.venv`; the Nix flake uses `py313`.
+- Run from the repository root so pytest loads `pyproject.toml`.
+- Do not use `default.nix` as current test-environment authority; it still
+  selects unsupported Python 3.12.
+
+Environment directory naming is not a harness invariant. Use an already
+verified active project environment when available. Otherwise, use an
+existing uv environment without syncing it:
+
+```text
+uv run --frozen --no-sync python -c 'import pathlib, sys, tractor; root = pathlib.Path.cwd().resolve(); mod = pathlib.Path(tractor.__file__).resolve(); print(sys.executable); print(mod); assert mod.is_relative_to(root)'
+```
+
+After module moves or collection failures, check collection with:
+
+```text
+uv run --frozen --no-sync pytest --collect-only -q tests/
+```
+
+Collection is not a mandatory precursor to every narrow run. Ask before
+provisioning or changing an environment.
+
+Before trusting CLI-selected runtime settings, inspect
+`TRACTOR_SPAWN_METHOD` and `TRACTOR_LOGLEVEL`. They override the spawn method
+and runtime log level passed by callers, so report active values with test
+results rather than claiming the CLI flags alone selected the runtime.
+
+## Pytest Configuration And Commands
+
+`pyproject.toml` configures:
+
+- `testpaths = ["tests"]` and `--rootdir=./tests`;
+- importlib import mode;
+- the `tractor._testing.pytest` plugin;
+- xonsh plugin disablement;
+- `--show-capture=no` and `--capture=fd`.
+
+Do not silently add `-x`, `--tb=short`, or `--no-header`; those are not
+project defaults. In a verified active environment, replace `uv run
+--frozen --no-sync pytest` below with `python -m pytest`.
+
+```text
+# Full suite
+uv run --frozen --no-sync pytest tests/
+
+# Narrow file
+uv run --frozen --no-sync pytest tests/test_local.py
+
+# Exact node
+uv run --frozen --no-sync pytest tests/discovery/test_registrar.py::test_reg_then_unreg
+
+# Keyword selection
+uv run --frozen --no-sync pytest tests/ -k 'cancel and not slow'
+
+# Previous failures
+uv run --frozen --no-sync pytest --lf
+```
+
+After verifying that the no-sync environment is current, these pytest
+arguments match the Linux TCP CI row:
+
+```text
+CI=1 uv run --frozen --no-sync pytest tests/ -rsx --spawn-backend=trio --tpt-proto=tcp --capture=fd
+```
+
+## Plugin Options And Matrices
+
+Supported spawn backends:
+
+- `trio` (default)
+- `mp_spawn`
+- `mp_forkserver`
+
+Do not advertise `subint`, `subint_forkserver`, or
+`main_thread_forkserver` as runnable backends. Supported transports are
+`tcp` (default) and `uds`. Run one transport per pytest session.
+`mp_forkserver` and UDS are POSIX-only.
+
+Other Tractor plugin options include:
+
+- `--tpdb` / `--debug-mode`
+- `--ll` / `--loglevel`
+- `--tl` / `--tractor-loglevel`
+- `--enable-stackscope`
+
+Examples:
+
+```text
+uv run --frozen --no-sync pytest tests/ipc/ --tpt-proto=uds
+uv run --frozen --no-sync pytest tests/test_spawning.py --spawn-backend=mp_spawn
+uv run --frozen --no-sync pytest tests/test_spawning.py --spawn-backend=mp_forkserver --capture=sys
+```
+
+CI currently exercises Python 3.13 with the `trio` backend: TCP and UDS on
+Linux and macOS, plus an informational TCP row on Windows whose pytest step
+uses `continue-on-error`.
+
+## Registry And Transport Isolation
+
+Tests requesting the `reg_addr` fixture use addresses randomized per session:
+an unreserved unprivileged loopback port for TCP or a unique socket name under
+the platform runtime directory for UDS. A TCP collision remains possible.
+
+The runtime fallback remains `127.0.0.1:1616` or `registry@1616.sock`.
+Inspect that fallback only when the selected test intentionally uses runtime
+defaults or a failure identifies that address. Do not perform a mandatory
+`:1616` preflight or assume UDS sockets live under `/tmp`.
+
+## Capture And Hang Diagnosis
+
+Normal capture is `fd`. Use `--capture=sys` with `mp_forkserver`; some tests
+switch to `capsys`, but the harness does not enforce that suite-wide.
+
+For a suspected capture interaction, compare only the exact node:
+
+```text
+uv run --frozen --no-sync pytest <node> --capture=sys
+uv run --frozen --no-sync pytest <node> -s
+```
+
+Treat `-s` as a diagnostic comparison, not a pass-equivalent workaround. Do
+not use it to reinterpret an ordinary captured pass. Interactive `--tpdb` or
+`tractor.pause()` sessions are different: they require a real TTY and disabled
+capture, normally `-s`.
+
+Do not add a global pytest timeout. `fail_after_w_trace` is Trio-cooperative;
+`afk_alarm_w_trace` is a POSIX main-thread `SIGALRM` hard backstop and can
+raise asynchronously. Use the latter only as a last resort, not as a generally
+Trio-safe timeout replacement.
+
+For live task-tree diagnosis:
+
+```text
+uv run --frozen --no-sync python -c 'import stackscope'
+uv run --frozen --no-sync pytest <node> --enable-stackscope --capture=sys
+kill -USR1 <pytest-pid>
+```
+
+The import check and pytest command must use the same environment. Do not send
+SIGUSR1 if the import fails or setup warns that stackscope or SIGUSR1 is
+unavailable: without the installed handler, SIGUSR1 normally terminates the
+target process. Signal a subactor only after separately confirming that it
+installed the same handler.
+
+Stackscope appends dumps to `/tmp/tractor-stackscope-<pid>.log`, including when
+pytest capture hides terminal output. SIGUSR1 stackscope is unavailable on
+Windows and degrades to a no-op there.
+
+When a trace guard actually fires and snapshot capture succeeds, it writes
+under `$XDG_CACHE_HOME/tractor/hung-dumps/`, falling back beneath
+`~/.cache/tractor/hung-dumps/`, and prints an end-of-session index. A normal
+non-timeout run creates no snapshot.
+
+## Cleanup And `tractor-reap`
+
+On Linux, normal pytest teardown discovers surviving descendants through
+`/proc`, sends SIGINT, waits three seconds, then escalates survivors to
+SIGKILL. It does not sweep shared memory and cannot run if pytest never reaches
+fixture teardown.
+
+Process discovery is a no-op off Linux. UDS PID liveness also depends on
+`/proc`; on macOS, recognized PID-named sockets can therefore be classified as
+dead without proof. The session-scoped autouse fixture currently passes those
+candidates directly to `reap_uds()` at teardown. Do not treat its non-Linux
+classification as proof of orphanhood or run concurrent live Tractor sessions
+against the same UDS bindspace.
+
+Use the CLI in inspection-only mode first:
+
+`--shm` and `--shm-only` are Linux/FreeBSD-only and raise
+`NotImplementedError` elsewhere. On other platforms, use the UDS-only command.
+
+```text
+uv run --frozen --no-sync scripts/tractor-reap -n
+uv run --frozen --no-sync scripts/tractor-reap --parent <pytest-pid> -n
+uv run --frozen --no-sync scripts/tractor-reap --shm --uds -n
+uv run --frozen --no-sync scripts/tractor-reap --uds-only -n
+```
+
+Direct `scripts/tractor-reap` execution is acceptable only after verifying its
+`python3` shebang resolves the intended project environment.
+
+Review every candidate before requesting a mutating run:
+
+- default orphan mode is not repository-scoped;
+- `--parent` trusts the supplied PID and can include non-Tractor children;
+- `--shm` scans all current-user candidate files, not just Tractor-named
+  files;
+- `--uds` treats `registry@1616.sock` as removable even if a live default UDS
+  registrar uses it.
+
+Dry-run output prints only the initially matched root PIDs. A mutating run can
+recursively expand those roots to additional descendants when `psutil` is
+available. Inspect the descendant process tree separately; `-n` is not exact
+signal-set parity and does not by itself authorize signaling unseen children.
+
+The canonical skill owns signaling and unlinking authorization.
+
+## Test Layout And Change Mapping
+
+| Changed area | Run first |
+|---|---|
+| `tractor/runtime/_runtime.py`, `_state.py`, `tractor/_root.py` | `tests/test_local.py`, `tests/test_root_runtime.py`, `tests/test_runtime.py`, `tests/test_rpc.py` |
+| `tractor/runtime/_portal.py`, `_rpc.py` | `tests/test_rpc.py`, `tests/test_cancellation.py` |
+| `tractor/runtime/_supervise.py` | `tests/test_cancellation.py`, `tests/test_spawning.py` |
+| `tractor/discovery/` | `tests/discovery/`, `tests/test_local.py` |
+| `tractor/ipc/` | `tests/ipc/`, `tests/test_2way.py`, `tests/test_shm.py` as relevant |
+| `tractor/spawn/` | `tests/test_spawning.py`, `tests/discovery/test_multi_program.py`, `tests/test_cancellation.py` |
+| `tractor/_context.py`, `_streaming.py` | `tests/test_context_stream_semantics.py`, `tests/test_advanced_streaming.py`, `tests/test_legacy_one_way_streaming.py` |
+| `tractor/to_asyncio.py` | `tests/test_infected_asyncio.py`, `tests/test_root_infect_asyncio.py` |
+| `tractor/msg/` | `tests/msg/` |
+| `tractor/devx/` | `tests/devx/`; debugger tests use pexpect and are comparatively slow |
+| `tractor/_exceptions.py` | `tests/test_remote_exc_relay.py`, `tests/test_reg_err_types.py`, `tests/test_inter_peer_cancellation.py`, `tests/test_cancellation.py`, `tests/msg/` |
+
+Current subdirectories include `discovery/`, `ipc/`, `msg/`, `devx/`, and
+`trionics/`. There is no `tests/spawn/` directory.
+
+## Expected Outcomes
+
+Do not maintain a blanket known-flaky exemption list. Classify only current
+explicit skip or xfail marks and exact expected signatures. Notable tracked
+outcomes include:
+
+- duplicate-name `n_dups=4` and `n_dups=8` variants in
+  `tests/discovery/test_multi_program.py` are non-strict xfails;
+- `tests/test_ringbuf.py` is module-skipped;
+- some documentation examples have explicit macOS-CI skips.
+
+A generic `TooSlowError` or `pexpect.TIMEOUT` is not enough to classify a
+failure as pre-existing.

.gitignore

@@ -168,3 +168,267 @@ gh/
 
 # LLM conversations that should remain private
 docs/conversations/
+
+# BEGIN ai.skillz: direct:symlink:claude:run-tests
+/.claude/skills/run-tests/SKILL.md
+# END ai.skillz: direct:symlink:claude:run-tests
+
+# BEGIN ai.skillz: direct:symlink:opencode:run-tests
+/.opencode/skills/run-tests/SKILL.md
+# END ai.skillz: direct:symlink:opencode:run-tests
+
+# BEGIN ai.skillz: direct:symlink:opencode:command:run-tests
+/.opencode/commands/run-tests.md
+# END ai.skillz: direct:symlink:opencode:command:run-tests
+
+# BEGIN ai.skillz: direct:symlink:claude:gish
+/.claude/skills/gish
+# END ai.skillz: direct:symlink:claude:gish
+
+# BEGIN ai.skillz: direct:symlink:opencode:gish
+/.opencode/skills/gish
+# END ai.skillz: direct:symlink:opencode:gish
+
+# BEGIN ai.skillz: direct:symlink:claude:resolve-conflicts
+/.claude/skills/resolve-conflicts
+# END ai.skillz: direct:symlink:claude:resolve-conflicts
+
+# BEGIN ai.skillz: direct:symlink:opencode:resolve-conflicts
+/.opencode/skills/resolve-conflicts
+# END ai.skillz: direct:symlink:opencode:resolve-conflicts
+
+# BEGIN ai.skillz: direct:symlink:claude:git-mgmt
+/.claude/skills/git-mgmt
+# END ai.skillz: direct:symlink:claude:git-mgmt
+
+# BEGIN ai.skillz: direct:symlink:opencode:git-mgmt
+/.opencode/skills/git-mgmt
+# END ai.skillz: direct:symlink:opencode:git-mgmt
+
+# BEGIN ai.skillz: runtime:open-wkt
+/wkts/
+# END ai.skillz: runtime:open-wkt
+
+# BEGIN ai.skillz: direct:symlink:claude:open-wkt
+/.claude/skills/open-wkt
+# END ai.skillz: direct:symlink:claude:open-wkt
+
+# BEGIN ai.skillz: direct:symlink:opencode:open-wkt
+/.opencode/skills/open-wkt
+# END ai.skillz: direct:symlink:opencode:open-wkt
+
+# BEGIN ai.skillz: direct:symlink:claude:close-wkt
+/.claude/skills/close-wkt
+# END ai.skillz: direct:symlink:claude:close-wkt
+
+# BEGIN ai.skillz: direct:symlink:opencode:close-wkt
+/.opencode/skills/close-wkt
+# END ai.skillz: direct:symlink:opencode:close-wkt
+
+# BEGIN ai.skillz: runtime:code-review
+.ai/code-review/reports/
+# END ai.skillz: runtime:code-review
+
+# BEGIN ai.skillz: direct:symlink:claude:code-review
+/.claude/skills/code-review
+# END ai.skillz: direct:symlink:claude:code-review
+
+# BEGIN ai.skillz: direct:symlink:opencode:code-review
+/.opencode/skills/code-review
+# END ai.skillz: direct:symlink:opencode:code-review
+
+# BEGIN ai.skillz: direct:symlink:claude:code-nav-refs
+/.claude/skills/code-nav-refs
+# END ai.skillz: direct:symlink:claude:code-nav-refs
+
+# BEGIN ai.skillz: direct:symlink:opencode:code-nav-refs
+/.opencode/skills/code-nav-refs
+# END ai.skillz: direct:symlink:opencode:code-nav-refs
+
+# BEGIN ai.skillz: runtime:code-review-changes
+.claude/review_context.md
+.claude/review_regression.md
+.claude/review_replies/
+# END ai.skillz: runtime:code-review-changes
+
+# BEGIN ai.skillz: direct:symlink:claude:code-review-changes
+/.claude/skills/code-review-changes
+# END ai.skillz: direct:symlink:claude:code-review-changes
+
+# BEGIN ai.skillz: direct:symlink:opencode:code-review-changes
+/.opencode/skills/code-review-changes
+# END ai.skillz: direct:symlink:opencode:code-review-changes
+
+# BEGIN ai.skillz: runtime:commit-msg
+.claude/skills/commit-msg/msgs/
+.claude/git_commit_msg_LATEST.md
+# END ai.skillz: runtime:commit-msg
+
+# BEGIN ai.skillz: direct:symlink:claude:commit-msg
+/.claude/skills/commit-msg/SKILL.md
+# END ai.skillz: direct:symlink:claude:commit-msg
+
+# BEGIN ai.skillz: direct:symlink:opencode:commit-msg
+/.opencode/skills/commit-msg/SKILL.md
+# END ai.skillz: direct:symlink:opencode:commit-msg
+
+# BEGIN ai.skillz: direct:symlink:claude:commit-plan
+/.claude/skills/commit-plan
+# END ai.skillz: direct:symlink:claude:commit-plan
+
+# BEGIN ai.skillz: direct:symlink:opencode:commit-plan
+/.opencode/skills/commit-plan
+# END ai.skillz: direct:symlink:opencode:commit-plan
+
+# BEGIN ai.skillz: direct:symlink:claude:dep-supersede-scan
+/.claude/skills/dep-supersede-scan
+# END ai.skillz: direct:symlink:claude:dep-supersede-scan
+
+# BEGIN ai.skillz: direct:symlink:opencode:dep-supersede-scan
+/.opencode/skills/dep-supersede-scan
+# END ai.skillz: direct:symlink:opencode:dep-supersede-scan
+
+# BEGIN ai.skillz: direct:symlink:claude:harness-perf
+/.claude/skills/harness-perf
+# END ai.skillz: direct:symlink:claude:harness-perf
+
+# BEGIN ai.skillz: direct:symlink:opencode:harness-perf
+/.opencode/skills/harness-perf
+# END ai.skillz: direct:symlink:opencode:harness-perf
+
+# BEGIN ai.skillz: direct:symlink:claude:inter-skill-review
+/.claude/skills/inter-skill-review
+# END ai.skillz: direct:symlink:claude:inter-skill-review
+
+# BEGIN ai.skillz: direct:symlink:opencode:inter-skill-review
+/.opencode/skills/inter-skill-review
+# END ai.skillz: direct:symlink:opencode:inter-skill-review
+
+# BEGIN ai.skillz: direct:symlink:claude:opencode-cleaning
+/.claude/skills/opencode-cleaning
+# END ai.skillz: direct:symlink:claude:opencode-cleaning
+
+# BEGIN ai.skillz: direct:symlink:opencode:opencode-cleaning
+/.opencode/skills/opencode-cleaning
+# END ai.skillz: direct:symlink:opencode:opencode-cleaning
+
+# BEGIN ai.skillz: direct:symlink:claude:plan-io
+/.claude/skills/plan-io
+# END ai.skillz: direct:symlink:claude:plan-io
+
+# BEGIN ai.skillz: direct:symlink:opencode:plan-io
+/.opencode/skills/plan-io
+# END ai.skillz: direct:symlink:opencode:plan-io
+
+# BEGIN ai.skillz: runtime:pr-msg
+.claude/skills/pr-msg/msgs/
+.claude/skills/pr-msg/pr_msg_LATEST.md
+# END ai.skillz: runtime:pr-msg
+
+# BEGIN ai.skillz: direct:symlink:claude:pr-msg
+/.claude/skills/pr-msg/SKILL.md
+/.claude/skills/pr-msg/references
+/.claude/skills/pr-msg/scripts
+# END ai.skillz: direct:symlink:claude:pr-msg
+
+# BEGIN ai.skillz: direct:symlink:opencode:pr-msg
+/.opencode/skills/pr-msg/SKILL.md
+/.opencode/skills/pr-msg/references
+/.opencode/skills/pr-msg/scripts
+# END ai.skillz: direct:symlink:opencode:pr-msg
+
+# BEGIN ai.skillz: direct:symlink:claude:prompt-io
+/.claude/skills/prompt-io
+# END ai.skillz: direct:symlink:claude:prompt-io
+
+# BEGIN ai.skillz: direct:symlink:opencode:prompt-io
+/.opencode/skills/prompt-io
+# END ai.skillz: direct:symlink:opencode:prompt-io
+
+# BEGIN ai.skillz: direct:symlink:claude:py-codestyle
+/.claude/skills/py-codestyle
+# END ai.skillz: direct:symlink:claude:py-codestyle
+
+# BEGIN ai.skillz: direct:symlink:opencode:py-codestyle
+/.opencode/skills/py-codestyle
+# END ai.skillz: direct:symlink:opencode:py-codestyle
+
+# BEGIN ai.skillz: runtime:taken-export
+.ai/taken/exports/
+# END ai.skillz: runtime:taken-export
+
+# BEGIN ai.skillz: direct:symlink:claude:taken-export
+/.claude/skills/taken-export
+# END ai.skillz: direct:symlink:claude:taken-export
+
+# BEGIN ai.skillz: direct:symlink:opencode:taken-export
+/.opencode/skills/taken-export
+# END ai.skillz: direct:symlink:opencode:taken-export
+
+# BEGIN ai.skillz: direct:symlink:claude:yt-url-lookup
+/.claude/skills/yt-url-lookup
+# END ai.skillz: direct:symlink:claude:yt-url-lookup
+
+# BEGIN ai.skillz: direct:symlink:opencode:yt-url-lookup
+/.opencode/skills/yt-url-lookup
+# END ai.skillz: direct:symlink:opencode:yt-url-lookup
+
+# BEGIN ai.skillz: direct:symlink:opencode:command:gish
+/.opencode/commands/gish.md
+# END ai.skillz: direct:symlink:opencode:command:gish
+
+# BEGIN ai.skillz: direct:symlink:opencode:command:resolve-conflicts
+/.opencode/commands/resolve-conflicts.md
+# END ai.skillz: direct:symlink:opencode:command:resolve-conflicts
+
+# BEGIN ai.skillz: direct:symlink:opencode:command:git-mgmt
+/.opencode/commands/git-mgmt.md
+# END ai.skillz: direct:symlink:opencode:command:git-mgmt
+
+# BEGIN ai.skillz: direct:symlink:opencode:command:open-wkt
+/.opencode/commands/open-wkt.md
+# END ai.skillz: direct:symlink:opencode:command:open-wkt
+
+# BEGIN ai.skillz: direct:symlink:opencode:command:close-wkt
+/.opencode/commands/close-wkt.md
+# END ai.skillz: direct:symlink:opencode:command:close-wkt
+
+# BEGIN ai.skillz: direct:symlink:opencode:command:code-review
+/.opencode/commands/code-review.md
+# END ai.skillz: direct:symlink:opencode:command:code-review
+
+# BEGIN ai.skillz: direct:symlink:opencode:command:code-review-changes
+/.opencode/commands/code-review-changes.md
+# END ai.skillz: direct:symlink:opencode:command:code-review-changes
+
+# BEGIN ai.skillz: direct:symlink:opencode:command:commit-msg
+/.opencode/commands/commit-msg.md
+# END ai.skillz: direct:symlink:opencode:command:commit-msg
+
+# BEGIN ai.skillz: direct:symlink:opencode:command:commit-plan
+/.opencode/commands/commit-plan.md
+# END ai.skillz: direct:symlink:opencode:command:commit-plan
+
+# BEGIN ai.skillz: direct:symlink:opencode:command:dep-supersede-scan
+/.opencode/commands/dep-supersede-scan.md
+# END ai.skillz: direct:symlink:opencode:command:dep-supersede-scan
+
+# BEGIN ai.skillz: direct:symlink:opencode:command:harness-perf
+/.opencode/commands/harness-perf.md
+# END ai.skillz: direct:symlink:opencode:command:harness-perf
+
+# BEGIN ai.skillz: direct:symlink:opencode:command:opencode-cleaning
+/.opencode/commands/opencode-cleaning.md
+# END ai.skillz: direct:symlink:opencode:command:opencode-cleaning
+
+# BEGIN ai.skillz: direct:symlink:opencode:command:pr-msg
+/.opencode/commands/pr-msg.md
+# END ai.skillz: direct:symlink:opencode:command:pr-msg
+
+# BEGIN ai.skillz: direct:symlink:opencode:command:taken-export
+/.opencode/commands/taken-export.md
+# END ai.skillz: direct:symlink:opencode:command:taken-export
+
+# BEGIN ai.skillz: direct:symlink:opencode:command:yt-url-lookup
+/.opencode/commands/yt-url-lookup.md
+# END ai.skillz: direct:symlink:opencode:command:yt-url-lookup

ai/prompt-io/claude/20260813T001102Z_27c34aeb_prompt_io.md

@@ -0,0 +1,136 @@
+---
+model: claude-opus-5
+service: claude
+session: 7b9c97c4-fff7-4ac4-97fb-35720453308e
+timestamp: 2026-08-13T00:11:02Z
+git_ref: 27c34aeb
+scope: docs+code
+substantive: true
+raw_file: 20260813T001102Z_27c34aeb_prompt_io.raw.md
+---
+
+## Prompt
+
+> draft hyper detailed implementation plans for [three]
+> prospective new transport (tpt) backends for tractor's `.ipc`
+> layer, from four GitHub issues: TIPC (gh #378) using built-in
+> linux socket API w/ `trio` interfacing, leveraging TIPC's
+> built-in discovery machinery; QUIC (gh #353) using the `iroh`
+> lib, ideally with the py asyncio support (via ffi) rewritten
+> for trio; wg (gh #482 and/or #443) with other shuttle-able
+> tpts, using `pyroute2`, as much trio wrapping as possible
+> where any other async support can be replaced.
+
+With constraints: "be only slightly speculative"; realistic given
+the linked info; the plans must be workable "by another
+model/provider without much discrepancy on the design or lib
+selections needed"; do it all in a new `/open-wkt` off main; and
+be "as modular and modern-async-python-as-nearly-functional-style
+as possible."
+
+Key follow-up prompts that changed the design mid-session:
+
+> wg requires binding a socket on both ends of the tunnel, the
+> new support we added to `py-multiaddr` in their #108 enables
+> declaring this in the maddr? […] so the
+> `/ip4/10.0.0.1/udp/666/ip4/192.168.0.13/tcp/80` bullet example
+> from #443 isn't correct for a wg tunnel maddr and would
+> actually end up looking like,
+> `/ip4/10.0.0.1/udp/666/wg/<pub-key>/ip4/192.168.0.13/tcp/80` no?
+
+> right, we're likely going to need less `tuple`-formed
+> `UnwrappedAddress` usage to make them all work and/or to extend
+> the unwrapped form with a proto-str, likely the key from
+> `multiaddrs` no?
+
+> better encapsulating `UnwrappedAddress` for use externally is
+> really way overdue; we should likely be encouraging (if not
+> preventing unwrapped) wrapped addrs always much like other
+> (std) libs, `ipaddress` comes to mind.
+
+> so iirc netns mgmt becomes a lower-level runtime config-api
+> rather then an actor-app-code API which can be dynamically
+> changed once the actor is up ya?
+
+## Response summary
+
+A research-and-planning branch, `ng_tpts_planning` — **no runtime
+code touched**. Five plan docs under `ai/tpt-backends/` plus a
+corrected `wg` example set under `examples/multihost/wg_lan/`,
+across 9 commits.
+
+Three design conclusions came out of the planning that outlive
+the branch:
+
+1. **`wg` is a bindspace, not a `MsgTransport`.** The `/wg/`
+   maddr segment is *infix*: bearer (underlay ip+udp
+   `ListenPort`, bound by the kernel) → `/wg/u<key>` (an
+   identity, bound by nothing) → overlay (the only part tractor
+   binds). Verified empirically by installing
+   `baudco/py-multiaddr@wg_support` (py-multiaddr#108) in a
+   throwaway venv and round-tripping each candidate form; gh
+   #482's original suffix form parses but is semantically
+   inverted.
+2. **Proto-key the `UnwrappedAddress` form**, spelled with the
+   `multiaddr` protocol names, dispatched via
+   `_address_types[addr[0]]`. Kills a whole collision class
+   (TIPC's `(str, int)` ≡ `TCPAddress`; iroh's `(str, str)`
+   swallowed by the UDS case) and is the recommended migration
+   *before* any new backend lands.
+3. **netns is a runtime/boot-time config API, not an app-code
+   one** — `setns(2)` is per-thread and won't move
+   already-created sockets, so there is deliberately no
+   `await actor.enter_netns(...)`.
+
+Also verified that `trio.SocketStream`/`SocketListener` are
+address-family agnostic (no `AF_*` check anywhere), which is what
+makes TIPC the cheapest of the three backends to add.
+
+Four related issues were annotated with the results (#378, #353,
+#482, #443); #443's body was rewritten to reflect the corrected
+grammar, with no existing checkbox state changed.
+
+## Files changed
+
+- `ai/tpt-backends/00_shared_backend_contract.md` — normative
+  backend duck-type contract, registration checklist, §1.1
+  proto-key conclusion
+- `ai/tpt-backends/01_tipc_backend.md` — TIPC plan; service
+  addressing, `TIPC_TOP_SRV` push registry, instance-collision
+  hazard, step-0 probe
+- `ai/tpt-backends/02_quic_iroh_backend.md` — `iroh` plan;
+  `uniffi`→`trio` bridge, listener/stream adapters, API-truth
+  table
+- `ai/tpt-backends/03_wg_tunnel_bindspace.md` — `wg`-as-bindspace
+  plan; verified maddr grammar, 3-owner split, netns reality
+- `ai/tpt-backends/README.md` — index
+- `examples/multihost/wg_lan/wg_maddr.py` — frozen `msgspec`
+  tunnelled addr + pure parse/render helpers; impure
+  `verify_wg_peer()` kept separate
+- `examples/multihost/wg_lan/host_a_srv.py` — host-A actor tree
+- `examples/multihost/wg_lan/host_b_client.py` — host-B dialer
+- `examples/multihost/wg_lan/README.md` — grammar, owner table,
+  setup, "what changed vs #482"
+
+## Human edits
+
+Substantial human steering rather than post-hoc editing; the
+corrections were applied by the model in-session after being
+challenged:
+
+- rejected an initial claim that `wg` has "nothing to bind at the
+  tunnel layer" and supplied the correct composed maddr form,
+  which forced a rewrite of plan 03 §3.2 and a retraction in the
+  already-posted #443 comment
+- rejected a supporting claim that `/ip4/../udp/443/quic-v1` was
+  "also composed"
+- directed the proto-key/`ipaddress`-discipline conclusion and
+  the netns-as-runtime-config framing, both of which were then
+  folded back into the docs
+- chose the commit boundaries and authored all commits; ran every
+  `git` mutation (commit, rebase, push) themselves
+
+One model-initiated correction pre-publication: a self-review
+downgraded two overconfident claims (the `uniffi`/asyncio thesis
+and TIPC duplicate-binder behaviour) to explicitly-flagged
+assumptions before the #353/#378 comments were posted.

ai/prompt-io/claude/20260813T001102Z_27c34aeb_prompt_io.raw.md

@@ -0,0 +1,165 @@
+---
+model: claude-opus-5
+service: claude
+timestamp: 2026-08-13T00:11:02Z
+git_ref: 27c34aeb
+diff_cmd: git diff main..ng_tpts_planning
+---
+
+# Raw output — next-gen tpt-backend implementation plans
+
+## Generated planning docs
+
+> `git diff main..ng_tpts_planning -- ai/tpt-backends/`
+
+Five markdown docs. `00_shared_backend_contract.md` is normative
+and the other three are written against it so they can be worked
+independently:
+
+- **`00_shared_backend_contract.md`** — the backend duck-type
+  (`<Proto>Address(msgspec.Struct, frozen=True)` + module-level
+  `start_listener()`/`close_listener()` + a
+  `Msgpack<Proto>Stream(MsgpackTransport)`), the
+  `inspect.getmodule(self.addr)` reflection in
+  `Endpoint.start_listener()` that forces the Address class and
+  its listener fns to share a module, a 10-item registration
+  checklist, the dep policy, the test-harness shape, and §1.1's
+  proto-key conclusion (below).
+- **`01_tipc_backend.md`** — service addressing via
+  `TIPC_ADDR_NAMESEQ` (bind/publish) and `TIPC_ADDR_NAME`
+  (connect/lookup), `TIPC_TOP_SRV` topology subscriptions as a
+  push-based registry, the `get_random()` instance-collision
+  hazard, and a step-0 capability-probe spike.
+- **`02_quic_iroh_backend.md`** — `iroh` over
+  `aioquic`/`quiche`/`trio-asyncio`, a `_uniffi_trio.py` bridge
+  built on `TrioToken.run_sync_soon()`, `trio.abc.Listener`/
+  `HalfCloseableStream` adapters, and an API-truth table to fill
+  in during step 0.
+- **`03_wg_tunnel_bindspace.md`** — `wg` as a *bindspace* rather
+  than a `MsgTransport`, a `TunnelledAddress` wrapper delegating
+  `.proto_key`/`.unwrap()` to `.inner`, `pyroute2` for layer B,
+  and `@acm`-managed netns/iface for layer C.
+- **`README.md`** — index.
+
+## Generated example code
+
+> `git diff main..ng_tpts_planning -- examples/multihost/wg_lan/`
+
+- `wg_maddr.py` — `WGTunnelledAddr(msgspec.Struct, frozen=True)`
+  carrying `bearer: tuple[str, int]`, `peer_pubkey: str`,
+  `inner: tuple[str, int]`, `inner_proto: Literal['tcp']`, plus a
+  `.maddr` property that re-renders the canonical form. Pure
+  helpers `mb_pubkey()`, `wg8_pubkey()`, `parse_wg_maddr()`, and
+  `_segments()` (with a marked stopgap for when the `wg` codec
+  isn't installed). `verify_wg_peer()` is impure **by design** and
+  kept out of the parse path.
+- `host_a_srv.py` / `host_b_client.py` — the two-host runs; both
+  pass only `addr.inner` to `open_nursery()`/`open_root_actor()`.
+- `README.md` — grammar, owner table, `#108`-branch install line,
+  tunnel setup, "what changed vs #482".
+
+## Verified findings (non-code, verbatim)
+
+### `trio` is address-family agnostic
+
+Read against the installed `trio`. `SocketStream`/`SocketListener`
+ctor checks are only "is a trio sock object" + `type ==
+SOCK_STREAM`, plus an `OSError`-**suppressed** `SO_ACCEPTCONN`
+probe. No `AF_*` check anywhere; `TCP_NODELAY`/`TCP_NOTSENT_LOWAT`
+are set under `suppress(OSError)`. A TIPC `SOCK_STREAM` sock should
+therefore drop straight into `trio.serve_listeners()` with the
+existing `MsgpackTransport` framing, making TIPC mostly
+table-registration boilerplate w/ zero new deps.
+
+### the `wg` maddr grammar — `/wg/` is infix, not suffix
+
+Installed `baudco/py-multiaddr@wg_support` (PR
+multiformats/py-multiaddr#108) into a throwaway venv and
+round-tripped every candidate form:
+
+| maddr | `[p.name for p in m.protocols()]` |
+| --- | --- |
+| `/ip4/1.2.3.4/udp/51820/wg/u<k>` | `['ip4','udp','wg']` |
+| `/ip4/../udp/../wg/u<k>/ip4/../tcp/..` | `['ip4','udp','wg','ip4','tcp']` |
+| `/ip4/10.0.11.1/tcp/1616/wg/u<k>` | `['ip4','tcp','wg']` |
+
+```
+/ip4/192.168.1.50/udp/51820/wg/u<A_pub>/ip4/10.0.11.1/tcp/1616
+\_______ bearer __________/\__ key __/\______ overlay ______/
+```
+
+Segments *before* `/wg/` are the bearer — the underlay
+`(ip, udp-port)` that `wg(8)` itself listens on (`ListenPort`).
+Segments *after* are the overlay endpoint, the only part tractor
+binds. The third row above is #482's original suffix form: it
+parses, but is semantically inverted.
+
+Three parts, three owners — and only one is an `Endpoint`:
+
+| part | bound by | in the runtime? |
+| --- | --- | --- |
+| bearer | kernel, via `wg-quick`/`pyroute2` | no |
+| `/wg/u<key>` | nothing — an identity | no, verified out-of-band |
+| overlay | `tractor`'s `IPCServer` | yes, as `.inner` |
+
+### proto-key-tagged `UnwrappedAddress`
+
+Shape-matching in `wrap_address()` does not survive four backends.
+TIPC's natural unwrapped form is a `(str, int)`, indistinguishable
+from `TCPAddress`; iroh's is a `(str, str)`, already swallowed by
+the existing UDS case (`case (_, filename) if type(filename) is
+str`). Ordering hacks and prefix-tagging only paper over it.
+
+Recommended prerequisite for all three backends: carry an explicit
+proto-key spelled with the `multiaddr` protocol name —
+`('tcp', host, port)`, `('unix', path)`,
+`('tipc', stype, inst, scope)` — so `wrap_address()` collapses to
+`_address_types[addr[0]]` and the collision class stops existing.
+This also makes the on-wire form agree with
+`mk_maddr()`/`parse_maddr()` instead of being an independent
+invention. It is a wire-format change (`SpawnSpec`,
+`_root_mailbox`, `_registry_addrs`) plus every fixture and
+downstream config, so it wants its own migration commit landed
+before any new backend — and it is the moment to stop handing raw
+tuples to users at all, making `Address` the public currency and
+`UnwrappedAddress` an internal serialization detail (the
+discipline `ipaddress` uses).
+
+### netns is a runtime-level config API
+
+`setns(2)` affects the calling thread only and does not move
+already-created sockets. So a netns is a spawn/boot-time input
+alongside `enable_transports`/`tpt_bind_addrs`, and there is
+deliberately no `await actor.enter_netns(...)` — a mid-life API
+would silently leave the IPC server bound in the old namespace.
+Corollary for layer B: pass `netns=` down to `pyroute2` rather
+than assuming a `trio.to_thread` worker inherits it.
+
+### `examples/` collection would have failed CI
+
+`tests/test_docs_examples.py` walks `examples/` recursively and
+subproc-runs every collected file asserting `rc == 0`. Its filter
+never checks the extension, so all four `wg_lan` files were
+collected — including `README.md`, which would have been run as
+`python README.md`. `'multihost' not in p[0]` was already in the
+exclusion list with no directory using it. Moving the set under
+`examples/multihost/wg_lan/` drops collection 24 → 20 with zero
+test changes; confirmed via `pytest --collect-only`.
+
+## Corrections applied during the session
+
+The human corrected two claims that had been asserted without
+verification, both since retracted in-place in the docs and in the
+posted issue comments:
+
+1. that `wg` has "nothing to bind at the tunnel layer, exactly one
+   bind" — wrong; a wg stack is genuinely composed, and the real
+   axis is *who owns* each layer's endpoint.
+2. that `/ip4/../udp/443/quic-v1` was "also composed" — wrong;
+   that is one endpoint with a protocol qualifier, not a tunnel.
+
+A self-review before publication also downgraded two
+overconfident claims to explicitly-flagged assumptions: the
+`uniffi`-uses-asyncio-only-as-executor thesis (contradicted that
+plan's own "do not guess from memory" step 0) and TIPC's
+duplicate-binder round-robin behaviour (unverified).

ai/prompt-io/opencode/20260811T233833Z_7cbd64ee_prompt_io.md

@@ -0,0 +1,30 @@
+---
+model: openai/gpt-5.6-sol
+service: opencode
+session: ses_0799212ebffe42arY96czXn89F
+timestamp: 2026-08-11T23:38:33Z
+git_ref: 7cbd64ee
+scope: code
+substantive: true
+raw_file: 20260811T233833Z_7cbd64ee_prompt_io.raw.md
+---
+
+## Prompt
+
+Open a new isolated worktree in the local Tractor repository and draft a fix
+for `BroadcastReceiver` reporting that a lagged one-slot consumer dropped
+zero values when one value had actually been displaced.
+
+## Response summary
+
+Corrected the off-by-one lag count while preserving cursor recovery and added
+deterministic narrow- and wider-window regressions for exact loss reporting.
+
+## Files changed
+
+- `tractor/trionics/_broadcast.py` - exact broadcast overrun count.
+- `tests/test_task_broadcasting.py` - lag count and recovery regression.
+
+## Human edits
+
+None - generated output follows the user's diagnosed edge case.

ai/prompt-io/opencode/20260811T233833Z_7cbd64ee_prompt_io.raw.md

@@ -0,0 +1,50 @@
+---
+model: openai/gpt-5.6-sol
+service: opencode
+timestamp: 2026-08-11T23:38:33Z
+git_ref: 7cbd64ee
+diff_cmd: git diff HEAD~1..HEAD
+---
+
+The user asked for a Tractor fix in a new isolated worktree after a live piker
+failure reported:
+
+```text
+tractor.trionics._broadcast.Lagged:
+Task `piker.brokers.ib.broker.handle_order_requests` overrun and
+dropped `0` values
+```
+
+Inspection showed the lag exception was valid but its count was off by one.
+`BroadcastReceiver.receive_nowait()` treats `seq` as a deque index. With a
+one-entry queue, index zero is the only retained value and `seq == 1` already
+means one value was displaced. The old `seq - maxlen` calculation therefore
+reported zero instead of one.
+
+> `git diff HEAD~1..HEAD -- tractor/trionics/_broadcast.py`
+
+Adjusted the lag count to `seq - maxlen + 1` and documented why the first
+invalid deque index must be included. The existing Tokio-style cursor reset
+remains unchanged.
+
+> `git diff HEAD~1..HEAD -- tests/test_task_broadcasting.py`
+
+Added a deterministic parameterized regression covering a one-slot queue
+with one dropped value and a three-slot queue with two dropped values. The
+test keeps the root receiver idle while a child subscriber synchronously
+drains each produced value, asserts the exact `Lagged` message, and proves the
+next receive resumes at the oldest retained item.
+
+Verification output:
+
+```text
+..                                                                       [100%]
+2 passed in 0.04s
+
+..........                                                               [100%]
+10 passed in 4.65s
+```
+
+The targeted import resolved to the new Tractor worktree. Python compilation
+and `git diff --check` passed. Adversarial review found no actionable issues;
+zero-capacity channels remain a pre-existing untested edge outside this fix.

ai/prompt-io/opencode/20260812T012324Z_06c4af17_prompt_io.md

@@ -0,0 +1,31 @@
+---
+model: openai/gpt-5.6-sol
+service: opencode
+session: ses_0799212ebffe42arY96czXn89F
+timestamp: 2026-08-12T01:23:24Z
+git_ref: 06c4af17
+scope: code
+substantive: true
+raw_file: 20260812T012324Z_06c4af17_prompt_io.raw.md
+---
+
+## Prompt
+
+Iteratively refine Tractor PR 490. For item one, correct broadcast statistics
+queue counts and Trio event checks, verify and review the exact change, then
+return a complete commit plan before proceeding.
+
+## Response summary
+
+Converted subscriber cursor indexes into clamped retained queue counts,
+removed deprecated event truthiness, and added deterministic state and
+deprecation regressions.
+
+## Files changed
+
+- `tractor/trionics/_broadcast.py` - accurate queue and waiter statistics.
+- `tests/test_task_broadcasting.py` - retained-count and event regression.
+
+## Human edits
+
+None - generated output follows the requested first iterative item.

ai/prompt-io/opencode/20260812T012324Z_06c4af17_prompt_io.raw.md

@@ -0,0 +1,37 @@
+---
+model: openai/gpt-5.6-sol
+service: opencode
+timestamp: 2026-08-12T01:23:24Z
+git_ref: 06c4af17
+diff_cmd: git diff HEAD~1..HEAD
+---
+
+After opening draft Tractor PR 490, the user requested an iterative pass over
+additional broadcast subsystem findings. The first item was to correct
+`BroadcastState.statistics()` queued counts and its deprecated Trio event
+truthiness check, then stop for a complete commit plan.
+
+> `git diff HEAD~1..HEAD -- tractor/trionics/_broadcast.py`
+
+Changed `queued_len_by_task` from raw deque cursor indexes to retained,
+receivable counts. Caught-up `-1` reports zero, valid indexes report index plus
+one, and lagged cursors clamp to the current retained queue length. Replaced
+`trio.Event` truthiness with an explicit `is not None` branch.
+
+> `git diff HEAD~1..HEAD -- tests/test_task_broadcasting.py`
+
+Added a deterministic statistics regression using actual sends and receives.
+It verifies caught-up and one-queued states, drives a root receiver beyond a
+three-slot retention window to prove clamping, and installs a real
+`trio.Event` while treating deprecations as errors.
+
+Verification output:
+
+```text
+...........                                                              [100%]
+11 passed in 5.76s
+```
+
+Python compilation and `git diff --check` passed. Initial adversarial review
+caught unclamped lagged cursors and an ineffective event test; both were
+fixed. Final review found no actionable issues.

ai/prompt-io/opencode/20260812T030608Z_1095e7f7_prompt_io.md

@@ -0,0 +1,33 @@
+---
+model: openai/gpt-5.6-sol
+service: opencode
+session: ses_0799212ebffe42arY96czXn89F
+timestamp: 2026-08-12T03:06:08Z
+git_ref: 1095e7f7
+scope: code
+substantive: true
+raw_file: 20260812T030608Z_1095e7f7_prompt_io.raw.md
+---
+
+## Prompt
+
+For Tractor PR 490 item two, make shared underlying receive failures wake and
+terminate every broadcast subscriber without losing retained values. Review,
+verify and return a complete commit plan before continuing.
+
+## Response summary
+
+Published ordinary receive failures as terminal broadcast state, introduced
+a public chained peer exception, kept control-flow exits transient while
+waking peers, documented the contract, and added deterministic regressions.
+
+## Files changed
+
+- `tractor/trionics/_broadcast.py` - terminal failure and peer wake protocol.
+- `tractor/trionics/__init__.py` - public peer exception export.
+- `docs/api/trionics.rst` - failure-delivery API contract.
+- `tests/test_task_broadcasting.py` - terminal and transient failure tests.
+
+## Human edits
+
+None - generated output follows the requested second iterative item.

ai/prompt-io/opencode/20260812T030608Z_1095e7f7_prompt_io.raw.md

@@ -0,0 +1,49 @@
+---
+model: openai/gpt-5.6-sol
+service: opencode
+timestamp: 2026-08-12T03:06:08Z
+git_ref: 1095e7f7
+diff_cmd: git diff HEAD~1..HEAD
+---
+
+The user requested the second iterative refinement for Tractor PR 490: ensure
+non-EOC failures from a shared underlying broadcast receiver do not leave peer
+subscribers blocked forever, then stop for a complete commit plan.
+
+> `git diff HEAD~1..HEAD -- tractor/trionics/_broadcast.py`
+
+Added shared terminal failure publication for ordinary `Exception` values.
+The receive owner gets the original exception; peers may drain retained
+values and then get a fresh `BroadcastReceiveError` chained from the original.
+Late subscribers observe the same terminal state without retrying the failed
+underlying receiver. Process-control and cancellation-like `BaseException`
+values wake peers but are re-raised without becoming durable channel state.
+
+> `git diff HEAD~1..HEAD -- tractor/trionics/__init__.py`
+
+Exported `BroadcastReceiveError` as the public peer-delivery exception.
+
+> `git diff HEAD~1..HEAD -- docs/api/trionics.rst`
+
+Documented `BroadcastReceiveError` and the owner-versus-peer delivery
+contract, including retained-value draining and late subscribers.
+
+> `git diff HEAD~1..HEAD -- tests/test_task_broadcasting.py`
+
+Added deterministic bounded regressions. One scripts a successful receive
+followed by `RuntimeError`, proving the root drains retained data, all current
+and late receivers observe terminal failure, and the source is not retried.
+The second scripts a custom `BaseException`, proving peers wake and take over
+the next source receive without retaining control-flow state.
+
+Verification output:
+
+```text
+.............                                                            [100%]
+13 passed in 5.62s
+```
+
+Python compilation and `git diff --check` passed. Iterative adversarial review
+drove independent peer exception wrappers, ordinary-versus-control-flow
+classification, bounded test completion, public docs, and the final catch-all
+peer wake. Final review found no issues.

ai/prompt-io/opencode/20260812T150027Z_c2a6ccef_prompt_io.md

@@ -0,0 +1,33 @@
+---
+model: openai/gpt-5.6-sol
+service: opencode
+session: ses_0799212ebffe42arY96czXn89F
+timestamp: 2026-08-12T15:00:27Z
+git_ref: c2a6ccef
+scope: code
+substantive: true
+raw_file: 20260812T150027Z_c2a6ccef_prompt_io.raw.md
+---
+
+## Prompt
+
+For Tractor PR 490 item three, prevent subscriber closure from waking another
+receiver's shared event or stranding peers. Preserve close-safe source-read
+ownership handoff, then review, verify and return a complete commit plan.
+
+## Response summary
+
+Introduced receiver-local wait/source cancellation, owner-specific handoff,
+and close precedence over shielded source values/errors/EOC. Clarified that
+private scope cancellation means explicit close while outer task cancellation
+remains `trio.Cancelled` for both source-read and peer-wait scopes, with
+deterministic receiver-close regressions.
+
+## Files changed
+
+- `tractor/trionics/_broadcast.py` - receiver-local close and ownership scopes.
+- `tests/test_task_broadcasting.py` - peer close and owner handoff regressions.
+
+## Human edits
+
+None - generated output follows the requested third iterative item.

ai/prompt-io/opencode/20260812T150027Z_c2a6ccef_prompt_io.raw.md

@@ -0,0 +1,51 @@
+---
+model: openai/gpt-5.6-sol
+service: opencode
+timestamp: 2026-08-12T15:00:27Z
+git_ref: c2a6ccef
+diff_cmd: git diff HEAD~1..HEAD
+---
+
+The user requested the third iterative refinement for Tractor PR 490: closing
+one broadcast subscriber must not set another receiver owner's shared event
+and create a runnable hot loop, then stop for a complete commit plan.
+
+> `git diff HEAD~1..HEAD -- tractor/trionics/_broadcast.py`
+
+Added receiver-local wait cancellation and source-read ownership scopes.
+Closing a non-owner waiting behind another source reader cancels only that
+receiver's private wait and maps it to `ClosedResourceError`; the shared event
+remains untouched. Closing the active source owner cancels only its private
+source-read scope, wakes peers after cleanup, and lets one peer take ownership.
+
+Source outcomes are captured inside the owner scope and classified only after
+checking close/cancel state. A cancellation-shielding source therefore cannot
+publish a returned value, ordinary error, or EOC after its owner was closed.
+The private scope's `cancel_called` bit is asserted to imply receiver closure;
+outer task cancellation remains `trio.Cancelled` and is not translated into
+`ClosedResourceError`. Owner-key comments document that only the receiver
+identified by `recv_ready[0]` may cancel the shared source-read scope. The
+same explicit-close invariant is enforced symmetrically for private peer-wait
+scope cancellation.
+
+> `git diff HEAD~1..HEAD -- tests/test_task_broadcasting.py`
+
+Added deterministic bounded regressions for both close positions. The
+non-owner test places two peers behind an active source read, closes one and
+proves only that peer exits while the shared event stays unset. The owner test
+closes a source owner whose receive shields cancellation and parameterizes a
+returned value, `RuntimeError`, and `EndOfChannel`; each discarded outcome
+hands the next source receive to the waiting root without terminal-state or
+EOC publication.
+
+Verification output:
+
+```text
+.................                                                        [100%]
+17 passed in 5.84s
+```
+
+Python compilation and `git diff --check` passed. Iterative adversarial review
+caught a waiting non-owner hang, cancellation-shielded source returns, and
+shielded source exceptions. All were fixed. Final review found no actionable
+issues.

ai/prompt-io/opencode/20260812T213117Z_51185487_prompt_io.md

@@ -0,0 +1,34 @@
+---
+model: openai/gpt-5.6-sol
+service: opencode
+session: ses_0799212ebffe42arY96czXn89F
+timestamp: 2026-08-12T21:31:17Z
+git_ref: 51185487
+scope: code
+substantive: true
+raw_file: 20260812T213117Z_51185487_prompt_io.raw.md
+---
+
+## Prompt
+
+For Tractor PR 490 item four, expose `raise_on_lag` through the public IPC and
+asyncio linked-channel subscription wrappers. Review, verify and return a
+complete commit plan before applying the API downstream in piker.
+
+## Response summary
+
+Added public per-subscription lag policy to both wrappers, preserved first-call
+root policy, documented the semantics, and covered forwarding plus real fan-out
+paths.
+
+## Files changed
+
+- `tractor/_streaming.py` - `MsgStream` lag policy forwarding.
+- `tractor/to_asyncio.py` - linked-channel lag policy forwarding.
+- `docs/guide/streaming.rst` - IPC fan-out policy docs.
+- `docs/guide/asyncio.rst` - linked-channel fan-out policy docs.
+- `tests/test_task_broadcasting.py` - wrapper policy regression.
+
+## Human edits
+
+None - generated output follows the requested fourth iterative item.

ai/prompt-io/opencode/20260812T213117Z_51185487_prompt_io.raw.md

@@ -0,0 +1,53 @@
+---
+model: openai/gpt-5.6-sol
+service: opencode
+timestamp: 2026-08-12T21:31:17Z
+git_ref: 51185487
+diff_cmd: git diff HEAD~1..HEAD
+---
+
+The user requested the fourth iterative refinement for Tractor PR 490: expose
+subscriber lag policy through the public `MsgStream.subscribe()` and
+`LinkedTaskChannel.subscribe()` wrappers, then stop for a complete commit
+plan. This enables piker to replace private receiver mutation.
+
+> `git diff HEAD~1..HEAD -- tractor/_streaming.py`
+
+Added `raise_on_lag: bool = True` to `MsgStream.subscribe()`. The first call
+passes the policy to the irreversibly allocated root broadcaster and its
+child; later calls configure each child independently while retaining the
+root's first-call policy.
+
+> `git diff HEAD~1..HEAD -- tractor/to_asyncio.py`
+
+Added equivalent lag-policy forwarding to `LinkedTaskChannel.subscribe()`.
+
+> `git diff HEAD~1..HEAD -- docs/guide/streaming.rst`
+
+> `git diff HEAD~1..HEAD -- docs/guide/asyncio.rst`
+
+Documented strict versus warn/drop/resume behavior, independent child policy,
+and first-call root policy for both wrapper types.
+
+> `git diff HEAD~1..HEAD -- tests/test_task_broadcasting.py`
+
+Added a parameterized wrapper-level regression using minimal receive-compatible
+handles. It verifies a first non-raising subscription configures root and
+child, then a later strict child does not mutate the sticky root policy.
+
+Verification output:
+
+```text
+...................                                                      [100%]
+19 passed in 5.71s
+
+....................                                                     [100%]
+20 passed in 6.88s
+
+.                                                                        [100%]
+1 passed in 0.86s
+```
+
+The second and third runs cover actual `MsgStream` and infected-asyncio
+`LinkedTaskChannel` fan-out respectively. Python compilation and
+`git diff --check` passed. Adversarial review found no actionable issues.

ai/prompt-io/opencode/20260813T181901Z_a2e0df4b_prompt_io.md

@@ -0,0 +1,36 @@
+---
+model: openai/gpt-5.6-sol
+service: opencode
+session: unavailable
+timestamp: 2026-08-13T18:19:01Z
+git_ref: a2e0df4b
+scope: code
+substantive: true
+raw_file: 20260813T181901Z_a2e0df4b_prompt_io.raw.md
+---
+
+## Prompt
+
+Continue Tractor PR 490 after the paired piker EMS consumer commit. Clean
+cancelled-task diagnostics, define or reject zero-buffer broadcast behavior,
+review and verify the change, then stop at a complete commit plan.
+
+## Response summary
+
+Bound cancellation diagnostics to receiver progress, terminal state and
+resource lifetime; made EOC durable across peers; released wrapper-owned root
+broadcasters without breaking graceful EOC or subclass overrides; and rejected
+non-positive fan-out retention capacity.
+
+## Files changed
+
+- `tractor/trionics/_broadcast.py` - diagnostic lifecycle, durable EOC and
+  buffer validation.
+- `tractor/_streaming.py` - safe `MsgStream` root broadcaster cleanup.
+- `tractor/to_asyncio.py` - linked-channel root broadcaster cleanup.
+- `tests/test_task_broadcasting.py` - cancellation, EOC, wrapper and capacity
+  regressions.
+
+## Human edits
+
+None - generated output follows the requested fifth iterative item.

ai/prompt-io/opencode/20260813T181901Z_a2e0df4b_prompt_io.raw.md

@@ -0,0 +1,58 @@
+---
+model: openai/gpt-5.6-sol
+service: opencode
+timestamp: 2026-08-13T18:19:01Z
+git_ref: a2e0df4b
+diff_cmd: git diff HEAD~1..HEAD
+---
+
+The user asked to continue after committing the paired piker EMS consumer
+fix. The next isolated Tractor PR 490 item was to clean cancelled-task
+diagnostics and define zero-buffer broadcast behavior, then review, test and
+stop at a complete commit plan.
+
+> `git diff HEAD~1..HEAD -- tractor/trionics/_broadcast.py`
+
+Made `BroadcastState.cancelled` transient: receiver progress and close clear
+that receiver's diagnostic, terminal EOC and shared receive failure clear all
+stale cancelled tasks, and durable EOC prevents peers from re-entering the
+closed source. `broadcast_receiver()` now rejects non-positive retention
+capacity before creating an unusable zero-length deque.
+
+> `git diff HEAD~1..HEAD -- tractor/_streaming.py`
+
+Made explicit `MsgStream.aclose()` release its internally allocated root
+broadcaster while preserving graceful receive-internal EOC teardown. Used a
+task-local marker so the public zero-argument `aclose()` signature and valid
+subclass overrides remain compatible.
+
+> `git diff HEAD~1..HEAD -- tractor/to_asyncio.py`
+
+Made `LinkedTaskChannel.aclose()` release its internally allocated root
+broadcaster before closing the underlying Trio receive channel.
+
+> `git diff HEAD~1..HEAD -- tests/test_task_broadcasting.py`
+
+Added synchronized regressions for transient child cancellation diagnostics,
+cross-receiver terminal cleanup, durable EOC peer wakeups, root broadcaster
+cleanup through both public wrappers, `MsgStream.aclose()` subclass
+compatibility, and zero-buffer rejection.
+
+Verification output:
+
+```text
+...........................                                              [100%]
+27 passed in 5.89s
+
+.                                                                        [100%]
+1 passed in 1.22s
+
+.                                                                        [100%]
+1 passed in 0.88s
+```
+
+The integration runs cover real `MsgStream` actor fan-out and infected-asyncio
+`LinkedTaskChannel` fan-out. Ruff, Python compilation and `git diff --check`
+passed. Repeated adversarial review found and resolved root close re-entrancy,
+cross-receiver terminal retention, durable-EOC and subclass-compatibility
+issues; final review reported no findings.

ai/prompt-io/opencode/20260822T022526Z_5562fd9a_prompt_io.md

@@ -0,0 +1,49 @@
+---
+model: openai/gpt-5.6-sol
+service: opencode
+session: 76c5d31c-5a2f-4503-9b16-410ee7f4fab3
+timestamp: 2026-08-22T02:25:26Z
+git_ref: eb3c99c9
+scope: config
+substantive: true
+raw_file: 20260822T022526Z_5562fd9a_prompt_io.raw.md
+---
+
+## Prompt
+
+Perform a full Tractor repository scan for related `ai.skillz` work,
+then correct the run-tests landing branch, prune unrelated `.gitignore`
+additions, preserve only the focused migration, and provide canonical
+deployment commands.
+
+## Response summary
+
+Audited all local branches, worktrees, affected-path history, deployment
+state, canonical skill dependencies, and current Tractor harness behavior.
+Corrected the local test reference where it overstated cleanup safety or
+omitted current environment, platform, debugger, timeout, and CI details.
+Narrowed the correction commit to three managed `run-tests` deployment
+blocks. A later dedicated commit records the complete generated `ai.skillz`
+deployment state.
+
+## Files changed
+
+- `.claude/skills/run-tests/test-harness-reference.md` - correct the
+  project-specific test and cleanup contract.
+- `.gitignore` - narrow the correction commit before the later dedicated
+  deployment-state expansion.
+- `ai/prompt-io/opencode/20260822T022526Z_5562fd9a_prompt_io.md` - record
+  the migration review provenance.
+- `ai/prompt-io/opencode/20260822T022526Z_5562fd9a_prompt_io.raw.md` -
+  preserve the unedited response record.
+
+## Human edits
+
+The human required an existing-work scan after duplicate implementation
+was discovered, approved correcting the landing branch during PR #481
+review, and directed removal or reconciliation of unrelated ignore rules.
+During PR #510 review, the human required the stackscope and shared-memory
+safety clarifications and immutable provenance pointers before landing. No
+direct source-line edits were made by the human. Copilot review then prompted
+the human to require explicit canonical deployment instructions and clarify
+the later `.gitignore` expansion.

ai/prompt-io/opencode/20260822T022526Z_5562fd9a_prompt_io.raw.md

@@ -0,0 +1,24 @@
+---
+model: openai/gpt-5.6-sol
+service: opencode
+timestamp: 2026-08-22T02:25:26Z
+git_ref: eb3c99c9
+diff_cmd: git diff eb3c99c9^..eb3c99c9
+---
+
+Audit Tractor's repository-wide `ai.skillz` state and correct the
+`wkt/ai_skillz_run_tests_landing` migration before landing it.
+
+> `git diff eb3c99c9^..eb3c99c9 -- .claude/skills/run-tests/test-harness-reference.md`
+
+The harness reference is corrected against current Tractor behavior:
+environment overrides, CI/platform matrices, randomized registry
+addresses, interactive debugger capture, timeout mechanisms, stackscope
+artifacts, reaper platform limits, dry-run expansion, and exact
+change-to-test mappings.
+
+> `git diff eb3c99c9^..eb3c99c9 -- .gitignore`
+
+Unrelated Taken and broad OpenCode command ignore additions are removed;
+only the managed canonical `run-tests` link and command blocks remain in
+this migration.

ai/prompt-io/opencode/20260828T014247Z_d06ee34b_prompt_io.md

@@ -0,0 +1,47 @@
+---
+model: openai/gpt-5.6-sol
+service: opencode
+session: pr490-landing-review
+timestamp: 2026-08-28T01:42:47Z
+git_ref: wkt/fix_broadcast_lag_count
+scope: code
+substantive: true
+raw_file: 20260828T014247Z_d06ee34b_prompt_io.raw.md
+---
+
+## Prompt
+
+Prime PR #490 for landing, review its concurrent broadcast changes,
+fix valid findings, and keep the landing pass focused.
+
+## Response summary
+
+Rejected an invalid mutable-default finding under `msgspec.Struct`
+semantics. Found and fixed a close-time race where concurrent receives
+on one `BroadcastReceiver` could overwrite its sole cancellation scope
+and strand an earlier waiter. The receiver now rejects overlapping
+calls with `trio.BusyResourceError`. Public docs now define each
+receiver as one logical subscription cursor and require one receiver
+per concurrent lossless consumer. Busy errors identify the active Trio
+task which owns the receiver.
+
+## Files changed
+
+- `tractor/trionics/_broadcast.py` - reject concurrent receives on one
+  receiver handle.
+- `tests/test_task_broadcasting.py` - prove the second call fails while
+  the original receive remains usable.
+- `docs/guide/streaming.rst` - document per-task receiver ownership and
+  explicit lag behavior.
+
+## Human edits
+
+The human selected PR #490 as the next landing target, rebased it onto
+current `main`, directed the agent to continue the landing pass, and
+required a faster, tightly scoped execution after the review expanded.
+The human then confirmed the per-receiver guard and required the public
+method, type, and surrounding guide to emphasize lossless per-consumer
+semantics. The human also required the regression's checkpoint loop to
+explain its synchronization and the busy error to report the active
+consumer task. The agent applied the source, test, and documentation
+changes; no direct manual edits to these files were identified.

ai/prompt-io/opencode/20260828T014247Z_d06ee34b_prompt_io.raw.md

@@ -0,0 +1,56 @@
+---
+model: openai/gpt-5.6-sol
+service: opencode
+timestamp: 2026-08-28T01:42:47Z
+git_ref: wkt/fix_broadcast_lag_count
+diff_cmd: git diff
+---
+
+## Prompt
+
+Get PR #490 primed and landed. Continue after rebasing, but keep the
+review and fix pass focused.
+
+## Response
+
+The mutable `{}` default is safe because `msgspec.Struct` creates a
+fresh dictionary per instance. A separate concurrency defect remains:
+two tasks can call `receive()` through one `BroadcastReceiver`, and the
+second call overwrites the only `_wait_scope`. Closing that receiver can
+then leave the first task blocked indefinitely.
+
+> `git diff -- tractor/trionics/_broadcast.py tests/test_task_broadcasting.py`
+
+Added a per-receiver in-progress guard that raises
+`trio.BusyResourceError` for an overlapping receive. Added a regression
+test that blocks the first receive, verifies the second call is
+rejected, and releases the source to prove the first call still works.
+The complete broadcast test file passes: 28 tests in 4.57 seconds.
+
+## Follow-up prompt
+
+Clarify that the guard is per receiver, preserves lossless consumer
+semantics, and document the contract on the public method, type, and
+surrounding guide.
+
+## Follow-up response
+
+> `git diff -- tractor/trionics/_broadcast.py docs/guide/streaming.rst`
+
+Documented that each `BroadcastReceiver` owns one logical cursor, each
+concurrent consumer needs its own subscribed receiver, overlapping
+calls on one handle raise `BusyResourceError`, and strict lag handling
+never skips values silently.
+
+## Second follow-up prompt
+
+Explain that the polling loop waits until the background receive is
+blocked before making the concurrent call, and include the first
+consumer task's runtime information in the busy exception.
+
+## Second follow-up response
+
+Replaced the boolean guard with the active `trio.lowlevel.Task`, added
+its name and representation to `BusyResourceError`, named the fixture
+task for a deterministic assertion, and documented the checkpoint-loop
+interleaving directly above the poll.

ai/tpt-backends/00_shared_backend_contract.md

@@ -0,0 +1,403 @@
+# `tractor.ipc` next-gen transport backends: the shared contract
+
+Status: design doc / implementation spec.
+Audience: any model or human implementing one of the three
+sibling plans in this directory.
+
+- [`01_tipc_backend.md`](./01_tipc_backend.md) — `AF_TIPC`
+  (gh #378)
+- [`02_quic_iroh_backend.md`](./02_quic_iroh_backend.md) — QUIC
+  via `iroh` FFI, uniffi-async rewritten onto `trio` (gh #353)
+- [`03_wg_tunnel_bindspace.md`](./03_wg_tunnel_bindspace.md) —
+  WireGuard (and other shuttle-able) tunnels as a *nested
+  bindspace* layer via `pyroute2` (gh #482, #443)
+
+This doc is the **normative** description of what a `tractor`
+transport backend *is* as of `main@83b34884`. Each sibling plan
+assumes it and only documents its own deltas. Read this first;
+do not re-derive it from the code.
+
+---
+
+## 0. Why a shared contract doc
+
+The three plans are meant to be implementable *independently and
+concurrently* by different models/providers without design
+drift. Everything they share — the backend duck-type, the
+registration tables, the test harness plumbing, the naming and
+code-style rules — lives here exactly once. If an implementer
+finds this doc disagrees with `main`, **the code wins**; fix this
+doc in the same PR.
+
+---
+
+## 1. The backend duck-type (empirical, from `_tcp.py`/`_uds.py`)
+
+A transport backend is **one module** under `tractor/ipc/`
+exposing exactly four things. There is no ABC to subclass and no
+plugin entrypoint; wiring is by explicit table registration
+(§2) plus one piece of reflection (§1.3).
+
+### 1.1 `class <Proto>Address(msgspec.Struct, frozen=True)`
+
+Structurally conforms to the `Address` `Protocol` in
+`tractor/discovery/_addr.py:82`. Required surface:
+
+| member | kind | notes |
+| --- | --- | --- |
+| `proto_key` | `ClassVar[str]` | the wire/registry key, e.g. `'tcp'`, `'uds'` |
+| `unwrapped_type` | `ClassVar[type]` | the primitive tuple shape |
+| `def_bindspace` | `ClassVar` | default bindspace value |
+| `is_valid` | `@property -> bool` | "is this a *dialable/bindable* addr" |
+| `bindspace` | `@property` | the "set of hosts"-ish scope (see below) |
+| `from_addr(cls, addr)` | `@classmethod` | primitive -> wrapped, `match`-based |
+| `unwrap(self)` | method | wrapped -> primitive (must be msgpack-native!) |
+| `get_random(cls, bindspace=...)` | `@classmethod` | per-subactor ephemeral addr |
+| `get_root(cls)` | `@classmethod` | host-singleton default registrar addr |
+| `__repr__` | method | `f'{type(self).__name__}[{...}]'` house style |
+
+Hard constraints learned from the existing two:
+
+- **`frozen=True`.** Addresses are dict keys
+  (`Server.epsdict()`, `Endpoint.peer_tpts`) and are compared by
+  value all over the runtime.
+- **`.unwrap()` output must round-trip through `msgspec` and
+  through `wrap_address()`.** It is what actually crosses the
+  wire in `SpawnSpec`/`_root_mailbox`/`_registry_addrs`, and it
+  is what `Actor.reg_addrs` and every test compares against. If
+  your unwrapped form is not *uniquely* pattern-matchable
+  against the other backends' forms in
+  `wrap_address()` (`_addr.py:230`), you have a bug that
+  manifests as the wrong transport being loaded — the file's own
+  `XXX NOTE` warns about precisely this.
+
+  ⚠️ **and shape-matching does not survive 4 backends.** Adding
+  TIPC and iroh breaks it outright: TIPC's natural form is a
+  `(str, int)` — indistinguishable from `TCPAddress` — and
+  iroh's is a `(str, str)`, which the *existing* UDS case
+  (`case (_, filename) if type(filename) is str`) already
+  swallows. Ordering hacks and prefix-tagging (an earlier
+  revision of plan 01 proposed `('tipc:<stype>:<scope>', inst)`)
+  paper over it at best.
+
+  **The fix, and the recommended prerequisite for all three
+  backends: make the unwrapped form carry an explicit
+  proto-key, using the `multiaddr` protocol name as the
+  canonical spelling** — `('tcp', host, port)`,
+  `('unix', path)`, `('udp', ...)`, `('tipc', stype, inst,
+  scope)`. Then `wrap_address()` collapses from an
+  order-sensitive `match` to `_address_types[addr[0]]`, and the
+  whole collision class stops existing. Note this *also* aligns
+  the on-wire form with `mk_maddr()`/`parse_maddr()`, so the two
+  representations stop being independent inventions.
+
+  Two consequences to plan for:
+  - it's a **wire-format change** (`SpawnSpec`,
+    `_root_mailbox`, `_registry_addrs`) plus every test fixture
+    and downstream config (`piker`'s `[network]` table). It
+    wants its **own migration commit, landed before any new
+    backend**, not smuggled into one.
+  - it's the moment to **stop handing raw unwrapped tuples to
+    users at all.** The long-term shape is: `Address` subtypes
+    are the public currency and `UnwrappedAddress` becomes an
+    internal serialization detail — the same discipline
+    `ipaddress` uses (you pass `IPv4Address`, not a 4-tuple).
+    Public API should accept `Address|maddr-str` and treat bare
+    tuples as legacy-tolerated input, ideally deprecated.
+- **`.get_random()` must be collision-free without a live
+  runtime.** See the `UDSAddress.get_random()` uuid-token
+  comment (`_uds.py:207-220`): with no `current_actor()` the
+  sockname degenerates to a pure fn of `(prefix, pid)` and two
+  calls in one proc alias. Mix in a `uuid4().hex[:8]` token.
+- **`.bindspace` semantics**: "the address' bindable space" —
+  ip/host for `tcp`, the socket-file *directory* for `uds`. For
+  the new backends: the TIPC *scope* (§1 of plan 01), the iroh
+  *ALPN + relay/discovery realm* (plan 02), the netns (plan 03).
+  `Address.namespace` is already spec'd in the Protocol as
+  "the if-available OS-specific network namespace key" and is
+  currently unimplemented by both backends — plan 03 is the
+  first real consumer.
+
+### 1.2 module-level listener lifecycle
+
+```python
+async def start_listener(
+    addr: <Proto>Address,
+    **kwargs,
+) -> trio.SocketListener   # or a trio.abc.Listener, see §3
+    ...
+
+def close_listener(          # OPTIONAL
+    addr: <Proto>Address,
+    lstnr: trio.abc.Listener,
+) -> None:
+    ...
+```
+
+`close_listener()` is optional; `Endpoint.close_listener()`
+(`_server.py:674`) `getattr`s it and treats absence as "closing
+is implicit". `uds` needs it (unlinks the sock-file), `tcp`
+does not.
+
+### 1.3 the ONE piece of reflection you must not break
+
+`Endpoint.start_listener()` (`_server.py:656`):
+
+```python
+tpt_mod: ModuleType = inspect.getmodule(self.addr)
+lstnr = await tpt_mod.start_listener(addr=self.addr)
+```
+
+The transport module is found by `inspect.getmodule()` **on the
+`Address` instance**. Therefore: *the `Address` class and its
+`start_listener()`/`close_listener()` MUST live in the same
+module.* Do not define the address type in `_types.py` or a
+`_addrs.py` and the listener elsewhere.
+
+Immediately after, the same method does:
+
+```python
+if (unwrapped := lstnr.socket.getsockname()) != self.addr.unwrap():
+    self.addr = self.addr.from_addr(unwrapped)
+```
+
+i.e. it assumes `lstnr.socket.getsockname()` exists and that its
+return value is a valid `from_addr()` input. This is fine for
+TIPC (§3 of plan 01) and **is the main integration hazard for
+iroh** (§3 of plan 02) — plans that break it must say so
+explicitly and propose the upstream `_server.py` patch.
+
+### 1.4 `class Msgpack<Proto>Stream(MsgpackTransport)`
+
+Subclass `tractor.ipc._transport.MsgpackTransport`. You inherit
+all framing (`<I` 4-byte little-endian length prefix),
+`msgspec` codec ctx-var lookup, `TransportClosed` normalization,
+`.drain()`, `__aiter__`. You implement only:
+
+| member | notes |
+| --- | --- |
+| `address_type` | the `<Proto>Address` class |
+| `layer_key: int` | OSI-ish layer, `4` for both current backends |
+| `maddr` `@property` | `-> Multiaddr\|str`, via `mk_maddr(self.raddr)` |
+| `connected(self) -> bool` | `tcp`/`uds` both use `self.stream.socket.fileno() != -1` |
+| `connect_to(cls, addr, prefix_size=4, codec=None, **kw)` | `@classmethod`, returns an instance |
+| `get_stream_addrs(cls, stream) -> (laddr, raddr)` | `@classmethod`, called from `MsgpackTransport.__init__` |
+
+`MsgpackTransport.__init__` requires the object passed as
+`stream` to satisfy:
+
+- `await stream.send_all(bytes)`
+- usable as `tricycle.BufferedReceiveStream(transport_stream=stream)`,
+  i.e. `await stream.receive_some(n)`
+- `trio.BrokenResourceError` / `trio.ClosedResourceError` /
+  `ValueError('...unclean EOF...')` on the failure paths that
+  `_iter_packets()` and `send()` already `match` on
+  (`_transport.py:221-304`, `:436-499`).
+
+That is **`trio.abc.Stream`, not `trio.SocketStream`**. The
+`MsgTransport` Protocol's `stream: trio.SocketStream`
+annotation (`_transport.py:83`) is a lie of convenience — the
+actual `MsgpackTransport.__init__` param is typed
+`trio.abc.Stream` and nothing in the msg path touches
+`.socket`. Only `connected()` (which each backend defines) and
+`Endpoint.start_listener()`'s `getsockname()` do.
+
+### 1.5 verified-good news for socket-family backends
+
+Both `trio.SocketStream` and `trio.SocketListener` are
+**address-family agnostic**. Verified against the installed
+`trio` (`trio/_highlevel_socket.py`): the only constructor
+checks are
+
+- `isinstance(socket, trio.socket.SocketType)`
+- `socket.type == SOCK_STREAM`
+- (listener) `getsockopt(SOL_SOCKET, SO_ACCEPTCONN)` is truthy,
+  with `OSError` **suppressed** (the macOS carve-out, which
+  also covers exotic families that reject the opt)
+
+There is no `AF_*` check and no `IPPROTO_TCP` hard dependency
+(`TCP_NODELAY`/`TCP_NOTSENT_LOWAT` are set under
+`suppress(OSError)`). Consequence: **any `SOCK_STREAM` family
+CPython can create — including `AF_TIPC` — drops straight into
+the existing `trio.SocketStream` + `trio.serve_listeners()`
+path.** This is why plan 01 is small and plan 02 is not.
+
+---
+
+## 2. Registration tables (the full wiring checklist)
+
+Adding a backend touches these and only these:
+
+1. `tractor/runtime/_state.py:46`
+   `TransportProtocolKey = Literal['tcp', 'uds', ...]` — add the
+   key. This `Literal` is the canonical set; `_testing/pytest.py`
+   drives `--tpt-proto` validation off `_addr._address_types`,
+   and the spawn-backend fixture already models the
+   "drive-the-set-from-the-Literal" pattern
+   (`pytest.py:870-880`) — do the same rather than hardcoding.
+2. `tractor/discovery/_addr.py:173` `_address_types: bidict` —
+   `{'<key>': <Proto>Address}`. Note it is a **`bidict`**, so
+   the mapping must stay 1:1.
+3. `tractor/discovery/_addr.py:181` `_default_lo_addrs` —
+   `'<key>': <Proto>Address.get_root().unwrap()`.
+   ⚠️ this dict is built at **import time**, so
+   `get_root()` must not require a live runtime, a loaded kernel
+   module, or network I/O. (`UDSAddress.def_bindspace =
+   get_rt_dir()` is the precedent for "cheap, pure, filesystem-
+   ish".) A backend whose root addr needs I/O must make this
+   entry lazy — propose that refactor explicitly.
+4. `tractor/discovery/_addr.py:230` `wrap_address()` `match` —
+   add a case iff your `unwrapped_type` isn't already uniquely
+   matched. **Preferably do the proto-key migration in §1.1
+   first**, after which this step becomes a one-line
+   `_address_types` entry instead of an order-sensitive `case`.
+5. `tractor/ipc/_types.py` — `Address` union alias,
+   `_msg_transports` list, `_key_to_transport[('msgpack', key)]`,
+   `_addr_to_transport[<Proto>Address]`.
+6. `tractor/ipc/_types.py:92` `transport_from_stream()` — the
+   `sock.family` `match`. For a non-socket stream type (iroh)
+   this needs a different discriminator; see plan 02 §3.3.
+7. `tractor/discovery/_multiaddr.py` —
+   `_tpt_proto_to_maddr`, and a `case` in both `mk_maddr()` and
+   `parse_maddr()`.
+8. `tractor/ipc/__init__.py` — re-export if the backend has a
+   public surface.
+9. `tractor/_testing/addr.py::get_rando_addr()` — per-proto
+   branch so the whole suite can run under `--tpt-proto <key>`.
+10. `pyproject.toml` — new deps go in an **optional extra**, never
+    in `[project].dependencies`. See §5.
+
+## 3. Where the `trio.SocketListener` assumption is load-bearing
+
+`_serve_ipc_eps()` (`_server.py:1041`) annotates
+`listener: trio.abc.Listener` and hands the list to
+`trio.serve_listeners(handler=handle_stream_from_peer,
+listeners=..., handler_nursery=stream_handler_tn)`.
+`trio.serve_listeners` itself is generic over
+`trio.abc.Listener`. So the *only* `SocketListener`-specific
+code in the server path is the `getsockname()` reconciliation in
+`Endpoint.start_listener()` (§1.3) and the type annotations.
+
+`handle_stream_from_peer()` (`_server.py:298`) then does
+`Channel.from_stream(stream)` →
+`transport_from_stream(stream)` → `sock.family` match (§2.6).
+
+**Therefore**: a non-socket backend needs (a) a
+`trio.abc.Listener` subclass, (b) a change to
+`Endpoint.start_listener()` to not blindly `getsockname()`, and
+(c) a change to `transport_from_stream()`'s discrimination.
+All three are small, upstream-able, and *should be landed as
+their own prep PR* before the backend itself — see plan 02 §3.
+
+## 4. Handshake / discovery invariants you inherit
+
+- Every accepted stream immediately does
+  `chan._do_handshake(aid=actor.aid)`; a peer that fails it is
+  logged at `runtime` and dropped, **not** raised
+  (`_server.py:334-365`). Discovery-sys "pings" rely on this,
+  so your `connect_to()` must raise something that normalizes
+  to `TransportClosed`/`ConnectionError` on a dead peer, never
+  a novel exception type.
+- `_root.py:381-406` fail-fasts when a `registry_addrs` entry's
+  `proto_key` is not in `enable_transports`. Your key must be
+  spellable in both.
+- `_root.py:256` currently enforces `len(enable_transports) == 1`.
+  Multi-tpt actors are a separate work item; none of these three
+  plans may depend on lifting it.
+- Sub-actor bind addrs come from
+  `_runtime.py:1600-1610`: for each key in the parent-supplied
+  `enable_transports`, `get_address_cls(key).get_random()`.
+  So `get_random()` runs *in the child, post-fork, pre-listen*.
+  Anything it needs (kernel module, netns membership, an iroh
+  secret key) must already be true at that moment.
+
+## 5. Dependency policy
+
+`[project].dependencies` stays lean (see the boot-latency work,
+gh #470: `import tractor` is budgeted at ~0.145s). Every new
+backend dep is an extra:
+
+```toml
+[project.optional-dependencies]
+tipc = []                       # stdlib-only!
+quic = ["iroh>=0.35"]           # pin per plan 02 §1
+wg   = ["pyroute2>=0.9"]        # pin per plan 03 §1
+```
+
+and every backend module must be **import-lazy**: a
+`tractor/ipc/_<proto>.py` that imports its 3rd-party dep at
+module scope must not be imported by `tractor/__init__.py`,
+`tractor/ipc/__init__.py`, or `tractor/discovery/_addr.py`'s
+import-time table construction. The `_addr._default_lo_addrs`
+eager-dict (§2.3) is the trap: keep the backend's `get_root()`
+dep-free, or make that table lazy.
+
+## 6. Test-harness plumbing (identical for all three)
+
+- `--tpt-proto <key>` (`_testing/pytest.py:409`) selects the
+  session-wide proto; the `tpt_proto` fixture mutates
+  `_state._def_tpt_proto` + `_runtime_vars['_enable_tpts']`
+  (`pytest.py:807-835`). Adding the key to `_address_types` is
+  what makes `--tpt-proto <key>` legal (`pytest.py:795-800`
+  asserts the lookup).
+- The **acceptance bar** for every backend is: the *entire*
+  existing suite passes under `--tpt-proto <key>`, unmodified.
+  That is the whole point of the abstraction. Backend-specific
+  unit tests go in `tests/ipc/test_each_tpt.py` (the existing
+  `test_uds_bindspace_created_implicitly` /
+  `test_uds_double_listen_raises_connerr` are the model).
+- Capability gating: each backend needs a **cheap, pure
+  predicate** + a `pytest.mark.skipif`, because these are all
+  environment-dependent. Verified example: on this dev box
+  `socket.socket(AF_TIPC, SOCK_STREAM)` raises
+  `OSError(97, 'Address family not supported by protocol')`
+  because the `tipc` module isn't loaded. Put the predicate in
+  the backend module (so apps can use it too), not in the test.
+- New pytest marks must be registered in `pyproject.toml`, per
+  the project's fix-warnings-at-source rule (gh #469).
+
+## 7. Code style (non-negotiable, matches the repo)
+
+- module header tagline: `# tractor: distributed structured
+  concurrency.` for **new** files (not the legacy
+  `structured concurrent "actors".` form the existing `_tcp.py`
+  carries).
+- AGPL header block copied verbatim from `_tcp.py`.
+- `from __future__ import annotations` first.
+- annotate *everything*, including locals:
+  `sockpath: Path = addr.sockpath`.
+- `match`/`case` over `isinstance` chains for address and
+  error dispatch.
+- multi-line call/`import` style with trailing commas.
+- never emit a whitespace-only line.
+- error messages are multi-line f-strings ending in `\n`, with
+  the `f'...\n' f'...\n'` implicit-concat layout and the
+  `>[`/`[>`/`<=(` nested-op sigils where a `nest_from_op()` is
+  in play.
+- prefer pure functions + module-level helpers over methods;
+  keep `Address` types data-only. Where a helper needs
+  scoped setup/teardown, it's an `@acm` — not a class with
+  `.start()`/`.stop()`.
+- pure getters: no `get_*(..., mutate=True)` flags; split into
+  a read-only getter and an explicit sibling setter.
+
+---
+
+## 8. Cross-plan sequencing
+
+The three are independent *except*:
+
+- plan 02 (iroh) needs the `_server.py` /
+  `transport_from_stream()` generalization (§3) — plan 01 does
+  **not**, and should therefore land first as the cheap proof
+  that the table-registration story works for a genuinely new
+  proto.
+- plan 03 (wg) composes *under* whatever L4 tpt is in use and
+  its netns work is what finally implements
+  `Address.namespace`. It can land before or after 02, but its
+  `TunnelledAddress` design must be reviewed against plan 02's
+  address shape so the "tunnelled maddr" grammar (gh #443)
+  covers `/…/quic-v1/p2p/…` inner addrs too.
+- All three want first-class `wg`/`quic`/`tipc` protos in
+  `py-multiaddr`; that upstream track is gh #483 and
+  multiformats/py-multiaddr#107/#108.

ai/tpt-backends/01_tipc_backend.md

@@ -0,0 +1,707 @@
+# Plan 01 — `TIPC` transport backend (`tractor/ipc/_tipc.py`)
+
+Tracks gh [#378]. Prereq reading:
+[`00_shared_backend_contract.md`](./00_shared_backend_contract.md).
+
+**Thesis**: TIPC is the *cheapest* new backend we can add and
+simultaneously the only one that gives us cluster-wide service
+discovery **for free, in the kernel**, replacing (for
+TIPC-capable deployments) the whole `tractor.discovery`
+registrar round-trip with a `bind()`/`connect()` on a
+*service name*. It is stdlib-only: zero new dependencies.
+
+[#378]: https://github.com/goodboy/tractor/issues/378
+
+---
+
+## 1. Why this is small: three verified facts
+
+1. **CPython already speaks TIPC.** `socket.AF_TIPC` plus 23
+   `TIPC_*` constants are present in the stdlib on Linux
+   (verified on the dev box, py3.13):
+   `AF_TIPC, SOL_TIPC, TIPC_ADDR_ID, TIPC_ADDR_NAME,
+   TIPC_ADDR_NAMESEQ, TIPC_CFG_SRV, TIPC_CLUSTER_SCOPE,
+   TIPC_CONN_TIMEOUT, TIPC_{CRITICAL,HIGH,MEDIUM,LOW}_IMPORTANCE,
+   TIPC_DEST_DROPPABLE, TIPC_IMPORTANCE, TIPC_NODE_SCOPE,
+   TIPC_PUBLISHED, TIPC_SRC_DROPPABLE, TIPC_SUBSCR_TIMEOUT,
+   TIPC_SUB_CANCEL, TIPC_SUB_PORTS, TIPC_SUB_SERVICE,
+   TIPC_TOP_SRV, TIPC_WAIT_FOREVER, TIPC_WITHDRAWN,
+   TIPC_ZONE_SCOPE`.
+   `sock.bind()/connect()/getsockname()` take/return the
+   5-tuple `(addr_type, v1, v2, v3, scope)` — the last element
+   is optional on input and defaults to `0`.
+2. **`trio` doesn't care about the address family.** Per
+   contract §1.5, `trio.SocketStream` and `trio.SocketListener`
+   only require a trio socket object of type `SOCK_STREAM`.
+   TIPC's `SOCK_STREAM` is a real connection-oriented reliable
+   byte stream. So we reuse `trio.SocketStream`,
+   `trio.SocketListener`, `trio.serve_listeners()`,
+   `MsgpackTransport`'s framing — *all of it*.
+3. **It is not available by default.** On this box
+   `socket.socket(AF_TIPC, SOCK_STREAM)` →
+   `OSError(97, 'Address family not supported by protocol')`
+   with no `tipc` in `/proc/modules`. `modprobe tipc` is
+   required; cross-node needs a bearer
+   (`tipc bearer enable media eth device <if>` or
+   `media udp name <n> localip <ip>`). Everything about this
+   plan's testability hinges on gating (§7).
+
+Non-goals: `SOCK_RDM`/`SOCK_DGRAM`/`SOCK_SEQPACKET` message
+modes, multicast fan-out, and TIPC group messaging. They are
+genuinely interesting for a future `tractor` broadcast/pubsub
+transport but they do **not** fit `MsgTransport`'s
+stream-of-length-prefixed-msgs shape. Note them in the
+follow-up issue, do not build them here.
+
+---
+
+## 2. `TIPCAddress`
+
+### 2.1 the three TIPC address flavours, and which we use
+
+| flavour | tuple | meaning |
+| --- | --- | --- |
+| `TIPC_ADDR_NAMESEQ` | `(type, lower, upper, scope)` | a *published range* — what a server `bind()`s |
+| `TIPC_ADDR_NAME` | `(type, instance, domain, scope)` | a *lookup* — what a client `connect()`s |
+| `TIPC_ADDR_ID` | `(node, ref, 0, scope)` | a concrete port id — the "physical" address |
+
+The design decision that makes this backend coherent:
+
+> **A `tractor` actor's TIPC address is a *service name*
+> `(type, instance)`; `bind()` publishes the singleton range
+> `(type, instance, instance)`; peers `connect()` by name and
+> the kernel resolves + load-balances. `TIPC_ADDR_ID` is only
+> ever an *observed* address (`getpeername()`), never a
+> user-facing one.**
+
+This is exactly the "leverage the built-in discovery machinery"
+ask in #378: publishing a bind *is* registration, and
+`connect()` on a name *is* a lookup, with no registrar actor in
+the loop.
+
+### 2.2 the struct
+
+```python
+class TIPCAddress(
+    msgspec.Struct,
+    frozen=True,
+):
+    _stype: int                    # TIPC "type" == service class
+    _instance: int                 # service instance within the type
+    _scope: int = TIPC_CLUSTER_SCOPE
+    # observed-only, never part of identity/equality-by-intent
+    maybe_node: int|None = None    # from TIPC_ADDR_ID getpeername()
+    maybe_ref: int|None = None
+
+    proto_key: ClassVar[str] = 'tipc'
+    unwrapped_type: ClassVar[type] = tuple[str, int]
+    def_bindspace: ClassVar[int] = TIPC_CLUSTER_SCOPE
+```
+
+**Unwrapped form** (the wire/`SpawnSpec` shape).
+
+TIPC's natural form is `(stype, instance, scope)` — but a
+2-tuple squeeze of it is a `(str, int)`, i.e. *the same coarse
+shape as `TCPAddress`*, so `wrap_address()`'s
+`case (str(), int())` steals it. This backend is therefore the
+forcing function for the contract-doc's conclusion (§1.1):
+
+> **make the unwrapped form carry an explicit proto-key, spelled
+> with the `multiaddr` protocol name.**
+
+```python
+def unwrap(self) -> tuple[str, int, int, int]:
+    return ('tipc', self._stype, self._instance, self._scope)
+```
+
+`wrap_address()` then dispatches `_address_types[addr[0]]` and
+the collision class disappears. **This is a prerequisite
+migration commit, not part of this backend** — see contract §1.1
+for its blast radius (wire format + every fixture + `piker`
+config) and for the follow-on "stop handing raw tuples to users
+at all, à la `ipaddress`" direction.
+
+⚠️ an earlier revision of this plan proposed a self-tagging
+`('tipc:<stype>:<scope>', instance)` string-prefix hack with an
+ordered `case` guard. **Dropped** — it papers over the problem,
+keeps `wrap_address()` order-sensitive, and doesn't help iroh's
+`(str, str)`-vs-UDS collision at all. Do not resurrect it.
+
+Note `TIPCAddress` is the first backend where `.unwrap()` is
+**not** a lossless view of the live socket — `maybe_node`/
+`maybe_ref` are observed metadata, exactly like
+`UDSAddress.maybe_pid` (which is likewise excluded from
+`.unwrap()`). Follow that precedent, including its `__repr__`
+treatment (`_uds.py:242`).
+
+### 2.3 how to pick `_stype` and `_instance`
+
+- `_stype` = a `tractor`-reserved service class. TIPC reserves
+  0..63 for internal use (`TIPC_TOP_SRV == 1`,
+  `TIPC_CFG_SRV == 0`). Use a module constant
+  `TRACTOR_STYPE: int = 0x74_72_00_00` ("tr\0\0") as the default
+  and make it overridable via `TIPCAddress._stype` so an app
+  can partition service classes. Document that two `tractor`
+  trees sharing a cluster **and** a `_stype` share a namespace.
+- `_instance` for `get_root()`: `1616` — mirrors the
+  `TCPAddress.get_root()` port and the `registry@1616.sock`
+  UDS filename, so the "1616 is tractor's registrar" idiom
+  holds across all backends.
+- `_instance` for `get_random()`: TIPC gives us no
+  kernel-assigned-instance analogue of `port=0`, so we must
+  choose. Use a *pure* fn of the actor identity so it is
+  reproducible and collision-free:
+  ```python
+  # 32-bit instance derived from the actor's uuid4 (+ pid when
+  # there's no live runtime, per the UDS precedent).
+  inst: int = int.from_bytes(
+      blake2b(seed.encode(), digest_size=4).digest(),
+      'big',
+  )
+  ```
+  where `seed = f'{actor.aid.name}@{pid}'` if
+  `current_actor(err_on_no_runtime=False)` else
+  `f'{prefix}.{uuid4().hex[:8]}@{pid}'`. Must avoid the reserved
+  low range: `inst = 64 + (inst % (2**32 - 64))`.
+  ⚠️ *unlike* `port=0`, a collision here surfaces as a
+  successful-but-shared publication (TIPC allows multiple
+  binders on the same name and round-robins!) rather than
+  `EADDRINUSE`. That is a silent-crosstalk failure mode; §7 has
+  the test that proves the 4-byte digest is enough and §9 has
+  the mitigation if it isn't.
+- `_scope`: `TIPC_NODE_SCOPE` for a same-host-only actor (the
+  UDS-equivalent), `TIPC_CLUSTER_SCOPE` (default) for
+  cluster-visible. **This is `.bindspace`**:
+  ```python
+  @property
+  def bindspace(self) -> int:
+      return self._scope
+  ```
+  It is the honest analogue of "the set of hosts this bind is
+  reachable from", which is precisely the docstring in
+  `Address.bindspace`. (`TIPC_ZONE_SCOPE` is deprecated/aliased
+  to cluster in modern kernels — accept it on input, normalize
+  to cluster, log at `transport` level.)
+
+### 2.4 `is_valid`
+
+```python
+@property
+def is_valid(self) -> bool:
+    return (
+        self._instance != 0
+        and
+        self._stype not in _tipc_reserved_stypes  # {0, 1, ...}
+        and
+        self._scope in (TIPC_NODE_SCOPE, TIPC_CLUSTER_SCOPE)
+    )
+```
+
+---
+
+## 3. Listener + stream
+
+### 3.1 `start_listener()`
+
+```python
+async def start_listener(
+    addr: TIPCAddress,
+    backlog: int = 128,
+    **kwargs,
+) -> SocketListener:
+    sock = trio.socket.socket(
+        socket.AF_TIPC,
+        socket.SOCK_STREAM,
+    )
+    # publish the singleton name-range == "register the service"
+    await sock.bind((
+        socket.TIPC_ADDR_NAMESEQ,
+        addr._stype,
+        addr._instance,
+        addr._instance,
+        addr._scope,
+    ))
+    sock.listen(backlog)
+    return SocketListener(sock)
+```
+
+Notes / hazards:
+
+- `bind()` on `AF_TIPC` is **not** a filesystem or port-table
+  operation and can't block on DNS, but keep it `await`ed
+  through `trio.socket` anyway for uniformity.
+- `backlog=128` matching `_uds.start_listener()`'s hard-won
+  value (see its comment at `_uds.py:317-331` re: concurrent
+  deregistration storms). Do not use `1`.
+- **no `close_listener()` needed** — nothing to unlink. Omit the
+  function entirely (contract §1.2: absence means implicit).
+  Withdrawal of the published name happens on socket close.
+- ⚠️ `SocketListener.__init__` will try
+  `getsockopt(SOL_SOCKET, SO_ACCEPTCONN)`. If TIPC rejects it,
+  trio's `except OSError: pass` covers us. Assert this in a
+  unit test rather than assuming.
+- Wrap the bind in a `_reraise_as_connerr()`-style `@cm` (copy
+  the `_uds.py:256` pattern) so `EADDRINUSE`-ish and
+  `EAFNOSUPPORT` become `ConnectionError` with the addr in the
+  message. `EAFNOSUPPORT` here means "kernel module not
+  loaded" and deserves a *specifically actionable* message:
+  `'TIPC unavailable — try `sudo modprobe tipc`\n'`.
+
+### 3.2 the `getsockname()` reconciliation
+
+`Endpoint.start_listener()` does
+`if lstnr.socket.getsockname() != self.addr.unwrap(): self.addr =
+self.addr.from_addr(unwrapped)`.
+
+For TIPC, `getsockname()` on a bound-but-listening socket
+returns a `TIPC_ADDR_ID`-flavoured 5-tuple (the port id), *not*
+the name-seq we bound. So the `!=` is **always true** and
+`from_addr()` will be handed a 5-tuple.
+
+Handle it inside `TIPCAddress.from_addr()` — do **not** patch
+`_server.py`:
+
+```python
+@classmethod
+def from_addr(cls, addr) -> TIPCAddress:
+    match addr:
+        # our own unwrapped form
+        case (str() as tag, int() as inst) if tag.startswith('tipc:'):
+            _, stype, scope = tag.split(':')
+            return TIPCAddress(int(stype), inst, int(scope))
+
+        # a kernel-observed TIPC_ADDR_ID 5-tuple: keep the
+        # *service* identity we already know and only annotate
+        # the observed port-id.
+        case (int() as atype, *rest) if atype == socket.TIPC_ADDR_ID:
+            ...
+```
+
+The `TIPC_ADDR_ID` case cannot reconstruct `(stype, instance)`
+— that info isn't in a port id. So `from_addr()` alone is
+insufficient for the reconciliation path. **Resolution**: make
+`from_addr()` raise a clear `ValueError` for the bare
+`TIPC_ADDR_ID` case, and instead prevent the reconciliation
+from firing by having `start_listener()` return a listener
+whose `getsockname()` we never need — i.e. land this two-line
+upstream fix in `_server.py:664`:
+
+```python
+if (
+    (unwrapped := lstnr.socket.getsockname()) != self.addr.unwrap()
+    and
+    self.addr.rebind_from_sockname   # ClassVar[bool] = True on tcp/uds
+):
+```
+
+with `TIPCAddress.rebind_from_sockname: ClassVar[bool] = False`
+(and `True` on `TCPAddress`/`UDSAddress`, preserving today's
+behaviour exactly). Rationale: the reconciliation exists *only*
+to learn the kernel-assigned port for `port=0` TCP binds (its
+own comment says so, `_server.py:662`); TIPC has no such
+late-binding, so opting out is semantically right rather than a
+hack. **Land this as its own commit, ahead of the backend**,
+with a test that `tcp`'s `port=0` behaviour is unchanged.
+
+Keep the observed port-id available anyway: annotate
+`ep.addr = ep.addr.with_port_id(*getsockname()[1:3])` (a pure
+`msgspec.structs.replace()` helper) purely for logging/repr.
+
+### 3.3 `MsgpackTIPCStream`
+
+```python
+class MsgpackTIPCStream(MsgpackTransport):
+    address_type = TIPCAddress
+    layer_key: int = 4
+
+    @property
+    def maddr(self) -> Multiaddr|str:
+        return mk_maddr(self.raddr)
+
+    def connected(self) -> bool:
+        return self.stream.socket.fileno() != -1
+
+    @classmethod
+    async def connect_to(
+        cls,
+        destaddr: TIPCAddress,
+        prefix_size: int = 4,
+        codec: MsgCodec|None = None,
+        **kwargs,
+    ) -> MsgpackTIPCStream:
+        sock = trio.socket.socket(AF_TIPC, SOCK_STREAM)
+        with close_on_error(sock):
+            # NOTE: connect by *name* -> kernel does the lookup,
+            # so this is our "discovery" call.
+            await sock.connect((
+                socket.TIPC_ADDR_NAME,
+                destaddr._stype,
+                destaddr._instance,
+                0,              # domain: 0 == "anywhere in scope"
+                destaddr._scope,
+            ))
+        return cls(
+            trio.SocketStream(sock),
+            prefix_size=prefix_size,
+            codec=codec,
+        )
+```
+
+- reuse `trio._highlevel_open_unix_stream.close_on_error` (the
+  UDS backend already imports it) or inline the equivalent
+  `try/except: sock.close(); raise`.
+- `SO_/TIPC_` opts worth setting and documenting:
+  - `setsockopt(SOL_TIPC, TIPC_IMPORTANCE, TIPC_HIGH_IMPORTANCE)`
+    for the *parent<->child* lifetime channel — this is a real
+    win TIPC gives us that TCP can't: the runtime's
+    supervision channel can outrank bulk app traffic under
+    congestion. Wire it as a `connect_to(..., importance=...)`
+    kwarg defaulted from a module constant, and have
+    `_runtime.py`'s parent-chan path pass the high value **in a
+    follow-up** (don't couple it to this PR).
+  - `TIPC_CONN_TIMEOUT` — the kernel-side connect timeout;
+    leave at default, we have `trio` cancel scopes.
+  - `TIPC_DEST_DROPPABLE = 0` on the connection so undeliverable
+    msgs come back as errors rather than being silently dropped.
+- **`connect_to()` on a name with no publisher**: TIPC returns
+  `ECONNREFUSED`/`EHOSTUNREACH` promptly (no SYN-timeout wait),
+  which is *better* discovery-ping behaviour than TCP. Confirm
+  the errno and make sure it surfaces as `ConnectionError`
+  (contract §4 — the registrar ping path depends on it).
+
+### 3.4 `get_stream_addrs()`
+
+```python
+@classmethod
+def get_stream_addrs(cls, stream) -> tuple[TIPCAddress, TIPCAddress]:
+    sock = stream.socket
+    # both return TIPC_ADDR_ID 5-tuples for a connected sock
+    l_id = sock.getsockname()
+    r_id = sock.getpeername()
+    ...
+```
+
+Problem: neither end's port-id tells us the *service name*. The
+`laddr`/`raddr` are used for logging, `Channel.raddr`,
+`Server._peers` keying-adjacent repr, and `maddr`. Design:
+
+- the **connecting** side knows the destaddr it dialled →
+  `connect_to()` overrides `_raddr` after construction with the
+  known-good `TIPCAddress`, exactly as
+  `MsgpackUDSStream.connect_to()` does for the peer-pid case
+  (`_uds.py:539-543`).
+- the **accepting** side does not know the peer's service name
+  from the socket. Two honest options:
+  - **(a) accept it: `raddr` carries only `(node, ref)`** via
+    `maybe_node`/`maybe_ref`, `_stype/_instance` set to a
+    sentinel `-1`, and `__repr__` renders
+    `TIPCAddress[<peer-node:0x...>:<ref>]`. The `Aid` from the
+    handshake already gives us the peer's logical identity, so
+    nothing in the runtime actually *needs* the peer's service
+    name. **Recommended.**
+  - (b) piggyback the peer's own bound name in the handshake.
+    Rejected for this PR: touches `Aid`/msg-spec.
+- `laddr` on the accepting side: the `Endpoint` knows its own
+  `addr`; but `get_stream_addrs()` is a `@classmethod` with only
+  the stream. Use `TIPC_ADDR_ID` for `laddr` too and let
+  `Endpoint.peer_tpts` keying (which is by *peer* addr) still
+  work. Verify nothing asserts `laddr == ep.addr` — grep for
+  `.laddr` uses before committing (`_server.py`'s
+  `con_status` logging, `Channel.pformat()`).
+
+---
+
+## 4. Multiaddr representation
+
+There is no `/tipc` in the multiaddr protocol table. Interim
+grammar, mirroring how `uds` maps to the spec-legal `/unix`:
+
+```
+/tipc/<stype>/<instance>            # scope implied = cluster
+/tipc/<stype>/<instance>/<scope>    # explicit
+```
+
+- `_tpt_proto_to_maddr['tipc'] = 'tipc'` and a `mk_maddr()`
+  `case 'tipc':` building the above.
+- `parse_maddr()` gets `case ['tipc']:` — but note
+  `py-multiaddr` will reject an unregistered protocol name
+  outright, so this **requires an upstream registration** (same
+  track as the `wg` work, gh #483 /
+  multiformats/py-multiaddr#107). Until that lands:
+  - `MsgpackTIPCStream.maddr` returns the **`str`** form (the
+    `MsgTransport.maddr` return type is already
+    `Multiaddr|str`, and `MsgpackUDSStream.maddr` already
+    exercises the `str` branch), and
+  - `parse_maddr()` special-cases the `/tipc/` prefix *before*
+    handing the string to `Multiaddr()`.
+  Document this as the reason gh #443's "standardize on
+  returning `Multiaddr` everywhere" item stays blocked.
+
+Propose `/tipc/` upstream as: name `tipc`, code TBD, size
+variable, value `<stype>:<instance>:<scope>` — or as three
+composed protos. Prefer *one* proto with a structured value so
+the maddr stays 2-segment like `/unix/...`.
+
+---
+
+## 5. Discovery: the actually-interesting part
+
+Two independently-shippable layers. **Layer A is in scope for
+the first PR; layer B is a fast-follow.**
+
+### 5.1 Layer A — "discovery by bind" (free)
+
+Because `bind(TIPC_ADDR_NAMESEQ)` publishes and
+`connect(TIPC_ADDR_NAME)` resolves, a `tractor` tree whose
+`registry_addrs` are TIPC service names needs **no registrar
+liveness at all** for the connect path: `find_actor()`'s
+"connect to the registrar and ask" becomes "connect to the
+service name directly". Concretely:
+
+- `tractor.discovery._api.find_actor()` etc. keep working
+  unchanged (they go through the registrar), *and*
+- a new, TIPC-only fast path becomes possible: derive an actor's
+  service name from its `(name, uuid)` and dial it without any
+  registrar hop.
+
+Do **not** build the fast path in PR 1. Instead, prove the
+property with a test (§7.4) and file the follow-up: it changes
+`discovery` semantics (name→instance derivation must be a
+documented, stable, cross-language-able hash) and deserves its
+own design.
+
+### 5.2 Layer B — the topology service (`TIPC_TOP_SRV`)
+
+This is what makes #378's "end game cluster proto" claim real:
+a *subscription* to name-table events, i.e. push-based
+`register`/`deregister` for free, replacing the registrar's
+polled `find_actor()`.
+
+Mechanics (verify each field against
+`linux/include/uapi/linux/tipc.h` + `net/tipc/topsrv.c` at
+implementation time — the struct layout below is from the uapi
+header and the byte-order caveat is real):
+
+```python
+# SOCK_SEQPACKET connected to the topology server
+sock = trio.socket.socket(AF_TIPC, SOCK_SEQPACKET)
+await sock.connect((
+    socket.TIPC_ADDR_NAME,
+    socket.TIPC_TOP_SRV,   # == 1
+    socket.TIPC_TOP_SRV,
+    0,
+))
+
+# struct tipc_subscr {
+#   struct tipc_name_seq seq;  /* 3 * __u32: type, lower, upper */
+#   __u32 timeout;             /* TIPC_WAIT_FOREVER == ~0 */
+#   __u32 filter;              /* TIPC_SUB_{PORTS,SERVICE,CANCEL} */
+#   char  usr_handle[8];
+# }                            /* == 28 bytes */
+_SUBSCR_FMT: str = '=IIIII8s'   # ⚠ 5*I is 20 -> use '=5I8s'
+```
+
+- **byte order**: the topology server historically accepts both
+  host and swapped order and auto-detects; modern kernels are
+  strict-ish. Pack native (`'='`) first, and if the server
+  closes the connection immediately, retry with `'>'`. Encode
+  that as a one-time probe helper
+  `_detect_topsrv_endianness()` cached at module level — and
+  put a `# ?TODO` pointing at `net/tipc/topsrv.c` for someone
+  to make it deterministic.
+- **events**: `struct tipc_event` is `event: u32`,
+  `found_lower: u32`, `found_upper: u32`,
+  `port: {ref: u32, node: u32}`, then the 28-byte subscription
+  echo → 40 bytes. `event ∈ {TIPC_PUBLISHED, TIPC_WITHDRAWN,
+  TIPC_SUBSCR_TIMEOUT}`.
+- **trio shape** — this is where the "nearly-functional,
+  modern-async" style pays off; expose it as an `@acm` yielding
+  a `trio` receive-channel of typed events, *not* a class:
+
+```python
+@acm
+async def open_topology_events(
+    stype: int = TRACTOR_STYPE,
+    lower: int = 0,
+    upper: int = 0xFFFFFFFF,
+    filter: int = TIPC_SUB_SERVICE,
+    timeout: int = TIPC_WAIT_FOREVER,
+    buf_size: int = 64,
+) -> AsyncGenerator[
+    trio.MemoryReceiveChannel[TIPCNameEvent],
+    None,
+]:
+    ...
+```
+
+  with `TIPCNameEvent(msgspec.Struct, frozen=True)` fields
+  `kind: Literal['published','withdrawn','timeout']`,
+  `addr: TIPCAddress`, `node: int`, `ref: int`. One
+  `trio.lowlevel`-free implementation: a nursery-spawned reader
+  task doing `await sock.recv(40)` in a loop and
+  `send_nowait()`ing decoded events, with the `@acm` closing the
+  socket on exit → reader gets `ClosedResourceError` → cancel
+  scope collapses. Standard `tractor` `@acm` discipline.
+- **consumer**: `tractor/discovery/_registry.py` gains an
+  optional "watch" mode so a registrar (or any actor) can keep
+  a live view of the actor set without polling. Sketch the
+  integration in the follow-up issue; do not wire it in PR 1.
+- **`SOCK_SEQPACKET` is fine here** because this socket never
+  goes through `MsgpackTransport` — it's a plain trio socket
+  used with `recv()`. The contract's "`SOCK_STREAM` only"
+  constraint applies to `MsgTransport` streams, not to this.
+
+---
+
+## 6. Commit sequencing (each independently reviewable + green)
+
+1. `_server.py`: add `Address.rebind_from_sockname:
+   ClassVar[bool]`, gate the `getsockname()` reconciliation on
+   it, `True` for tcp/uds. Test: tcp `port=0` unchanged.
+2. `tractor/ipc/_tipc.py`: `TIPCAddress` + `is_tipc_available()`
+   predicate + `start_listener()`. No transport yet.
+   Tests: address round-trip (`unwrap`/`from_addr`/`wrap_address`),
+   `get_random()` uniqueness, bind/listen + `SO_ACCEPTCONN`
+   tolerance, `EAFNOSUPPORT` → actionable `ConnectionError`.
+3. `MsgpackTIPCStream` + `connect_to()` + `get_stream_addrs()`.
+   Test: two `trio` tasks in one proc exchange a msg over
+   `Msgpack` framing (no `tractor` runtime).
+4. registration tables (contract §2 items 1-6, 9) +
+   `pyproject.toml` mark/extra. Test: full suite under
+   `--tpt-proto tipc` (§7.3).
+5. maddr support (`str` form + prefix special-case) + docs.
+6. `open_topology_events()` @acm + its tests (layer B).
+7. docs page + `docs/` example.
+
+Per project convention, a reproducing/guard test lands in its
+own commit **before** the fix it guards.
+
+---
+
+## 7. Testing
+
+### 7.1 the capability predicate (in `_tipc.py`, public)
+
+```python
+def is_tipc_available() -> bool:
+    '''
+    True iff this kernel can create an `AF_TIPC` socket, i.e.
+    the `tipc` module is loaded.
+
+    '''
+    try:
+        socket.socket(socket.AF_TIPC, socket.SOCK_STREAM).close()
+        return True
+    except OSError:
+        return False
+```
+
+Cache it in a module global (it can't change without a
+`modprobe`, and a cold call costs a syscall). Pure predicate, no
+side effects, no logging.
+
+### 7.2 gating
+
+- `pytest.mark.tipc` registered in `pyproject.toml`.
+- module-level
+  `pytestmark = pytest.mark.skipif(not is_tipc_available(),
+  reason='`tipc` kernel module not loaded (`modprobe tipc`)')`
+  in `tests/ipc/test_tipc.py`.
+- `--tpt-proto tipc` with no module must fail **loudly and
+  early** with the actionable message, not with 400 confusing
+  timeouts. Add the check to the `tpt_protos` fixture's existing
+  per-proto validation loop (`_testing/pytest.py:795`): if the
+  chosen `Address` type exposes an `is_available()`-style
+  classmethod, call it and `pytest.fail()` with its reason.
+  Generalize (don't special-case tipc) — plans 02/03 need the
+  same hook.
+
+### 7.3 CI
+
+- add a job matrix entry `--tpt-proto tipc` that runs
+  `sudo modprobe tipc` in a `before` step. GH's
+  `ubuntu-latest` runners do allow `modprobe tipc` (the module
+  ships with the standard Ubuntu kernel package); verify in a
+  throwaway workflow before wiring the matrix. If it turns out
+  to be unavailable, fall back to a container job with
+  `--privileged`/`--cap-add NET_ADMIN`, and mark the job
+  `continue-on-error` until it's proven stable.
+- cross-node TIPC (bearer) cannot be CI'd; cover it with a
+  documented manual smoke test in the docs page, in the style
+  of gh #482's LAN examples.
+
+### 7.4 backend-specific tests worth writing
+
+- **name-publication is discovery**: bind a listener on
+  `(stype, inst)`, then from a second task `connect()` by name
+  and assert it lands — *without* any `tractor` registrar.
+- **`get_random()` collision resistance**: 10k `get_random()`
+  calls with no live runtime → 10k distinct `_instance`s.
+  (This is the silent-crosstalk risk from §2.3; if the 4-byte
+  digest ever collides in this test, escalate to §9.)
+- **round-robin surprise**: two listeners bound to the *same*
+  `(stype, inst)` both succeed (TIPC allows it) and connects
+  distribute. Assert the observed behaviour and reference it
+  from the `get_random()` docstring so the next reader knows
+  why the hash matters.
+- **scope isolation**: a `TIPC_NODE_SCOPE` bind is not visible
+  to a cluster-scope lookup from another node (manual/marked).
+- **importance opt** round-trips via `getsockopt`.
+- **graceful + abrupt close** produce `TransportClosed` with the
+  same `loglevel` classification as tcp/uds — i.e. re-run the
+  relevant `tests/ipc/test_each_tpt.py` cases parametrized over
+  the new proto rather than writing new ones.
+
+---
+
+## 8. Deployment / docs deliverable
+
+A `docs/` page (and/or an `examples/` script) covering:
+
+```bash
+# single host, node-scope only
+sudo modprobe tipc
+tipc node get addr
+
+# multi-host over ethernet (pairs beautifully with plan 03's wg)
+sudo tipc bearer enable media eth device eth0
+# ...or over UDP when L2 isn't available:
+sudo tipc bearer enable media udp name uc localip 10.0.11.1
+tipc link list
+tipc nametable show          # <- see tractor's published services!
+```
+
+`tipc nametable show` displaying live `tractor` actors is the
+single best demo this backend has; lead with it.
+
+---
+
+## 9. Known risks + escalations
+
+| risk | mitigation |
+| --- | --- |
+| `_instance` hash collision → silent crosstalk (two actors share a service name, TIPC round-robins connects between them) | §7.4 test; if it bites, add a post-bind verification handshake, or bump to a 6-byte digest folded into `(stype_low, instance)` |
+| kernel/module unavailability everywhere (dev boxes, macOS, CI) | hard gating (§7.2); TIPC is explicitly an *opt-in cluster* transport, never a default |
+| `getsockname()` returns port-id not name | the `rebind_from_sockname` opt-out (§3.2), landed first |
+| unregistered `/tipc` multiaddr proto | `str` maddr fallback (§4) + upstream track gh #483 |
+| stale docs (#378 notes tipc.io docs may be out of date) | treat `include/uapi/linux/tipc.h` + `net/tipc/` as the only normative source; cite file+symbol in code comments |
+| `SOCK_SEQPACKET` topology framing byte-order | probe helper + `?TODO` (§5.2) |
+
+## 10. Follow-up issue seeds
+
+- **register `/tipc` in the multiaddr spec**, mirroring the `wg`
+  track (multiformats/py-multiaddr#107/#108 + gh #483). Same
+  shape of work: propose the proto + code, land a codec in
+  `py-multiaddr`, then drop our `str`-maddr fallback (§4). Worth
+  filing *alongside* the `wg` spec-submission issue so both
+  proposals go up together rather than as one-offs.
+- registrar-less discovery fast path via name derivation (§5.1)
+- `TIPC_TOP_SRV`-driven push registry in
+  `discovery/_registry.py` (§5.2)
+- `TIPC_IMPORTANCE` for the parent<->child lifetime channel
+  (§3.3) — genuinely novel supervision QoS, no other backend
+  can do it
+- TIPC multicast / group messaging as a *broadcast* transport
+  for `tractor.trionics` fan-out (explicitly not `MsgTransport`)
+- dual-link resiliency / multi-homing (#378's "hybrid dual link")
+  once bearers are scripted in the docs

ai/tpt-backends/02_quic_iroh_backend.md

@@ -0,0 +1,566 @@
+# Plan 02 — QUIC backend via `iroh` FFI, uniffi-async rewritten onto `trio`
+
+Tracks gh [#353]. Prereq reading:
+[`00_shared_backend_contract.md`](./00_shared_backend_contract.md).
+
+**Thesis**: the value of `iroh` over "just QUIC" is
+`NodeId`-addressed, NAT-traversing, relay-fallback endpoints —
+i.e. a `tractor` actor tree that spans hosts *without* a
+reachable listening socket. The cost is that `iroh`'s python
+surface is `uniffi`-generated **asyncio** and its listener is not
+a socket. This plan spends its complexity budget in exactly two
+places: a `trio`-native uniffi future bridge, and a
+`trio.abc.Listener`/`Stream` adapter pair. Everything else is
+contract boilerplate.
+
+[#353]: https://github.com/goodboy/tractor/issues/353
+
+---
+
+## 1. Library selection (decided, with the rejected alternatives)
+
+**Chosen: `iroh` (PyPI, from `n0-computer/iroh-ffi`), pinned to
+a single minor.** The `iroh` python package is a `uniffi`
+binding over the rust `iroh` crate (QUIC via `quinn`/`noq`).
+
+Rejected, and why — record these so the next implementer doesn't
+relitigate:
+
+- **`aioquic`** (sans-io + asyncio): genuinely trio-portable
+  (`hypercorn` already pairs its sans-io core with a trio UDP
+  server, see the links in #353) and dependency-light. But it
+  gives us *only* QUIC — no NodeId identity, no hole punching,
+  no relay. We'd be reimplementing iroh's whole reason for
+  existing. **Keep as the documented fallback** if the FFI
+  bridge (§2) proves unmaintainable; the `MsgTransport` and
+  `Listener` adapters from §3 are ~90% reusable against an
+  `aioquic` core, which is a deliberate design property of this
+  plan.
+- **`quiche` / `quinn` via a hand-rolled PyO3 ext**: strictly
+  more work than reusing `iroh-ffi`, and puts us in the
+  build-wheels business.
+- **`trio-asyncio`**: viable *shortcut* to run the asyncio-shaped
+  bindings under trio, and `tractor` already ships
+  infected-asyncio machinery (`tractor.to_asyncio`,
+  `tests/test_infected_asyncio.py`). Rejected as the *primary*
+  design because it makes every IPC send/recv cross a
+  loop-boundary shim in the hot path, and because #353 asks
+  explicitly for the asyncio support to be "rewritten for trio".
+  **But**: build it first as the throwaway spike (§6 step 0) to
+  de-risk the iroh API surface before writing the bridge.
+
+Version pinning: `iroh` moves fast and has had breaking
+API renames across minors. Pin `iroh>=X.Y,<X.Y+1` in a `quic`
+extra, and **write down the exact resolved version + the
+generated `iroh/_uniffi*` module layout** in the module
+docstring, because §2 depends on generated-code internals.
+
+**Step 0 of implementation is an API-truth pass**: install the
+pinned `iroh`, `python -c "import iroh; help(iroh)"`, and record
+in this doc's §1.1 the real names of: endpoint builder, secret
+key type, `connect`/`accept`, bi-stream open/accept, the
+send/recv methods and their exact signatures/return types, and
+whether they're `async def`. Everything below uses *provisional*
+names and must be reconciled. Do not skip this; do not guess
+from memory.
+
+### 1.1 API-truth table (fill in during step 0)
+
+| concept | provisional name | actual (fill in) |
+| --- | --- | --- |
+| secret key | `iroh.SecretKey.generate()` | |
+| endpoint builder | `iroh.Endpoint.builder(...).bind()` | |
+| node id | `endpoint.node_id() -> str` | |
+| node addr (relay + direct) | `iroh.NodeAddr` | |
+| dial | `await endpoint.connect(node_addr, alpn)` | |
+| accept conn | `await endpoint.accept()` | |
+| open bi-stream | `await conn.open_bi()` | |
+| accept bi-stream | `await conn.accept_bi()` | |
+| send | `await send_stream.write_all(b)` | |
+| recv | `await recv_stream.read(n) -> bytes\|None` | |
+| half-close | `await send_stream.finish()` | |
+
+---
+
+## 2. The `trio`-native uniffi future bridge (`tractor/ipc/_uniffi_trio.py`)
+
+### 2.1 what uniffi actually generates
+
+`uniffi`'s async support does not use asyncio *semantically* —
+it uses asyncio only as the *executor* for a poll loop. The
+generated python for an `async fn` is, in shape:
+
+1. call `_uniffi_..._<method>(...)` → returns an opaque
+   `RustFuture` handle (a `void*`/`u64`).
+2. loop: call
+   `ffi_..._rust_future_poll_<T>(handle, callback, callback_data)`.
+   The callback is a C-ABI fn pointer invoked **from an
+   arbitrary rust thread** with a poll-result code
+   (`READY`/`MAYBE_READY`).
+3. the generated glue's callback resolves an
+   `asyncio.Future` via `loop.call_soon_threadsafe(...)`; the
+   coroutine awaits it, then re-polls.
+4. on ready: `ffi_..._rust_future_complete_<T>(handle,
+   &call_status)` → the value; then
+   `ffi_..._rust_future_free_<T>(handle)`.
+
+**The asyncio dependency is confined to step 3.** That is the
+whole insight: the bridge is ~40 lines.
+
+### 2.2 the trio version
+
+```python
+async def await_rust_future(
+    poll: Callable,      # ffi_..._rust_future_poll_<T>
+    complete: Callable,  # ffi_..._rust_future_complete_<T>
+    free: Callable,      # ffi_..._rust_future_free_<T>
+    handle: int,
+    lift: Callable[[Any], Any],
+) -> Any:
+    '''
+    Drive a `uniffi` rust-future to completion on the current
+    `trio` task, bridging rust-thread wakeups via
+    `TrioToken.run_sync_soon()`.
+
+    '''
+    token = trio.lowlevel.current_trio_token()
+    while True:
+        wake = trio.Event()
+        # NOTE, invoked from a *rust* thread!
+        def _cb(_data, poll_code):
+            token.run_sync_soon(wake.set)
+
+        cb = _UNIFFI_FUTURE_CALLBACK(_cb)   # keep a strong ref!
+        poll(handle, cb, 0)
+        await wake.wait()
+        if <poll_code was READY>:
+            break
+    try:
+        status = _UniffiRustCallStatus.default()
+        res = complete(handle, status)
+        _uniffi_check_call_status(status)   # reuse generated helper
+        return lift(res)
+    finally:
+        free(handle)
+```
+
+Critical details, each a real bug if missed:
+
+- **`token.run_sync_soon()` is the only trio API callable from a
+  foreign thread**, and it is documented as such. Use it; do
+  *not* use `trio.from_thread.run_sync` (requires a trio thread
+  context) and do not touch the `Event` directly from the
+  callback.
+- **the poll code must reach the trio side.** Capture it in a
+  `nonlocal`/1-slot list written by the callback *before*
+  `run_sync_soon`, since the callback owns the value. Handle
+  `MAYBE_READY` by re-polling (the loop above does).
+- **keep the `ctypes` callback object alive** across the await —
+  a GC'd `CFUNCTYPE` trampoline is a segfault. Bind it to a
+  local *and* make sure the local outlives the `poll()` call
+  window.
+- **cancellation.** `await wake.wait()` is a trio checkpoint, so
+  a `Cancelled` can fire while rust still owns the future. On
+  cancel we must still `free(handle)` — and per uniffi, the
+  correct sequence is to call the generated
+  `ffi_..._rust_future_cancel_<T>(handle)` then continue
+  polling to completion before `free`. Wrap the whole thing so
+  the cancel path does:
+  `with trio.CancelScope(shield=True): cancel(handle); <drain
+  poll loop>; free(handle)`. **Bounded** shield (add a
+  `trio.move_on_after()` with a module-level constant) so a
+  wedged rust future can't make an actor un-cancellable —
+  `tractor` is SC-first and an unbounded shield here would
+  violate that.
+- **`trio.lowlevel.current_trio_token()`** must be captured on
+  the trio side (not in the callback).
+
+### 2.3 how to apply it to the generated bindings
+
+Do **not** fork/vendor the generated `iroh` python. Instead ship
+a *narrow* re-dispatch shim:
+
+- write `tractor/ipc/_uniffi_trio.py` with `await_rust_future()`
+  plus a `@cm patch_uniffi_for_trio()` that monkey-patches the
+  generated module's single async-driver entrypoint (in current
+  uniffi that's `_uniffi_rust_call_async` / `_rust_call_async`,
+  one function) to the trio implementation.
+- verify at import time that the expected symbol exists and
+  raise a clear, actionable error naming the pinned `iroh`
+  version if not. A silent fallback to asyncio would be a
+  nightmare to debug.
+- **plan for this to break on `iroh`/`uniffi` upgrades.** Mitigate
+  with (a) a unit test that drives one trivial `iroh` async call
+  under bare `trio.run()` and asserts no event loop was ever
+  created (`asyncio.get_event_loop_policy()` untouched /
+  `asyncio._get_running_loop() is None`), and (b) a docstring
+  pointing at the uniffi codegen template this mirrors.
+
+If step 0 reveals the generated code is *structurally* hostile
+to this (e.g. `asyncio` imported and used at module scope for
+more than the driver), fall back to option (b): run iroh under
+`tractor.to_asyncio` infected mode and open the follow-up to
+revisit. Say so in the PR rather than fighting it.
+
+---
+
+## 3. Mapping QUIC onto `MsgTransport`
+
+### 3.1 the layering decision
+
+QUIC natively multiplexes streams inside one connection. The
+mapping that preserves *all* existing `tractor` semantics with
+the least new code:
+
+```
+iroh Endpoint      ==  one per actor (process)     -> the "listener"
+iroh Connection    ==  one per peer actor           -> pooled
+iroh bi-stream     ==  one `Channel`/`MsgTransport`  -> 1:1
+```
+
+- keep the 4-byte `<I` length-prefix framing **unchanged**. It's
+  redundant-ish over a QUIC stream but it means
+  `MsgpackTransport` is reused verbatim, and framing is cheap.
+  Revisit only after it works.
+- **one-task-per-stream** falls out naturally, which is exactly
+  the #353 note about QUIC sub-stream QoS/cancellation fitting
+  `trio`.
+- `layer_key: int = 4` still (QUIC is L4-ish); note in a comment
+  that this backend is really 4+security+multiplex.
+
+**Connection pooling** is the one place we add state the other
+backends don't have: dialing the same peer twice should reuse
+the `Connection` and open a second bi-stream. Implement as a
+module-level `dict[NodeId, Connection]` guarded by a
+`trio.Lock`... **no** — that's a per-process cache with
+lifetime/teardown hazards. Instead reuse the codebase's existing
+idiom: `tractor.trionics.maybe_open_context()` keyed on the
+node-id, which already solves exactly this (one-cached-resource-
+per-key, refcounted, teardown-on-last-exit) and whose teardown
+semantics were just hardened (gh #488). Use it; do not hand-roll
+a cache. Anything concurrency-subtle here should get the
+`conc-anal` skill run over it.
+
+### 3.2 `IrohAddress`
+
+```python
+class IrohAddress(
+    msgspec.Struct,
+    frozen=True,
+):
+    _node_id: str                 # 32B ed25519 pubkey, hex or z32
+    _alpn: str = 'tractor/0'      # the bindspace!
+    # optional dial hints; NOT part of identity
+    maybe_relay_url: str|None = None
+    maybe_direct_addrs: tuple[str, ...] = ()
+
+    proto_key: ClassVar[str] = 'iroh'   # ?or 'quic'; see §3.2.1
+    unwrapped_type: ClassVar[type] = tuple[str, str]
+    def_bindspace: ClassVar[str] = 'tractor/0'
+```
+
+- **`.unwrap() -> (node_id_str, alpn_str)`** — a `(str, str)`
+  tuple, which is *unambiguously distinct* from
+  `TCPAddress`'s `(str, int)`. But careful:
+  `wrap_address()`'s UDS case is
+  `case (_, filename) if type(filename) is str` — which
+  **already catches `(str, str)`**. So the iroh `case` MUST be
+  ordered *before* the UDS case and guarded, e.g.
+  `case (str() as nid, str() as alpn) if _is_node_id(nid):`
+  with `_is_node_id()` a cheap length+alphabet check. Add a
+  regression test asserting a UDS `(dir, filename)` pair still
+  wraps to `UDSAddress` — this is the exact "wrong transport
+  loaded" hazard `_addr.py:214` warns about.
+- `.bindspace` → `self._alpn`. This is the honest analogue:
+  the ALPN is the set of endpoints willing to talk to you, and
+  two `tractor` deployments sharing an iroh network are
+  separated by ALPN exactly as two UDS deployments are
+  separated by directory. Include a `tractor` version/proto
+  epoch in the default ALPN so incompatible runtimes can't
+  handshake.
+- `.is_valid` → node-id parses, alpn non-empty.
+- **`get_root()` is the hard one.** There is no
+  well-known-port analogue: an iroh node id is a *keypair*, so
+  "the host's default registrar addr" requires a *persisted
+  secret key*. Design:
+  - the root/registrar's secret key lives at
+    `get_rt_dir() / 'iroh_registrar.key'` (0600), created on
+    first use.
+  - `get_root()` must stay **pure and import-time-safe**
+    (contract §2.3: `_default_lo_addrs` is built at import!).
+    So `get_root()` *reads* the key file if present and
+    otherwise returns an `IrohAddress` with
+    `_node_id=''`/sentinel, and the **generation** happens in
+    an explicit sibling — `ensure_registrar_key() ->
+    IrohAddress` — called from the listen path. Pure getter,
+    explicit setter; do not smuggle key generation into
+    `get_root()`.
+  - this almost certainly means `_default_lo_addrs` must become
+    lazy for this backend. **Land that refactor as its own prep
+    commit** (a `default_lo_addrs()` that computes per-call
+    instead of the import-time dict) — it also unblocks plan
+    03's netns-scoped defaults.
+- `get_random()`: generate a fresh `SecretKey` per subactor and
+  return its node-id. Note this runs post-fork pre-listen
+  (contract §4) and costs an ed25519 keygen (~µs, fine). The
+  *secret* can't live in a frozen `Address`, so it must be
+  stashed where the listen path can find it: a module-level
+  `dict[node_id, SecretKey]` populated by `get_random()` and
+  consumed+popped by `start_listener()`. Ugly but honest;
+  document it and note the alternative (thread the key through
+  `Endpoint`) as a follow-up.
+
+#### 3.2.1 `proto_key`: `'iroh'` vs `'quic'`
+
+Use **`'quic'`** for the `proto_key`/`--tpt-proto` name and
+name the module `_quic.py`, with `iroh` as the *implementation*.
+Rationale: it keeps the door open for the `aioquic` fallback
+(§1) without a user-visible rename, and it matches how `uds` is
+a proto name rather than a lib name. Put `iroh`-specific bits
+behind an internal `_iroh` submodule if the file gets big.
+
+### 3.3 the `trio.abc` adapters — where the real work is
+
+Contract §3 says a non-socket backend needs three upstream
+generalizations. Land them **as a prep PR, before any iroh
+code**, so they can be reviewed on their own merits with
+tcp/uds still the only backends:
+
+1. **`Endpoint.start_listener()` must not assume
+   `.socket.getsockname()`.** Use the same
+   `Address.rebind_from_sockname: ClassVar[bool]` gate that
+   plan 01 §3.2 introduces — coordinate so it lands once. (If
+   plan 01 lands first, this is free.)
+2. **`transport_from_stream()` (`_types.py:92`) must not assume
+   `trio.SocketStream`.** Replace the `sock.family` match with:
+   check `isinstance(stream, trio.SocketStream)` → existing
+   family match; else look for a
+   `stream.tpt_key: ClassVar[MsgTransportKey]` attribute on the
+   adapter and use it. Keeps the existing path byte-identical
+   and makes new stream types self-describing (a much better
+   shape than growing an `isinstance` ladder).
+3. **type annotations**: `handle_stream_from_peer(stream:
+   trio.SocketStream)` → `trio.abc.Stream`; `Endpoint._listener:
+   SocketListener|None` → `trio.abc.Listener|None`;
+   `MsgTransport.stream: trio.SocketStream` →
+   `trio.abc.Stream`. Annotation-only, zero behaviour change.
+
+Then the adapters:
+
+```python
+class QuicMsgStream(trio.abc.HalfCloseableStream):
+    '''
+    A single `iroh` bi-directional QUIC stream presented as
+    a `trio` byte-stream so `MsgpackTransport` can frame over
+    it unmodified.
+
+    '''
+    tpt_key: ClassVar[MsgTransportKey] = ('msgpack', 'quic')
+
+    def __init__(self, conn, send, recv) -> None: ...
+    async def send_all(self, data: bytes) -> None: ...
+    async def wait_send_all_might_not_block(self) -> None: ...
+    async def receive_some(self, max_bytes: int|None = None) -> bytes: ...
+    async def send_eof(self) -> None: ...
+    async def aclose(self) -> None: ...
+```
+
+Non-negotiable behaviours (each maps to a `match` case that
+already exists in `_transport.py` and must keep working):
+
+- `receive_some()` returns `b''` at clean EOF →
+  `MsgpackTransport._iter_packets()` sees `header == b''` and
+  raises `TransportClosed(loglevel='transport')`. **This is the
+  graceful-disconnect path the whole runtime relies on**; get it
+  right first.
+- a reset/aborted stream → raise `trio.BrokenResourceError`.
+- use after local close → raise `trio.ClosedResourceError`
+  (ideally with `'another task closed this fd'`-equivalent text
+  absent, so the `raise_on_report` branch at
+  `_transport.py:290` stays quiet).
+- `send_all()` on a closed peer → `trio.BrokenResourceError`.
+- honour `trio`'s one-task-per-direction rule: guard with
+  `trio._util.ConflictDetector` equivalents (or just document +
+  assert), because `MsgpackTransport` already serializes sends
+  with a `StrictFIFOLock` but recvs are single-task by
+  construction.
+- **buffering**: if iroh's `read()` doesn't support
+  "read up to n", `receive_some()` must maintain an internal
+  leftover buffer. Note `MsgpackTransport` wraps us in
+  `tricycle.BufferedReceiveStream` anyway, so `receive_some()`
+  just needs *some* nonzero-progress contract.
+
+```python
+class QuicListener(trio.abc.Listener):
+    '''
+    Accepts iroh `Connection`s and yields one `QuicMsgStream`
+    per accepted bi-stream, so `trio.serve_listeners()` spawns
+    one `handle_stream_from_peer()` per `Channel`.
+
+    '''
+    async def accept(self) -> QuicMsgStream: ...
+    async def aclose(self) -> None: ...
+```
+
+The accept-side subtlety: `trio.abc.Listener.accept()` yields
+one stream per call, but iroh gives us *connections* which then
+yield *streams*. So `QuicListener` needs an internal
+`trio.MemoryReceiveChannel[QuicMsgStream]` fed by a background
+task-pair (one task accepting connections, one per connection
+accepting bi-streams). `trio.abc.Listener` has no nursery, so:
+make the listener **constructed by an `@acm`** that owns the
+nursery, and have `start_listener()` be that `@acm`'s driver.
+
+⚠️ this collides with `Endpoint.start_listener()` being a plain
+`async def` returning a listener. Two options:
+- **(a)** hang the nursery off the `Endpoint`'s existing
+  `listen_tn` — `_serve_ipc_eps()` already creates `listen_tn`
+  and passes it into every `Endpoint` (`_server.py:1063-1074`),
+  and `Endpoint.listen_tn` is right there. So
+  `start_listener()` can `self.listen_tn.start_soon(...)` the
+  acceptor tasks. **Recommended**: no upstream signature change,
+  correct lifetime (dies with the ep group), and it's why
+  `listen_tn` is on the struct in the first place.
+- (b) change `start_listener()` to a `@acm`. Bigger blast
+  radius; only if (a) proves insufficient.
+
+Since `start_listener()` is called via
+`inspect.getmodule(addr)` with only `addr=` (contract §1.3),
+option (a) needs the `Endpoint` itself. Either add `ep=` to the
+module-level `start_listener()` call signature (all backends
+ignore it except quic → small upstream change, do it as part of
+the prep PR and make it keyword-only with a default) or have
+`QuicListener.accept()` lazily spawn via
+`trio.lowlevel.current_task().parent_nursery` (**rejected** —
+fragile, implicit). Do the explicit `ep=` kwarg.
+
+### 3.4 `maddr`
+
+Multiaddr already standardizes the pieces:
+
+```
+/ip4/<h>/udp/<p>/quic-v1                     # direct
+/ip4/<h>/udp/<p>/quic-v1/p2p/<node-id>       # direct + identity
+/dns/<relay-host>/tcp/443/tls/ws/p2p/<node>  # relay-ish
+```
+
+- primary form: `/p2p/<node-id>` alone is a legal maddr and is
+  the *only* required component for iroh dialling — relay +
+  direct addrs are discovery hints. So `mk_maddr()` emits
+  `/p2p/<node_id>` and, when known, prefixes the direct
+  `/ip4/../udp/../quic-v1/`.
+- `/p2p/` values are multihash-encoded peer ids; an iroh node-id
+  is a raw ed25519 key. Converting requires the identity
+  multihash + libp2p key protobuf wrapper. **Decide**: emit the
+  raw node-id under a *tractor-local* `/iroh/<node-id>` segment
+  (needs upstream registration, same track as `wg`/`tipc`,
+  gh #483) rather than pretending to be a libp2p peer-id we
+  can't round-trip. Return the `str` form until upstream lands
+  (`MsgTransport.maddr` is `Multiaddr|str`).
+- this backend is the strongest argument for gh #443's
+  **tunnelled/composed maddr** item: `/ip4/../udp/../quic-v1/..`
+  *is* a composed stack. Cross-reference plan 03 §5 so the two
+  grammars land compatibly.
+
+---
+
+## 4. Discovery integration
+
+- iroh's node-id addressing means the `tractor` registrar can
+  hold `IrohAddress`es that are **reachable from anywhere** with
+  no port-forwarding — that is the headline feature. The
+  registrar itself works unchanged.
+- iroh has its own discovery (DNS/pkarr/mdns). **Out of scope**;
+  note in the follow-up that `tractor.discovery` could
+  eventually delegate to it, which would be the direct analogue
+  of plan 01's TIPC-topology idea.
+- relay servers: default to n0's public relays for the demo,
+  document self-hosting (docs.iroh.computer's dedicated-infra
+  page is linked from #353), and make the relay set a
+  `start_listener()` kwarg.
+
+## 5. Security note
+
+QUIC is TLS-1.3-always and iroh authenticates by node-id, so
+this backend is the first `tractor` transport with real
+transport security and peer authentication. Two things follow:
+1. an **allowlist hook** — an actor should be able to reject
+   inbound connections from unknown node-ids *before* the
+   `Aid` handshake. Natural home: a predicate kwarg on
+   `start_listener()`, evaluated in `QuicListener`'s connection
+   acceptor task. Sketch it; ship it in PR 1 if cheap (it is).
+2. do **not** claim any security property for the other
+   backends by association. `tcp`/`uds`/`tipc` remain
+   unauthenticated; that's what plan 03 (wg) is for.
+
+## 6. Commit sequencing
+
+0. **spike (throwaway, not committed)**: drive iroh under
+   `trio-asyncio`/`tractor.to_asyncio`, echo bytes over a
+   bi-stream between two procs. Fills in §1.1. Timebox it.
+1. prep PR: annotation widening + `rebind_from_sockname` gate +
+   `transport_from_stream()` `tpt_key` dispatch + `ep=` kwarg on
+   `start_listener()` + lazy `default_lo_addrs()`. **No new
+   backend.** Full suite green on tcp *and* uds.
+2. `_uniffi_trio.py` + its tests (drive one iroh async call
+   under bare `trio.run()`; assert no asyncio loop; assert
+   cancellation frees the future).
+3. `QuicMsgStream` + tests against a *loopback* iroh endpoint
+   pair in one process (no `tractor` runtime): send/recv, clean
+   EOF → `b''`, reset → `BrokenResourceError`, use-after-close
+   → `ClosedResourceError`.
+4. `QuicListener` + `start_listener()` + `IrohAddress` +
+   key-file mgmt.
+5. `MsgpackQuicStream(MsgpackTransport)` + `connect_to()` +
+   `maybe_open_context()` connection pooling.
+6. registration tables + `--tpt-proto quic` + full suite.
+7. maddr + docs + a two-host example (pairs with #482's format).
+
+## 7. Testing
+
+- capability predicate `is_quic_available()` → `iroh` importable
+  *and* the uniffi driver symbol present at the pinned version.
+  Same `pytest.fail`-early hook as plan 01 §7.2.
+- **the acceptance bar is the same**: whole suite green under
+  `--tpt-proto quic`. Expect this to shake out real bugs in the
+  adapters (esp. teardown ordering and `TransportClosed`
+  classification) — that's the point.
+- expect to need **timeout headroom**: iroh endpoint bind +
+  first connect (relay discovery) is orders of magnitude slower
+  than a UDS bind. Before touching any test deadline, rule out
+  the CPU-throttle false-positive (see the project's
+  `env_cpu_throttle_masquerades_as_regression` note); then, if
+  real, add a per-proto timeout multiplier to the test harness
+  rather than editing individual tests.
+- a no-network test mode: iroh with relays disabled +
+  loopback direct addrs only, so CI doesn't depend on n0's
+  infra. **Make this the default in CI**; mark the relay tests
+  `pytest.mark.net` and keep them out of the default run.
+- leak checks: assert every `SecretKey`/`Endpoint` is closed on
+  actor teardown (an `Endpoint` left open holds UDP sockets and
+  relay connections; a leak here shows up as hung tests, not
+  errors).
+
+## 8. Risks
+
+| risk | mitigation |
+| --- | --- |
+| uniffi codegen internals shift on upgrade | pinned minor, symbol assertion at import, the "no asyncio loop" test, documented fallback to `to_asyncio` |
+| rust-thread callback → trio wakeup mishandled (segfault / lost wakeup / un-cancellable task) | strong ref on the ctypes trampoline; `run_sync_soon` only; **bounded** shielded cancel-drain; run the `conc-anal` skill over the bridge |
+| `iroh` wheel availability for 3.13/3.14 on linux+macos | verify in step 0; if missing, that alone may force the `aioquic` fallback |
+| QUIC latency/jitter destabilizes the existing suite's timing assumptions | per-proto timeout multiplier, relay-less CI mode |
+| `(str, str)` unwrapped form collides with UDS in `wrap_address()` | guarded case ordered first + explicit regression test (§3.2) |
+| scope creep into iroh's docs/blobs/gossip crates | this backend is `Endpoint`+`Connection`+bi-streams only; anything else is a separate issue |
+
+## 9. Follow-up issue seeds
+
+- `tractor.discovery` delegating to iroh discovery (DNS/pkarr/mdns)
+- per-`Context` QUIC sub-streams: today one `Channel` == one
+  stream; QUIC would let each `tractor.Context` own its own
+  stream with independent flow-control and cancellation — this
+  is the genuinely novel win #353 gestures at, and it's a
+  runtime-layer change, not a transport one
+- unreliable QUIC datagrams for a lossy-ok broadcast transport
+  (pairs with plan 01's TIPC-multicast seed)
+- node-id allowlist → a real `tractor` authz story
+- `aioquic` sans-io backend reusing §3's adapters

ai/tpt-backends/03_wg_tunnel_bindspace.md

@@ -0,0 +1,487 @@
+# Plan 03 — WireGuard (and other tunnels) as a *nested bindspace* via `pyroute2`
+
+Tracks gh [#482] + the tunnelled-maddr item of [#443].
+Prereq reading:
+[`00_shared_backend_contract.md`](./00_shared_backend_contract.md).
+
+**Thesis**: WireGuard is **not** a `MsgTransport`. It is an
+interface-layer tunnel that is transparent to `socket(2)`, so
+the correct abstraction is a *bindspace* — a scoped,
+`@acm`-managed network context that an existing L4 transport
+(`tcp`, and later `quic`/`tipc`-over-UDP-bearer) binds *inside*.
+This plan implements `Address.namespace` (spec'd but unused
+since day one) and the composed/tunnelled maddr grammar, with
+`pyroute2` as the netlink codec and as much of the I/O moved
+onto `trio` as the library's sans-io layer allows.
+
+[#482]: https://github.com/goodboy/tractor/issues/482
+[#443]: https://github.com/goodboy/tractor/issues/443
+
+---
+
+## 1. What exists today (verified, per #482)
+
+- `wrap_address()` accepts maddr `str`s (leading-`/` dispatch,
+  `_addr.py:262`) but `parse_maddr()` only knows
+  `/ip4|ip6/<h>/tcp/<p>` and `/unix/<p>`; a `.../wg/u<key>`
+  maddr raises `ValueError('Unsupported multiaddr protocol
+  combo')`.
+- there is no `wg` proto in the multiaddr *spec* yet, but
+  multiformats/py-multiaddr#108 (key form `u<base64url>`) is
+  **merged** as of 2026-07-28 (`f86519da`) — and unreleased, the
+  latest `0.2.0` predating it. Spec registration is still tracked
+  by multiformats/py-multiaddr#107 and gh #483.
+- so **today's deployable story is declarative**: run `wg-quick`
+  out-of-band, parse the maddr, strip to the overlay
+  `(host, port)`, verify the pubkey against the live tunnel,
+  hand the overlay addr to `registry_addrs=`/`tpt_bind_addrs=`.
+  #482 already contains working example code for exactly this.
+- `Address.namespace` exists in the Protocol
+  (`_addr.py:94-101`, "the if-available OS-specific network
+  namespace key") and **no backend implements it**. This plan is
+  its first consumer.
+
+## 2. Three layers, three PRs
+
+| layer | what | dep | ships |
+| --- | --- | --- | --- |
+| **A. declarative** | commit #482's examples; `parse_maddr()` learns `/wg/u<key>` → overlay `Address` + verified pubkey | `multiaddr` (already), `wg(8)` CLI | first |
+| **B. `pyroute2` read/verify** | replace the `subprocess.run(['sudo','wg','show'])` shelling with netlink queries | `pyroute2` extra | second |
+| **C. `@acm` lifecycle** | create/configure/tear down wg ifaces + netns *from the runtime*, as nested bindspaces; implement `Address.namespace` | `pyroute2` + `CAP_NET_ADMIN` | third |
+
+Each is independently valuable and independently reviewable.
+**Do not attempt C first** — the interesting design (nested
+bindspace `@acm`s) is only well-posed once A has pinned the
+address grammar and B has proven the netlink path under trio.
+
+---
+
+## 3. Layer A — declarative `wg` maddrs
+
+### 3.1 the address shape
+
+The decision: **a wg segment annotates an existing address, it
+does not create a new address type.** Two candidate encodings;
+**pick (a)**:
+
+- **(a) `TunnelledAddress` wrapper** (recommended):
+  ```python
+  class TunnelledAddress(
+      msgspec.Struct,
+      frozen=True,
+  ):
+      overlay: Address         # e.g. TCPAddress
+      tunnel: WGTunnelSpec     # proto-specific, frozen
+  ```
+  with `.proto_key` **delegating to `overlay.proto_key`** so every
+  existing table lookup (`_addr_to_transport`,
+  `enable_transports` guard at `_root.py:391`,
+  `transport_from_addr()`) keeps working untouched, and
+  `.unwrap()` delegating to `overlay.unwrap()` so **nothing new
+  crosses the wire**. `.namespace` and `.bindspace` come from
+  the tunnel spec. The wrapper is stripped (`→ .overlay`) at the
+  moment of bind/connect.
+  - ⚠️ `is_wrapped_addr()` (`_addr.py:194`) tests
+    `type(addr) in _address_types.values()` — a `bidict` of
+    proto_key→type. `TunnelledAddress` isn't in it and must not
+    be (it's not 1:1 with a proto). So either add an explicit
+    `isinstance(addr, TunnelledAddress)` clause there, or give
+    the wrapper a marker and test structurally. Do the former;
+    it's two lines and honest.
+  - the reflection in `Endpoint.start_listener()`
+    (`inspect.getmodule(self.addr)`) would resolve to the
+    *wrapper's* module, not the transport's. **So the wrapper
+    must be unwrapped before it reaches `Endpoint`** — i.e. by
+    the bindspace `@acm` (layer C) or by `parse_maddr()`
+    (layer A). State this loudly in the docstring; it's the #1
+    way to get this wrong.
+- (b) add fields to each existing `Address` type. Rejected:
+  duplicates tunnel logic per-backend and pollutes `.unwrap()`.
+
+```python
+class WGTunnelSpec(
+    msgspec.Struct,
+    frozen=True,
+):
+    peer_pubkey: str          # std-base64 `wg(8)` form
+    iface: str = 'wg0'
+    netns: str|None = None
+    # layer-C-only fields, unset in layer A
+    maybe_endpoint: tuple[str, int]|None = None
+    maybe_allowed_ips: tuple[str, ...] = ()
+```
+
+### 3.2 `parse_maddr()`/`mk_maddr()`
+
+Grammar — **verified** against py-multiaddr#108, first on the
+`baudco/py-multiaddr@wg_support` branch and re-verified after it
+merged upstream (`multiformats/py-multiaddr@f86519da`); all three
+forms below parse *and* round-trip. Note the codec also validates
+that the key decodes to exactly 32 bytes, so a truncated key is a
+`StringParseError`, not a silently-mangled parse:
+
+```
+/ip4/192.168.1.50/udp/51820/wg/u<A_pub>/ip4/10.0.11.1/tcp/1616
+\_______ bearer __________/\__ key __/\______ overlay ______/
+ underlay, wg `ListenPort`             the ONLY part we bind
+```
+
+The `/wg/` segment is **infix, not suffix** — the segments
+*before* it are the wg **bearer** (the underlay `(ip, udp-port)`
+that `wg(8)` itself listens on, per the codec docstring's own
+`/ip4/1.2.3.4/udp/51820/wg/{key}` example), and the segments
+*after* are the **overlay** endpoint that `tractor` binds.
+
+⚠️ **CORRECTION** — an earlier revision of this plan (and the
+examples in gh #482) used a *suffix* form
+`/ip4/10.0.11.1/tcp/1616/wg/u<key>`. That parses, but it is
+semantically inverted: it puts the overlay addr where the bearer
+belongs, `tcp` where wg's `udp` `ListenPort` goes, and declares
+no overlay endpoint at all. `parse_wg_maddr()` in
+`examples/multihost/wg_lan/` now rejects it with an actionable
+error.
+Observed protocol-name lists, for writing the `match`:
+
+| maddr | `[p.name for p in m.protocols()]` |
+| --- | --- |
+| `/ip4/1.2.3.4/udp/51820/wg/u<k>` | `['ip4','udp','wg']` |
+| `/ip4/../udp/../wg/u<k>/ip4/../tcp/..` | `['ip4','udp','wg','ip4','tcp']` |
+
+- so the three parts have **three different owners**, and only the
+  third is an `Endpoint`:
+
+  | part | bound by | in the runtime? |
+  | --- | --- | --- |
+  | bearer | kernel, via `wg-quick`/`pyroute2` | no |
+  | `/wg/u<key>` | nothing — it's an identity | no, verified out-of-band |
+  | overlay | `tractor`'s `IPCServer` | **yes**, as `.overlay` |
+
+  This owner-split is the real axis of the design, *not* whether
+  the maddr stack is "composed" (it is).
+- ⚠️ **CORRECTION**, an earlier draft of this section specced a
+  hand-rolled `_peel_tunnel_segs(proto_names) -> (bearer_names,
+  tunnel_specs, overlay_names)`. **Do not write it.**
+  `py-multiaddr` already ships the whole tunnel compose/peel API
+  and it was simply missed here — see its README "En/decapsulate"
+  and "Tunneling" sections, and gh #443's 2nd bullet which links
+  them. Verified against the pinned rev:
+
+  | need | API |
+  | --- | --- |
+  | isolate the bearer | `ma.decapsulate_code(P_WG)` |
+  | drop the overlay, keep bearer+key | `ma.decapsulate(overlay_ma)` |
+  | per-seg maddrs | `ma.split()` |
+  | rejoin a seg tail | `Multiaddr.join(*segs)` |
+  | read the key | `ma.value_for_protocol('wg')` |
+  | recompose | `bearer.encapsulate(key).encapsulate(overlay)` |
+
+  `.decapsulate_code()` handles the infix `/wg/` seg cleanly
+  *because* it cuts on proto-code and never tries to match an
+  addr value — the key seg has no addr of its own. This is the
+  same NIH trap gh #429 existed to close, one layer up.
+
+- ⚠️ `value_for_protocol('ip4')` on a *full* tunnelled maddr
+  silently returns the **first** match, i.e. the bearer's host.
+  Always call it on a peeled sub-maddr, never the whole stack.
+
+- `parse_maddr()` gains a case on
+  `[('ip4'|'ip6'), 'udp', 'wg', ('ip4'|'ip6'), <overlay-l4>]` →
+  peel w/ the API above, decode the multibase key to std-base64,
+  and return `TunnelledAddress(overlay=..., tunnel=WGTunnelSpec(
+  ...))` w/ the bearer recorded in the spec.
+- keep the existing 2-proto cases byte-identical; add the new
+  case *after* them.
+- nesting (wg-in-wg) falls out of `.decapsulate_code()` cutting
+  at the *last* occurrence — peel repeatedly rather than
+  recursing through a bespoke splitter.
+- `mk_maddr()` inverse for `TunnelledAddress` is just
+  `.encapsulate()` composition; don't rebuild `str`s by hand.
+- **pending an upstream release**: py-multiaddr#108 is merged, so
+  `Multiaddr('/…/wg/u…')` parses — but off a `[tool.uv.sources]`
+  `rev` pin, since no release carries the codec. Gate the tests
+  on `_have_wg_maddr_proto()`, implemented as
+  `protocols.protocol_with_name('wg')` under
+  `except ProtocolNotFoundError`. Do **not** probe by parsing a
+  dummy like `Multiaddr('/wg/uAAAA')` — the codec enforces a
+  32-byte key, so that raises even when the proto *is* known. Do
+  **not** hand-roll a `wg` parser in `tractor` — the whole point
+  of #429 was dropping the NIH parser.
+
+### 3.3 verification helper (pure, composable)
+
+Port #482 §2's helpers into `tractor/discovery/_tunnel.py` as
+*pure functions* + one impure probe, cleanly separated:
+
+```python
+def parse_wg_maddr(maddr: str) -> TunnelledAddress: ...   # pure
+def wg8_pubkey(multibase_key: str) -> str: ...            # pure
+def verify_wg_peer(spec: WGTunnelSpec) -> bool: ...       # impure probe
+```
+
+In layer A `verify_wg_peer()` may shell out (`wg show <if>
+peers`), but it must be a *single* function so layer B swaps
+only its body. Never call it implicitly from
+`wrap_address()`/`parse_maddr()` — parsing must stay pure and
+side-effect-free; verification is the *caller's* explicit step
+(and later, the bindspace `@acm`'s).
+
+### 3.4 deliverables
+
+- `examples/` scripts distilled from #482 §§3-5 (this is the
+  unchecked "commit examples from ^" bullet in #443). They live
+  under `examples/multihost/` — `test_docs_examples.py` walks
+  `examples/` recursively and runs every collected file as a
+  subproc asserting `rc == 0` (it doesn't even filter by
+  extension, so a stray `README.md` would be `python`-run too),
+  and `'multihost' not in p[0]` is already in its exclusion
+  list. Anything needing a real second host or a live tunnel
+  belongs there.
+- a `docs/` page: tunnel setup, the maddr form, the two-host
+  run. Keep prose in the docs; keep the examples runnable and
+  minimal.
+- tests: maddr round-trip, `TunnelledAddress` delegation
+  (`proto_key`/`unwrap` identical to overlay), `wrap_address()`
+  regression (a tunnelled maddr `str` → `TunnelledAddress`; a
+  plain one → unchanged), and **a real end-to-end over a
+  locally-created wg pair** gated on `CAP_NET_ADMIN` (see §5.3).
+
+---
+
+## 4. Layer B — `pyroute2` under `trio`
+
+### 4.1 the library situation (verify at implementation time)
+
+`pyroute2` ≥0.9 rewrote its core onto **asyncio**
+(`AsyncIPRoute`; the sync `IPRoute` wraps it with its own loop).
+It also ships a `WireGuard` netlink (generic-netlink) module
+supporting `.set(iface, private_key=..., peer={...})` and
+`.info(iface)`, plus `pyroute2.netns` / `NetNS` for namespaces,
+and `IPRoute.link('add', kind='wireguard', ifname=...)`.
+
+Three integration options, in increasing trio-nativeness:
+
+- **(1) `trio.to_thread.run_sync()` around the sync API.**
+  Netlink ops here are one-shot, sub-millisecond, and happen at
+  bind/teardown time only — *not* in the msg hot path. This is
+  the **correct default**: it's ~10 lines, uses a battle-tested
+  API, and costs nothing where it's used.
+- **(2) sans-io: `trio.socket` + pyroute2's message codecs.**
+  `pyroute2`'s message classes
+  (`pyroute2.netlink.rtnl.*`, `pyroute2.netlink.generic.wireguard.wgmsg`)
+  encode/decode independently of its I/O core. So a
+  `tractor/ipc/_netlink.py` with a small trio `NetlinkSocket`
+  (`trio.socket.socket(AF_NETLINK, SOCK_RAW|SOCK_DGRAM, proto)`,
+  `sendto`/`recv`, seq/pid matching, `NLMSG_DONE`/`NLMSG_ERROR`
+  handling) + pyroute2 codecs is very achievable and is the
+  honest reading of "as much trio wrapping as possible where any
+  other async support can be replaced".
+  **Do this for the paths we actually need** (link add/del,
+  addr add, wg get/set, netns bind) and *only* those — a
+  general netlink client is out of scope.
+- (3) reimplement the codecs. Never.
+
+**Recommended split**: ship (1) first so layer B is a small,
+reviewable, behaviour-preserving swap of `verify_wg_peer()`'s
+body; then land (2) as a follow-up commit for the read path
+(`wg get`, `link get`) where the sans-io surface is smallest,
+and keep (1) for the privileged mutating ops. Measure before
+converting anything else — there is no perf argument here, only
+a "no foreign event loop in a trio actor" argument, which (1)
+already satisfies (a thread is not an event loop).
+
+Explicitly **do not** pull in `trio-asyncio` for pyroute2: it
+would be the one place in the runtime where an asyncio loop
+exists for no reason.
+
+### 4.2 API shape
+
+Pure-ish, functional, `@acm` for anything with teardown:
+
+```python
+async def read_wg_peers(
+    iface: str = 'wg0',
+    netns: str|None = None,
+) -> tuple[str, ...]: ...          # base64 pubkeys
+
+async def read_wg_pubkey(iface: str = 'wg0', ...) -> str: ...
+```
+
+and `verify_wg_peer()` becomes a thin composition over the two.
+Note the pure-getter rule: no `read_wg_peers(..., create=True)`.
+
+---
+
+## 5. Layer C — nested bindspace `@acm`s + `Address.namespace`
+
+This is the part #443 and `multiaddr_declare_eps.md` actually
+ask for: *"for any tunneled maddr-`str`-entry we deliver a
+data-structure which can easily be passed to nested `@acm`s
+which consecutively setup nested net bindspaces for binding the
+endpoint addrs"*.
+
+### 5.1 the composition
+
+```python
+@acm
+async def open_bindspace(
+    addr: TunnelledAddress,
+) -> AsyncGenerator[Address, None]:
+    '''
+    Enter the net-bindspace implied by `addr`'s tunnel stack,
+    yielding the *overlay* `Address` ready to bind/connect.
+
+    Nests: one `@acm` per tunnel segment, outermost-first, so
+    a 2-deep stack is just two nested `async with`s and the
+    teardown order is guaranteed by `trio`.
+
+    '''
+```
+
+with per-tunnel-kind implementations:
+
+```python
+@acm
+async def open_netns(name: str) -> AsyncGenerator[None, None]: ...
+@acm
+async def open_wg_iface(spec: WGTunnelSpec) -> AsyncGenerator[WGTunnelSpec, None]: ...
+```
+
+and a driver that folds a list of specs into nested contexts
+(`contextlib.AsyncExitStack` for the N-deep case). The
+`parse_endpoints()` API (`_multiaddr.py:153`) is the front door:
+it already returns `dict[name, list[Address]]` and the
+`multiaddr_declare_eps.md` sketch anticipates the recursive
+`dict[str, list[Address]]|dict[...]` return for tunnelled
+entries. Extend it to carry the tunnel stack, not to *enter* it.
+
+### 5.2 `Address.namespace`, at last
+
+- `TunnelledAddress.namespace` → `(kind, id)` e.g.
+  `('netns', 'tractor-wg0')`.
+- **and** the existing backends should implement it as `None`
+  explicitly (they currently just don't define it), so the
+  Protocol stops lying.
+- consumers to audit: nothing reads `.namespace` today — so
+  adding it is safe, but the *point* is that
+  `Endpoint`/`Server.pformat()` should start showing it (there's
+  already a `# !TODO, always be ns aware!` +
+  `f'|_netns: {netns}\n'` placeholder sitting in
+  `Endpoint.pformat()`, `_server.py:645`). Fill that in; it's
+  the cheapest possible proof the layer is wired.
+
+### 5.3 the netns/process reality — read this before designing
+
+**The headline consequence, stated up front**: netns is a
+**runtime-level config API, not an actor-app-code API.** It is
+declared as part of how an actor process is *brought up* — a
+spawn-time/boot-time input alongside `enable_transports` and
+`tpt_bind_addrs` — and it is **not** dynamically re-enterable by
+app code once the actor is live. There is deliberately no
+`await actor.enter_netns(...)`. Two hard reasons, both below:
+`setns(2)` doesn't retroactively move existing sockets, and it's
+per-thread rather than per-process. Anything that *looks* like a
+mid-life API here would be a footgun that silently leaves the IPC
+server bound in the old namespace.
+
+- `setns(2)` with `CLONE_NEWNET` affects **the calling thread
+  only**, and sockets already created keep their original netns.
+  A trio actor is effectively single-threaded for our purposes,
+  so "enter the netns, *then* bind" works — but any
+  `to_thread` worker (§4.1 option 1!) is in the **original**
+  netns unless it also `setns`. Concretely: a wg query issued
+  via `trio.to_thread` will hit the wrong namespace. Either
+  pass `netns=` down to `pyroute2` (which does the
+  fork/setns dance itself) or pin a dedicated worker. **This is
+  the single subtlest bug in this plan — write the test first.**
+- entering a netns is *process-global-ish and irreversible-ish*
+  in practice. Therefore: **netns membership belongs to the
+  actor process, decided before the runtime binds**, not to a
+  mid-life `@acm`. Design:
+  - the root/parent decides the netns for a subactor and passes
+    it in the spawn spec (there's already
+    `enable_transports`/`accept_addrs` plumbing at
+    `_runtime.py:1595-1615` — the netns rides alongside).
+  - the child, in `_runtime.async_main()` **before**
+    `IPCServer.listen_on()`, enters it.
+  - the mid-life `@acm` form is then only for the *root* /
+    single-actor case, and for iface creation (which is
+    genuinely scoped).
+  - document the constraint rather than hiding it; a
+    `RuntimeError` if `open_netns()` is entered after any
+    listener exists.
+- privileges: iface/netns creation needs `CAP_NET_ADMIN`.
+  Never `sudo` from inside the runtime. Two supported modes:
+  (i) pre-provisioned out-of-band (layers A/B — the default,
+  and what #482 documents), (ii) runtime-managed when the
+  process already holds the cap. Detect with a cheap
+  `os.geteuid()==0 or CAP_NET_ADMIN in /proc/self/status`
+  probe and *fail loudly with an actionable message* otherwise.
+- teardown must be idempotent and tolerant: an iface/netns
+  already gone must not strand the rest of the teardown — the
+  exact lesson `_uds.close_listener()`'s `FileNotFoundError`
+  tolerance and `_serve_ipc_eps()`'s per-ep `try/except`
+  encode. Mirror both.
+
+### 5.4 tests for layer C
+
+- unit: fold-N-tunnel-specs-into-nested-`@acm`s, with fakes; assert
+  enter/exit ordering (outermost-last-out) via a trace list.
+- integration, gated on `CAP_NET_ADMIN` (skip otherwise, and in
+  CI run it in a `--cap-add NET_ADMIN` container job): create two
+  netns + a wg pair entirely in-process, boot a `tractor` root in
+  one and a subactor in the other, `find_actor()` across the
+  tunnel. This is a *fantastic* test to have and is fully
+  self-contained — no second host, no `sudo` in the test body.
+- the `to_thread`-netns-mismatch regression from §5.3, written
+  **first** (red), then the fix (green), per project convention.
+
+---
+
+## 6. "Other shuttle-able tpts"
+
+The generalization the #482 follow-up gestures at: once
+`TunnelledAddress` + `open_bindspace()` exist, the same
+machinery covers any iface-layer tunnel `pyroute2` can drive —
+`ipip`/`gre`/`sit`/`vxlan`/`geneve`/`bridge`/`veth`. Keep
+`WGTunnelSpec` as *one* frozen struct among a
+`TunnelSpec = WGTunnelSpec|VxlanTunnelSpec|...` union with a
+`kind: ClassVar[str]`, and dispatch `open_*` by `match` on it.
+Design for it now (union + `match`), implement only `wg` +
+`netns`. `veth`-pairs-in-netns is the natural second one because
+it makes the §5.4 integration test possible without wg at all —
+consider doing it *first* for exactly that reason.
+
+## 7. Non-goals
+
+- no wg userspace implementation, no key exchange, no
+  `wg-quick` reimplementation (config-file parsing is
+  out of scope; take structured input).
+- no persistence of private keys beyond what layer C's iface
+  creation needs (and that stays in `get_rt_dir()`, 0600).
+- macOS/Windows: layers B/C are Linux-only. Layer A (declarative)
+  works anywhere `wg` does. Gate accordingly and say so in the
+  docs — do not silently no-op.
+
+## 8. Risks
+
+| risk | mitigation |
+| --- | --- |
+| `to_thread` worker runs in the wrong netns | §5.3; pass `netns=` to pyroute2 or pin a worker; test-first |
+| py-multiaddr#108 merged but unreleased | `[tool.uv.sources]` `rev` pin + `_have_wg_maddr_proto()` gate; layer A's overlay-addr path works regardless |
+| `TunnelledAddress` leaks into `Endpoint` and breaks `inspect.getmodule()` | unwrap at parse/bindspace boundary; assert `not isinstance(ep.addr, TunnelledAddress)` in `Endpoint.__post_init__` |
+| privileged ops in a library | never `sudo`; explicit cap probe + actionable error; pre-provisioned is the default |
+| pyroute2 0.9 asyncio core drags a loop into the actor | option (1) is a *thread*, not a loop; forbid `trio-asyncio` here (§4.1) |
+| netns teardown strands actor teardown | idempotent/tolerant teardown mirroring `_uds.close_listener()` |
+
+## 9. Follow-up issue seeds
+
+- `veth`-in-netns bindspace (unblocks capless-ish integration
+  testing, and is a great local multi-"host" test rig)
+- composed/tunnelled maddr grammar shared with plan 02's
+  `/…/quic-v1/…` stacks (gh #443)
+- `wg` proto into the multiaddr **spec** (gh #483), then flip
+  `MsgTransport.maddr` to always return `Multiaddr` (the third
+  #443 bullet)
+- runtime-managed wg key rotation / peer add-remove as a
+  `tractor` service actor — the natural "actor that owns the
+  network" demo

ai/tpt-backends/README.md

@@ -0,0 +1,54 @@
+# next-gen `tractor.ipc` transport backend plans
+
+Implementation specs for three prospective `.ipc` transport
+backends, written so each can be worked independently (by a
+different model/provider) without design or lib-selection drift.
+
+**Read [`00_shared_backend_contract.md`](./00_shared_backend_contract.md)
+first** — it is the normative description of what a `tractor`
+transport backend *is* as of `main@83b34884` (the backend
+duck-type, the 10-item registration checklist, the test-harness
+plumbing, the code-style rules). The three plans assume it and
+document only their own deltas.
+
+| plan | issue | dep | size | lands |
+| --- | --- | --- | --- | --- |
+| [01 — TIPC](./01_tipc_backend.md) | [#378] | **none** (stdlib) | small | first |
+| [02 — QUIC/`iroh`](./02_quic_iroh_backend.md) | [#353] | `iroh` (uniffi FFI) | large | needs a prep PR |
+| [03 — `wg` bindspace](./03_wg_tunnel_bindspace.md) | [#482], [#443] | `pyroute2` | medium, 3 layers | layer A now |
+
+Headline conclusions:
+
+- **TIPC is the cheap win.** Verified: `trio.SocketStream` and
+  `trio.SocketListener` are address-family agnostic (only
+  `SOCK_STREAM` + a trio socket), and CPython ships `AF_TIPC` +
+  23 `TIPC_*` constants. So the backend is ~one module of
+  contract boilerplate, zero new deps, and it buys
+  *kernel-native* service discovery: `bind()` publishes,
+  `connect()`-by-name resolves — no registrar in the loop.
+  (`modprobe tipc` is required; hard-gate everything.)
+- **QUIC's cost is entirely in two adapters**, not in QUIC. The
+  `iroh` python bindings are `uniffi`-generated asyncio, but the
+  asyncio dependency is confined to *one* future-poll callback —
+  a ~40-line `trio` bridge (`TrioToken.run_sync_soon`) replaces
+  it. The second cost is that an iroh listener isn't a socket,
+  which needs a small, independently-reviewable prep PR to
+  `_server.py`/`_types.py`.
+- **WireGuard is not a transport.** It's an iface-layer tunnel,
+  so it belongs as a *nested bindspace* (`TunnelledAddress` +
+  `open_bindspace()` `@acm`s) wrapping whatever L4 tpt is in
+  use — which is also what finally implements the long-spec'd
+  `Address.namespace`, and what generalizes to
+  `veth`/`vxlan`/`gre`.
+
+Ordering rationale: plan 01 first as the cheap proof the
+table-registration story generalizes to a genuinely new proto;
+plan 03 layer A is already deployable-today doc/example work;
+plan 02 last (and gated on its prep PR). Plans 01 and 02 both
+want the same `Address.rebind_from_sockname` gate — whichever
+lands first ships it.
+
+[#378]: https://github.com/goodboy/tractor/issues/378
+[#353]: https://github.com/goodboy/tractor/issues/353
+[#482]: https://github.com/goodboy/tractor/issues/482
+[#443]: https://github.com/goodboy/tractor/issues/443

docs/api/trionics.rst

@@ -40,6 +40,9 @@ Broadcast fan-out
 .. autoexception:: Lagged
    :show-inheritance:
 
+.. autoexception:: BroadcastReceiveError
+   :show-inheritance:
+
 A single-producer, many-consumer broadcast layer over any
 ``trio``-style receive channel: non-lossy for the *fastest*
 consumer while slower consumers raise :class:`Lagged` (a
@@ -48,6 +51,13 @@ internal ring. This is exactly the machinery behind
 :meth:`tractor.MsgStream.subscribe` — see
 ``examples/streaming_broadcast_fanout.py``.
 
+If the shared underlying receiver raises an ordinary exception, the
+subscriber which owned that receive gets the original failure.
+Waiting peers drain their retained values and then raise
+:class:`BroadcastReceiveError`, with the original failure available
+as ``__cause__``. Later subscribers observe the same terminal state
+without retrying the failed underlying receiver.
+
 ExceptionGroup helpers
 ----------------------
 

docs/guide/asyncio.rst

@@ -209,6 +209,11 @@ The underlying broadcast machinery is lazily allocated on first
 use and is *not* reversible for the channel's remaining lifetime,
 so only reach for it when you actually want the fan-out.
 
+As with ``MsgStream``, pass ``raise_on_lag=False`` for a consumer
+which may warn, drop old values and resume from the retained window.
+Each child chooses independently; the first subscription also fixes
+the linked channel's root receive policy.
+
 One-shot calls with ``run_task()``
 ----------------------------------
 When you just want a single ``asyncio`` result and no streaming

docs/guide/streaming.rst

@@ -171,6 +171,20 @@ keeps pace with the *fastest* subscriber; a task falling more
 than the buffered window behind has its next receive raise
 ``tractor.trionics.Lagged`` to say it lost data.
 
+Each ``BroadcastReceiver`` is one logical subscription cursor, so
+give every concurrent consumer task its own receiver. Overlapping
+``receive()`` calls on the same handle raise
+``trio.BusyResourceError`` instead of racing that cursor. In strict
+mode values are never skipped silently: the consumer either reads
+each retained value in sequence or receives an explicit ``Lagged``
+error after exceeding the buffer window.
+
+Pass ``raise_on_lag=False`` when a consumer may drop old values and
+resume from the oldest retained item instead. The receiver logs the
+overrun rather than raising. Each child subscription chooses its own
+policy; the first call also fixes the policy of the stream's root
+receive handle because broadcaster allocation is irreversible.
+
 The broadcast handle stays duplex btw: it proxies ``send()``
 through to the underlying stream, so each subscriber task can
 keep talking upstream while consuming its fan-out copy.

examples/multihost/wg_lan/README.md

@@ -0,0 +1,176 @@
+# `tractor` over a WireGuard tunnel, declared as one maddr
+
+A two-host LAN setup: a `tractor` actor tree on host A, dialed
+from host B, with the endpoint declared as a single `wg`
+multiaddr.
+
+Supersedes the example set in gh
+[#482](https://github.com/goodboy/tractor/issues/482) — see
+[what changed](#what-changed-vs-482).
+
+> **Why `examples/multihost/`?** `tests/test_docs_examples.py`
+> walks `examples/` recursively and runs everything it collects
+> as a subproc, asserting `rc == 0`. These need a real second
+> host and a live `wg` tunnel, so they can't satisfy that;
+> `'multihost' not in p[0]` is already in the test's exclusion
+> list, which is what keeps them out of CI.
+
+## the maddr form
+
+```
+/ip4/192.168.1.50/udp/51820/wg/u<A_pub>/ip4/10.0.11.1/tcp/1616
+\____ wg bearer ___________/\__ key __/\____ tractor ep _____/
+ underlay, wg `ListenPort`              overlay, on the wg iface
+ (kernel/`wg(8)` owns it)               (the ONLY part tractor binds)
+```
+
+Three parts, three different owners:
+
+| part | who binds it | in the runtime? |
+| --- | --- | --- |
+| `/ip4/../udp/51820` bearer | kernel via `wg-quick`/`pyroute2` | no |
+| `/wg/u<key>` | nothing — it's an identity | no, verified out-of-band |
+| `/ip4/../tcp/1616` overlay | `tractor`'s `IPCServer` | **yes**, as `.overlay` |
+
+Verified against py-multiaddr
+[#108](https://github.com/multiformats/py-multiaddr/pull/108):
+this composed form parses and round-trips
+(`['ip4','udp','wg','ip4','tcp']`).
+
+## requirements
+
+py-multiaddr #108 is **merged** (2026-07-28) but ships in no
+release yet — the latest `0.2.0` (2026-03-17) predates it and has
+no `wg` codec. So `pyproject.toml` carries a temporary
+`[tool.uv.sources]` `rev` pin at the merge commit, and a plain
+
+```bash
+uv sync
+```
+
+gets you a `wg`-aware `multiaddr`. That pin goes away once a
+release carries the codec. You also need `multibase`:
+
+```bash
+uv pip install multibase
+```
+
+Without the codec `parse_wg_maddr()` raises immediately with an
+actionable message — there is deliberately **no** degraded
+hand-split fallback. `_have_wg_maddr_proto()` is the predicate.
+
+Every peel and re-compose here goes through `py-multiaddr`'s own
+tunnel API (`.decapsulate_code()`, `.split()`, `.join()`,
+`.encapsulate()`, `.value_for_protocol()`) rather than any
+bespoke segment slicing — see its README "En/decapsulate" and
+"Tunneling" sections. gh #429 was about *dropping* our NIH
+parser, and that applies to peeling a tunnel stack just as much
+as to decoding one proto.
+
+## 0. tunnel setup (out-of-band, both hosts)
+
+Host A is the service host (underlay e.g. `192.168.1.50`), host B
+your workstation. Overlay net `10.0.11.0/24`.
+
+```bash
+umask 077
+wg genkey | tee wg_priv.key | wg pubkey > wg_pub.key
+```
+
+`/etc/wireguard/wg0.conf` on **host A**:
+
+```ini
+[Interface]
+PrivateKey = <A_priv>
+Address = 10.0.11.1/24
+ListenPort = 51820
+```
+```ini
+[Peer]
+PublicKey = <B_pub>
+AllowedIPs = 10.0.11.2/32
+```
+
+on **host B**:
+
+```ini
+[Interface]
+PrivateKey = <B_priv>
+Address = 10.0.11.2/24
+```
+```ini
+[Peer]
+PublicKey = <A_pub>
+Endpoint = 192.168.1.50:51820
+AllowedIPs = 10.0.11.1/32
+PersistentKeepalive = 25
+```
+
+Note how `ListenPort` and `Endpoint` are exactly the maddr's
+bearer segment, and `[Interface] Address` is its overlay host.
+
+```bash
+sudo wg-quick up wg0   # both hosts
+ping -c1 10.0.11.1     # from B
+```
+
+## 1. get your pubkey into the maddr
+
+```bash
+python -c "
+import base64, multibase
+key = open('wg_pub.key').read().strip()
+print(multibase.encode('base64url', base64.b64decode(key)).decode())
+"
+```
+
+Paste the `u...` output into `WG_MADDR` in both scripts (they use
+the same string — A's bearer, A's key, A's overlay ep).
+
+## 2. run
+
+```bash
+# host A
+python host_a_srv.py
+
+# host B
+python host_b_client.py
+```
+
+`host_a_srv.py` must be importable on host B too, since
+`portal.run()` refs the fn by module path — standard `tractor`
+RPC semantics.
+
+## what changed vs #482
+
+Four corrections, all from
+`ai/tpt-backends/03_wg_tunnel_bindspace.md`:
+
+1. **the maddr semantics were inverted.** #482 used
+   `/ip4/10.0.11.1/tcp/1616/wg/u<key>` — that parses, but it puts
+   the *overlay* addr where the bearer belongs and `tcp` where
+   wg's `udp` `ListenPort` goes, and it declares no overlay ep at
+   all. `parse_wg_maddr()` now rejects it with an actionable
+   error.
+2. **parsing is pure.** #482's helper had the key-check adjacent
+   to the parse; `verify_wg_peer()` is now a separate, explicitly
+   composed step that the caller invokes. A parser that shells
+   out is a nasty surprise.
+3. **no `sudo`.** #482 ran `sudo wg show`; a library/example must
+   never escalate. `wg show` works unprivileged for read on most
+   setups; if yours needs root, run the script as root rather
+   than embedding `sudo`.
+4. **no new `Address` proto-type.** The tunnel rides *beside* the
+   overlay addr in a frozen `WGTunnelledAddr`, and only `.overlay`
+   crosses into `open_nursery()`. #482 §6 floated a `WGAddress`
+   registered in `_address_types` — that table is a `bidict`
+   (1:1 proto-key↔type) and `_addr_to_transport` wants a
+   `MsgTransport` per addr-type, which `wg` doesn't have.
+
+## next
+
+`WGTunnelledAddr` is deliberately example-local. Promoting it to
+`tractor.discovery` as a `TunnelledAddress` whose
+`.proto_key`/`.unwrap()` delegate to `.overlay`, plus
+`open_bindspace()` `@acm`s that create/tear down the iface +
+netns via `pyroute2`, is layers A→C of the plan doc.

examples/multihost/wg_lan/host_a_srv.py

@@ -0,0 +1,61 @@
+# tractor: distributed structured concurrency.
+'''
+Host A: the service host, reachable over a `wg` tunnel.
+
+Binds `tractor`'s registrar + an `echo_srv` sub-actor on the
+tunnel's *overlay* addr, declared as a single `wg` maddr.
+
+'''
+from __future__ import annotations
+
+import tractor
+import trio
+
+from wg_maddr import (
+    parse_wg_maddr,
+    verify_wg_peer,
+    WGTunnelledAddr,
+)
+
+# bearer = host A's underlay `(ip, wg ListenPort)`
+# key    = host A's OWN tunnel pubkey
+# overlay = the ep `tractor` binds, on the wg iface's addr
+WG_MADDR: str = (
+    '/ip4/192.168.1.50/udp/51820'
+    '/wg/u<A_pub_b64url>'
+    '/ip4/10.0.11.1/tcp/1616'
+)
+
+
+async def echo(msg: str) -> str:
+    actor = tractor.current_actor()
+    return f'{actor.aid.name!r} echoes: {msg}'
+
+
+async def main():
+    addr: WGTunnelledAddr = parse_wg_maddr(WG_MADDR)
+    assert verify_wg_peer(addr), (
+        f'wg pubkey from maddr not active on wg0 !\n'
+        f'maddr: {WG_MADDR}\n'
+        f'key: {addr.peer_pubkey}\n'
+    )
+    print(
+        f'wg bearer (kernel-owned): {addr.bearer}\n'
+        f'tractor overlay ep: {addr.overlay}\n'
+    )
+    async with tractor.open_nursery(
+        # XXX only `.overlay` crosses into the runtime; the bearer
+        # + key are iface-layer concerns `tractor` never binds.
+        registry_addrs=[addr.overlay],
+        enable_transports=[addr.overlay_proto],
+    ) as an:
+        await an.start_actor(
+            'echo_srv',
+            enable_modules=[__name__],
+        )
+        print(f'echo_srv up on\n  {addr.maddr}\n')
+        await trio.sleep_forever()
+
+
+if __name__ == '__main__':
+    trio.run(main)

examples/multihost/wg_lan/host_b_client.py

@@ -0,0 +1,52 @@
+# tractor: distributed structured concurrency.
+'''
+Host B: workstation dialing host A's actor tree through the
+`wg` tunnel.
+
+'''
+from __future__ import annotations
+
+import tractor
+import trio
+
+from host_a_srv import echo  # noqa: F401  (RPC refs it by mod path)
+from wg_maddr import (
+    parse_wg_maddr,
+    verify_wg_peer,
+    WGTunnelledAddr,
+)
+
+# same maddr as host A: A's bearer, A's key, A's overlay ep
+WG_MADDR: str = (
+    '/ip4/192.168.1.50/udp/51820'
+    '/wg/u<A_pub_b64url>'
+    '/ip4/10.0.11.1/tcp/1616'
+)
+
+
+async def main():
+    addr: WGTunnelledAddr = parse_wg_maddr(WG_MADDR)
+    assert verify_wg_peer(addr), (
+        f'wg pubkey from maddr not a peer on wg0 !\n'
+        f'maddr: {WG_MADDR}\n'
+    )
+    async with (
+        tractor.open_root_actor(
+            name='wg_client',
+            registry_addrs=[addr.overlay],
+            enable_transports=[addr.overlay_proto],
+        ),
+        tractor.find_actor(
+            'echo_srv',
+            registry_addrs=[addr.overlay],
+        ) as portal,
+    ):
+        res: str = await portal.run(
+            echo,
+            msg='hello over wg!',
+        )
+        print(res)
+
+
+if __name__ == '__main__':
+    trio.run(main)

examples/multihost/wg_lan/wg_maddr.py

@@ -0,0 +1,299 @@
+# tractor: distributed structured concurrency.
+r'''
+Parse `wg`-tunnelled multiaddrs into `tractor`-ready addrs.
+
+The canonical form (per py-multiaddr #108, verified against its
+upstream merge) nests the *overlay* endpoint **after** the `/wg/`
+segment:
+
+    /ip4/10.0.0.1/udp/51820/wg/u<key>/ip4/10.0.11.1/tcp/1616
+    \_______ wg bearer ______/\_ key _/\____ tractor ep _____/
+      (underlay, wg
+       `ListenPort`)
+
+Naming follows `py-multiaddr`'s own encapsulation model, where
+earlier segments *wrap* later ones (`.encapsulate()` appends), so
+the two roles are:
+
+- **bearer**: the segs *before* `/wg/`, i.e. the underlay
+  `(ip, udp-port)` that `wg(8)` itself listens on. Nothing in
+  `tractor` ever binds this — the kernel/`wg` iface owns it.
+- **overlay**: the segs *after*, i.e. the addr `tractor` actually
+  binds/dials. The only part the runtime ever sees.
+
+We deliberately avoid `inner`/`outer` for these two: in a *call*
+stack "inner" reads as higher-up and later-called, whereas here
+the encapsulated addr is bound *first* and sits deeper in the
+maddr — two opposite intuitions on one word.
+
+`/wg/u<key>` itself carries the tunnel peer's Curve25519 pubkey
+as multibase base64url (std base64 from `wg(8)` contains `/` and
+so can't go in a `/`-delimited maddr). It binds nothing at all;
+it's an identity, verified out-of-band.
+
+XXX NOTE, `tractor`'s own `parse_maddr()` can't parse this yet
+(`ValueError('Unsupported multiaddr protocol combo')`), which is
+why this module exists: peel here, hand `.overlay` to the
+runtime.
+
+Design rules this module follows (see
+`ai/tpt-backends/03_wg_tunnel_bindspace.md`):
+
+- **let `py-multiaddr` do the parsing**. Every peel/compose goes
+  through `.decapsulate_code()`, `.split()`, `.join()`,
+  `.encapsulate()` and `.value_for_protocol()`. We hand-roll no
+  segment splitting whatsoever — the whole point of gh #429 was
+  dropping the NIH parser, and that applies to *peeling a tunnel
+  stack* every bit as much as to decoding a single proto.
+- **parsing is pure**. `parse_wg_maddr()` does no I/O, no
+  `subprocess`, no netlink. A parser that shells out is a nasty
+  surprise.
+- **verification is an explicit, separate step**. The caller
+  composes `verify_wg_peer()` when it wants it; nothing implicit.
+- **no new `Address` proto-type**. `wg` gets no entry in
+  `tractor.discovery._addr._address_types` (a `bidict`, so 1:1
+  proto-key<->type) bc it has no `MsgTransport` of its own. The
+  tunnel is a *bindspace*, so we carry it beside the overlay
+  addr and strip to `.overlay` at bind/dial time.
+
+'''
+from __future__ import annotations
+import base64
+import subprocess
+from typing import Literal
+
+import msgspec
+from multiaddr import Multiaddr
+from multiaddr.protocols import P_WG
+
+
+IPProto = Literal['ip4', 'ip6']
+
+
+class WGTunnelledAddr(
+    msgspec.Struct,
+    frozen=True,
+):
+    '''
+    A `wg`-tunnelled endpoint: the underlay bearer, the tunnel
+    peer key, and the overlay addr `tractor` binds/dials.
+
+    '''
+    # underlay, owned by `wg(8)`/the kernel — NEVER bound by us
+    bearer: tuple[str, int]
+
+    # tunnel peer pubkey in the std-base64 `wg(8)` form, i.e.
+    # directly comparable to `wg show <if> peers` output
+    peer_pubkey: str
+
+    # overlay ep: an `UnwrappedAddress` as accepted by
+    # `tractor.discovery.wrap_address()`
+    overlay: tuple[str, int]
+    overlay_proto: Literal['tcp'] = 'tcp'
+
+    # kept so `.as_multiaddr()` re-renders the same ip family it
+    # was parsed from, rather than assuming v4
+    bearer_ip: IPProto = 'ip4'
+    overlay_ip: IPProto = 'ip4'
+
+    def as_multiaddr(self) -> Multiaddr:
+        '''
+        Re-compose the canonical `Multiaddr`, bearer outward-in,
+        using `.encapsulate()` exactly as py-multiaddr's own
+        tunneling example does.
+
+        '''
+        b_host, b_port = self.bearer
+        o_host, o_port = self.overlay
+        return (
+            Multiaddr(f'/{self.bearer_ip}/{b_host}/udp/{b_port}')
+            .encapsulate(
+                Multiaddr(f'/wg/{mb_pubkey(self.peer_pubkey)}')
+            )
+            .encapsulate(
+                Multiaddr(
+                    f'/{self.overlay_ip}/{o_host}'
+                    f'/{self.overlay_proto}/{o_port}'
+                )
+            )
+        )
+
+    @property
+    def maddr(self) -> str:
+        '''
+        The canonical maddr `str` form.
+
+        '''
+        return str(self.as_multiaddr())
+
+
+def mb_pubkey(wg8_key: str) -> str:
+    '''
+    `wg(8)` std-base64 pubkey -> multibase base64url (`u`-prefixed).
+
+    '''
+    import multibase
+    raw: bytes = base64.b64decode(wg8_key)
+    return multibase.encode('base64url', raw).decode('ascii')
+
+
+def wg8_pubkey(mb_key: str) -> str:
+    '''
+    Inverse of `mb_pubkey()`: multibase -> `wg(8)` std-base64.
+
+    '''
+    import multibase
+    raw: bytes = multibase.decode(mb_key)
+    return base64.b64encode(raw).decode('ascii')
+
+
+_wg_proto_known: bool|None = None
+
+
+def _have_wg_maddr_proto() -> bool:
+    '''
+    True iff the installed `py-multiaddr` knows the `/wg/` proto,
+    i.e. carries py-multiaddr#108.
+
+    Merged upstream 2026-07-28 (`f86519da`) but in no release as
+    of `0.2.0`, hence the `[tool.uv.sources]` `rev` pin in
+    `pyproject.toml`.
+
+    Pure predicate; result cached since it can't change without a
+    reinstall.
+
+    '''
+    global _wg_proto_known
+    if _wg_proto_known is None:
+        from multiaddr.protocols import protocol_with_name
+        from multiaddr.exceptions import ProtocolNotFoundError
+        try:
+            protocol_with_name('wg')
+            _wg_proto_known = True
+        except ProtocolNotFoundError:
+            _wg_proto_known = False
+
+    return _wg_proto_known
+
+
+def parse_wg_maddr(
+    maddr: str|Multiaddr,
+) -> WGTunnelledAddr:
+    '''
+    Peel a `wg`-tunnelled maddr into its bearer/key/overlay
+    parts. Pure — no I/O.
+
+    Every cut is made by `py-multiaddr`, so a malformed maddr
+    (incl. a `wg` key that isn't exactly 32B) raises out of
+    `Multiaddr()` rather than yielding a struct quietly built
+    from garbage segs.
+
+    '''
+    if not _have_wg_maddr_proto():
+        raise RuntimeError(
+            f'Installed `py-multiaddr` has no `/wg/` proto!\n'
+            f'Needs py-multiaddr#108, merged upstream but not\n'
+            f'yet released; a `uv sync` picks up the pinned rev.\n'
+            f'maddr: {maddr!r}\n'
+        )
+
+    ma: Multiaddr = (
+        maddr
+        if isinstance(maddr, Multiaddr)
+        else Multiaddr(maddr)
+    )
+    segs: list[Multiaddr] = ma.split()
+    names: list[str] = [
+        proto.name
+        for seg in segs
+        for proto in seg.protocols()
+    ]
+    if 'wg' not in names:
+        raise ValueError(
+            f'Not a `wg`-tunnelled maddr, no `/wg/` segment ??\n'
+            f'maddr: {ma}\n'
+        )
+
+    # NOTE, `.decapsulate_code()` cuts at the LAST occurrence of
+    # the proto and keeps the *prefix*, which is exactly the
+    # bearer. It handles `/wg/` cleanly precisely bc it cuts on
+    # proto-code and never tries to match an addr value — the
+    # key seg has no addr of its own.
+    bearer_ma: Multiaddr = ma.decapsulate_code(P_WG)
+    overlay_ma: Multiaddr = Multiaddr.join(
+        *segs[names.index('wg') + 1:]
+    )
+
+    match [proto.name for proto in bearer_ma.protocols()]:
+        case [('ip4' | 'ip6') as b_ip, 'udp']:
+            bearer = (
+                bearer_ma.value_for_protocol(b_ip),
+                int(bearer_ma.value_for_protocol('udp')),
+            )
+        case _:
+            raise ValueError(
+                f'Bad `wg` bearer, expected `/ip4|ip6/<h>/udp/<p>`\n'
+                f'got: {bearer_ma}\n'
+                f'from maddr: {ma}\n'
+            )
+
+    match [proto.name for proto in overlay_ma.protocols()]:
+        case [('ip4' | 'ip6') as o_ip, ('tcp') as l4]:
+            overlay = (
+                overlay_ma.value_for_protocol(o_ip),
+                int(overlay_ma.value_for_protocol(l4)),
+            )
+        case []:
+            raise ValueError(
+                f'`wg` maddr declares no overlay endpoint!\n'
+                f'A bare `/…/wg/<key>` names only the tunnel; '
+                f'append the ep `tractor` should bind, e.g.\n'
+                f'  {ma}/ip4/10.0.11.1/tcp/1616\n'
+            )
+        case _:
+            raise ValueError(
+                f'Unsupported `wg` overlay proto combo\n'
+                f'got: {overlay_ma}\n'
+                f'from maddr: {ma}\n'
+            )
+
+    return WGTunnelledAddr(
+        bearer=bearer,
+        peer_pubkey=wg8_pubkey(ma.value_for_protocol('wg')),
+        overlay=overlay,
+        overlay_proto=l4,
+        bearer_ip=b_ip,
+        overlay_ip=o_ip,
+    )
+
+
+def verify_wg_peer(
+    addr: WGTunnelledAddr,
+    iface: str = 'wg0',
+) -> bool:
+    '''
+    True iff `addr.peer_pubkey` is a configured peer (or our own
+    pubkey) on `iface`.
+
+    IMPURE + explicit by design: never called from
+    `parse_wg_maddr()`.
+
+    ?TODO, per plan-03 layer B, swap this body for `pyroute2`
+    (keeping the signature) — and note `setns(2)` is *per-thread*,
+    so a query issued via `trio.to_thread` lands in the ORIGINAL
+    netns unless `netns=` is passed down.
+
+    '''
+    def _wg(*args: str) -> str:
+        return subprocess.run(
+            ['wg', 'show', iface, *args],
+            capture_output=True,
+            text=True,
+            check=True,
+        ).stdout
+
+    return (
+        addr.peer_pubkey in _wg('peers').split()
+        or
+        addr.peer_pubkey == _wg('public-key').strip()
+    )

pyproject.toml

@@ -169,6 +169,17 @@ sync_pause = {requires-python = ">=3.13, <3.14"}
 # linux kernel networking
 # 'pyroute2
 
+# XXX TEMP, the `/wg/u<key>` maddr proto is MERGED upstream (in
+# py-multiaddr#108, 2026-07-28) but is in NO release yet; the
+# latest `0.2.0` (2026-03-17) predates the merge by ~4 months.
+# Pinned by `rev` (not `branch`) so CI stays reproducible.
+#
+# Drop this pin (and bump the `multiaddr` dep floor above) the
+# moment a release carries the `wg` codec; the only consumer is
+# `examples/multihost/wg_lan/`.
+# |_https://github.com/multiformats/py-multiaddr/pull/108
+multiaddr = { git = 'https://github.com/multiformats/py-multiaddr.git', rev = 'f86519daaa21699023d0037c58cdff600313dd09' }
+
 # ------ tool.uv.sources ------
 
 [tool.uv]

tests/test_task_broadcasting.py

@@ -8,14 +8,18 @@ from contextlib import (
 from functools import partial
 from itertools import cycle
 import time
+from types import SimpleNamespace
 from typing import Optional
+import warnings
 
 import pytest
 import trio
 from trio.lowlevel import current_task
 import tractor
+from tractor.to_asyncio import LinkedTaskChannel
 from tractor.trionics import (
     broadcast_receiver,
+    BroadcastReceiveError,
     Lagged,
     collapse_eg,
 )
@@ -307,6 +311,847 @@ def test_subscribe_errors_after_close():
     trio.run(main)
 
 
+@pytest.mark.parametrize(
+    ('size', 'sent', 'dropped'),
+    [
+        (1, 2, 1),
+        (3, 5, 2),
+    ],
+)
+def test_lagged_reports_exact_drop_count(
+    size: int,
+    sent: int,
+    dropped: int,
+) -> None:
+    '''
+    `Lagged` must report every value outside the retained window.
+
+    `BroadcastReceiver.receive_nowait()` previously subtracted the
+    queue length from an already-invalid deque index without counting
+    that first displaced value. A one-slot queue therefore claimed it
+    dropped zero values after two sends. Keep one root receiver idle
+    while a child subscriber drains every produced value, then prove
+    the lag error reports the exact overrun and positions the root at
+    the oldest value still retained by `BroadcastState.queue`.
+
+    '''
+    async def main() -> None:
+        tx, rx = trio.open_memory_channel(size)
+        brx = broadcast_receiver(rx, size)
+
+        async with brx.subscribe() as fast:
+            for value in range(sent):
+                await tx.send(value)
+                assert await fast.receive() == value
+
+            match = rf'dropped `{dropped}` values'
+            with pytest.raises(Lagged, match=match):
+                await brx.receive()
+
+            assert await brx.receive() == sent - size
+
+    trio.run(main)
+
+
+def test_broadcast_statistics_report_queued_counts() -> None:
+    '''
+    `BroadcastState.statistics()` must report counts, not indexes.
+
+    Each `BroadcastState.subs` value is the deque index of a
+    receiver's next unread value, with `-1` meaning caught up. The
+    statistics API returned these indexes directly, so one queued
+    value appeared as zero and every positive count was one short.
+    Keep one root receiver idle while a child synchronously receives
+    four produced values. Prove the root count advances through one
+    and three retained values, then remains clamped to the three-slot
+    retention window after lagging.
+
+    Finally install an actual unwaited `trio.Event` in
+    `BroadcastState.recv_ready` while treating deprecations as errors.
+    This proves statistics checks `None` explicitly instead of using
+    deprecated `trio.Event` truthiness.
+
+    '''
+    async def main() -> None:
+        tx, rx = trio.open_memory_channel(3)
+        brx = broadcast_receiver(rx, 3)
+
+        async with brx.subscribe() as child:
+            state = brx._state
+            assert state.statistics()['queued_len_by_task'] == {
+                brx.key: 0,
+                child.key: 0,
+            }
+
+            await tx.send(0)
+            assert await child.receive() == 0
+            assert state.statistics()['queued_len_by_task'] == {
+                brx.key: 1,
+                child.key: 0,
+            }
+
+            for value in range(1, 4):
+                await tx.send(value)
+                assert await child.receive() == value
+
+            state.recv_ready = (child.key, trio.Event())
+            with warnings.catch_warnings():
+                warnings.simplefilter('error', DeprecationWarning)
+                stats = state.statistics()
+
+            assert stats['queued_len_by_task'] == {
+                brx.key: 3,
+                child.key: 0,
+            }
+            assert stats['tasks_waiting'] == 0
+            state.recv_ready = None
+
+    trio.run(main)
+
+
+def test_cancelled_reader_diagnostics_are_transient() -> None:
+    '''
+    Cancelled-reader diagnostics must not retain stale `Task`s.
+
+    `BroadcastState.cancelled` previously accumulated every source
+    owner cancelled during `BroadcastReceiver.receive()`. Even after
+    that receiver successfully read again or its subscription closed,
+    `BroadcastState.statistics()` retained the old `Task`, reporting
+    stale state and keeping the completed task alive.
+
+    Cancel one child's source read under a receiver-local scope and
+    verify its task is reported. Reuse that same receiver for one
+    successful read to prove progress clears the entry. Cancel it once
+    more, then leave the subscription and prove close also removes the
+    diagnostic while the root receiver remains registered.
+
+    '''
+    async def main() -> None:
+        tx, rx = trio.open_memory_channel(1)
+        brx = broadcast_receiver(rx, 1)
+        cancel_scope = trio.CancelScope()
+        child_key: int
+        child_task = None
+
+        async with brx.subscribe() as child:
+            child_key = child.key
+
+            async def cancel_source_read() -> None:
+                nonlocal child_task
+                child_task = current_task()
+                with cancel_scope:
+                    await child.receive()
+                assert cancel_scope.cancelled_caught
+
+            async with trio.open_nursery() as nursery:
+                nursery.start_soon(cancel_source_read)
+                while brx._state.recv_ready is None:
+                    await trio.lowlevel.checkpoint()
+                cancel_scope.cancel()
+
+            stats = brx._state.statistics()
+            assert child_task is not None
+            assert stats['tasks_cancelled'] == {
+                child_key: child_task,
+            }
+
+            await tx.send(1)
+            assert await child.receive() == 1
+            assert not brx._state.cancelled
+
+            cancel_scope = trio.CancelScope()
+            async with trio.open_nursery() as nursery:
+                nursery.start_soon(cancel_source_read)
+                while brx._state.recv_ready is None:
+                    await trio.lowlevel.checkpoint()
+                cancel_scope.cancel()
+
+            assert child_key in brx._state.cancelled
+
+        assert child_key not in brx._state.cancelled
+        assert brx.key in brx._state.subs
+
+    trio.run(main)
+
+
+@pytest.mark.parametrize(
+    'terminal_exc',
+    [
+        trio.EndOfChannel(),
+        RuntimeError('terminal source failure'),
+    ],
+    ids=['end-of-channel', 'receive-error'],
+)
+def test_terminal_broadcast_clears_cancelled_tasks(
+    terminal_exc: Exception,
+) -> None:
+    '''
+    Terminal broadcast state must release every cancelled `Task`.
+
+    A receiver which owned and cancelled a source read can leave its
+    task in `BroadcastState.cancelled`. If another receiver later gets
+    EOC or a terminal source failure, no subscriber can make source
+    progress to clear that stale diagnostic. Clearing only the terminal
+    owner's key therefore retained the first receiver's completed task.
+
+    Cancel a child during the first controlled source read, then let
+    the root own a second read which raises EOC or `RuntimeError`.
+    Prove each terminal path clears the other receiver's diagnostic
+    before propagating its exact source outcome.
+
+    '''
+    class TerminalReceiver:
+        '''
+        Block one cancellable read, then raise a terminal outcome.
+
+        '''
+        def __init__(self) -> None:
+            self.calls = 0
+            self.first_started = trio.Event()
+
+        async def receive(self) -> None:
+            '''
+            Drive cancellation followed by terminal source state.
+
+            '''
+            self.calls += 1
+            if self.calls == 1:
+                self.first_started.set()
+                await trio.sleep_forever()
+
+            raise terminal_exc
+
+    async def main() -> None:
+        source = TerminalReceiver()
+        brx = broadcast_receiver(source, 1)
+        cancel_scope = trio.CancelScope()
+
+        async with brx.subscribe() as child:
+            async def cancel_child_read() -> None:
+                with cancel_scope:
+                    await child.receive()
+                assert cancel_scope.cancelled_caught
+
+            async with trio.open_nursery() as nursery:
+                nursery.start_soon(cancel_child_read)
+                await source.first_started.wait()
+                cancel_scope.cancel()
+
+            assert child.key in brx._state.cancelled
+            with pytest.raises(type(terminal_exc)) as exc_info:
+                await brx.receive()
+            assert exc_info.value is terminal_exc
+            assert not brx._state.cancelled
+
+    trio.run(main)
+
+
+def test_end_of_channel_is_terminal_for_waiting_peer() -> None:
+    '''
+    EOC must not let an awakened peer re-enter the closed source.
+
+    `BroadcastState.eoc` was set when one source owner received EOC,
+    but neither receive path consulted it. A peer waiting behind that
+    owner therefore woke, saw no queued value, and started a second
+    source read. Cancellation at that checkpoint could repopulate
+    `BroadcastState.cancelled` after the broadcast became terminal.
+
+    Block one child in the sole source read while the root waits on its
+    event, then release EOC. Both receivers must terminate from that
+    one source call, and the root's later receive must replay EOC
+    immediately without retaining cancellation diagnostics.
+
+    '''
+    class EOCReceiver:
+        '''
+        Publish one controlled EOC and reject any second source read.
+
+        '''
+        def __init__(self) -> None:
+            self.calls = 0
+            self.started = trio.Event()
+            self.release = trio.Event()
+
+        async def receive(self) -> None:
+            '''
+            Block the only valid source read until EOC release.
+
+            '''
+            self.calls += 1
+            assert self.calls == 1
+            self.started.set()
+            await self.release.wait()
+            raise trio.EndOfChannel
+
+    async def main() -> None:
+        source = EOCReceiver()
+        brx = broadcast_receiver(source, 1)
+        outcomes: list[str] = []
+
+        async with brx.subscribe() as child:
+            async def receive_eoc(
+                receiver,
+                name: str,
+            ) -> None:
+                with pytest.raises(trio.EndOfChannel):
+                    await receiver.receive()
+                outcomes.append(name)
+
+            async with trio.open_nursery() as nursery:
+                nursery.start_soon(receive_eoc, child, 'child')
+                await source.started.wait()
+                nursery.start_soon(receive_eoc, brx, 'root')
+
+                _, event = brx._state.recv_ready
+                while not event.statistics().tasks_waiting:
+                    await trio.lowlevel.checkpoint()
+                source.release.set()
+
+            with pytest.raises(trio.EndOfChannel):
+                await brx.receive()
+
+        assert sorted(outcomes) == ['child', 'root']
+        assert source.calls == 1
+        assert not brx._state.cancelled
+
+    trio.run(main)
+
+
+def test_msgstream_eoc_close_preserves_aclose_override() -> None:
+    '''
+    Internal EOC cleanup must preserve the public `aclose()` contract.
+
+    Passing a new private keyword from `MsgStream.receive()` to
+    `self.aclose()` broke subclasses whose compatible override kept
+    the original zero-argument signature. Use a minimal subclass which
+    records virtual dispatch and delegates to the base implementation.
+    Drive graceful EOC through the real root broadcaster and prove the
+    override runs without closing that active root re-entrantly.
+
+    '''
+    class Stream(tractor.MsgStream):
+        '''
+        Record public close dispatch with the established signature.
+
+        '''
+        close_calls = 0
+
+        async def aclose(self):
+            '''
+            Delegate closure without accepting private arguments.
+
+            '''
+            self.close_calls += 1
+            return await super().aclose()
+
+    class PldRx:
+        '''
+        Delegate source receive and terminate the close drain.
+
+        '''
+        def __init__(self, rx) -> None:
+            self._rx = rx
+
+        async def recv_pld(self, **kwargs):
+            '''
+            Receive directly from the test source channel.
+
+            '''
+            return await self._rx.receive()
+
+        def recv_msg_nowait(self, **kwargs):
+            '''
+            Report EOC to finish `MsgStream.aclose()` draining.
+
+            '''
+            raise trio.EndOfChannel
+
+    async def main() -> None:
+        tx, rx = trio.open_memory_channel(1)
+        ctx = SimpleNamespace(
+            cid='test-context',
+            _pld_rx=PldRx(rx),
+            send_stop=lambda: trio.lowlevel.checkpoint(),
+            side='caller',
+            peer_side='callee',
+            maybe_raise=lambda **kwargs: None,
+        )
+        stream = Stream(ctx, rx)
+
+        async with stream.subscribe():
+            await tx.aclose()
+            with pytest.raises(trio.EndOfChannel):
+                await stream.receive()
+            assert stream.close_calls == 1
+            assert not stream._broadcaster._closed
+
+    trio.run(main)
+
+
+@pytest.mark.parametrize(
+    'close_wrapper',
+    [
+        tractor.MsgStream.aclose,
+        LinkedTaskChannel.aclose,
+    ],
+    ids=['msg-stream', 'linked-task-channel'],
+)
+def test_wrapper_close_clears_root_cancelled_task(
+    close_wrapper,
+) -> None:
+    '''
+    Public stream close must release root cancellation diagnostics.
+
+    Root broadcasters allocated by `MsgStream.subscribe()` and
+    `LinkedTaskChannel.subscribe()` are private implementation state.
+    If their source receive was cancelled, callers had no public way
+    to close the root, so wrapper teardown retained the completed
+    `Task` in `BroadcastState.cancelled` indefinitely.
+
+    Cancel a root source read, attach that broadcaster to a minimal
+    public wrapper, and close it through each real `aclose()` method.
+    The root receiver and its task diagnostic must both be removed;
+    for `MsgStream`, pre-close the source to cover its idempotent early
+    return path.
+
+    '''
+    async def main() -> None:
+        _, rx = trio.open_memory_channel(1)
+        brx = broadcast_receiver(rx, 1)
+        cancel_scope = trio.CancelScope()
+
+        async def cancel_source_read() -> None:
+            with cancel_scope:
+                await brx.receive()
+            assert cancel_scope.cancelled_caught
+
+        async with trio.open_nursery() as nursery:
+            nursery.start_soon(cancel_source_read)
+            while brx._state.recv_ready is None:
+                await trio.lowlevel.checkpoint()
+            cancel_scope.cancel()
+
+        assert brx.key in brx._state.cancelled
+
+        if close_wrapper is tractor.MsgStream.aclose:
+            ctx = SimpleNamespace(cid='test-context')
+            wrapper = tractor.MsgStream(ctx, rx)
+            wrapper._broadcaster = brx
+            await rx.aclose()
+        else:
+            wrapper = SimpleNamespace(
+                _broadcaster=brx,
+                _from_aio=rx,
+            )
+
+        await close_wrapper(wrapper)
+        assert brx.key not in brx._state.subs
+        assert brx.key not in brx._state.cancelled
+
+    trio.run(main)
+
+
+def test_broadcast_rejects_zero_buffer_size() -> None:
+    '''
+    A broadcaster must retain at least one value for peer fan-out.
+
+    `collections.deque(maxlen=0)` silently discards every appended
+    value, so `broadcast_receiver(..., 0)` allowed the source owner to
+    receive while peer cursors advanced into an always-empty queue.
+    Their lag recovery then reset to index `-1` and recursively retried
+    without any retained value to consume.
+
+    Construct a rendezvous memory channel and prove broadcaster setup
+    rejects its zero capacity synchronously with a clear public error,
+    before any receiver is registered or source receive can begin.
+
+    '''
+    _, rx = trio.open_memory_channel(0)
+    with pytest.raises(
+        ValueError,
+        match='`max_buffer_size` must be greater than zero',
+    ):
+        broadcast_receiver(rx, 0)
+
+
+def test_underlying_receive_failure_wakes_all_subscribers() -> None:
+    '''
+    A shared receive failure must terminate every broadcast receiver.
+
+    Previously, only `EndOfChannel` and receiver cancellation woke
+    peer tasks waiting on `BroadcastState.recv_ready`. If the shared
+    underlying receiver raised another error, its owner propagated
+    the failure and cleared the event while every peer remained
+    blocked forever.
+
+    Script one successful receive followed by a controlled
+    `RuntimeError`. Let a fast child own both underlying receives
+    while the root first drains its retained value and then waits on
+    the child's second receive. Release the failure only after both
+    tasks have reached those positions. Both exact errors prove the
+    peer was awakened without losing buffered data. A later
+    subscriber proves the terminal failure remains published for new
+    receivers instead of retrying the failed underlying channel.
+
+    '''
+    class FailingReceiver:
+        '''
+        Return one value, then fail after deterministic release.
+
+        '''
+        def __init__(self) -> None:
+            self.calls: int = 0
+            self.failure_started = trio.Event()
+            self.release_failure = trio.Event()
+
+        async def receive(self) -> int:
+            '''
+            Drive the scripted success-then-failure sequence.
+
+            '''
+            self.calls += 1
+            if self.calls == 1:
+                return 1
+
+            self.failure_started.set()
+            await self.release_failure.wait()
+            raise RuntimeError('underlying receive failed')
+
+    async def main() -> None:
+        source = FailingReceiver()
+        brx = broadcast_receiver(source, 3)
+        child_error: list[RuntimeError] = []
+        root_error: list[BroadcastReceiveError] = []
+        late_error: list[BroadcastReceiveError] = []
+        root_drained = trio.Event()
+
+        async def receive_child() -> None:
+            async with brx.subscribe() as child:
+                assert await child.receive() == 1
+                try:
+                    await child.receive()
+                except RuntimeError as exc:
+                    child_error.append(exc)
+
+        async def receive_root() -> None:
+            assert await brx.receive() == 1
+            root_drained.set()
+            try:
+                await brx.receive()
+            except BroadcastReceiveError as exc:
+                root_error.append(exc)
+
+        with trio.fail_after(1):
+            async with trio.open_nursery() as nursery:
+                nursery.start_soon(receive_child)
+                await source.failure_started.wait()
+
+                nursery.start_soon(receive_root)
+                await root_drained.wait()
+
+                source.release_failure.set()
+
+        assert source.calls == 2
+        assert [str(exc) for exc in child_error] == [
+            'underlying receive failed',
+        ]
+        assert [str(exc) for exc in root_error] == [
+            'Shared broadcast receiver failed',
+        ]
+        assert child_error[0] is not root_error[0]
+        assert root_error[0].__cause__ is child_error[0]
+
+        async with brx.subscribe() as late:
+            with pytest.raises(
+                BroadcastReceiveError,
+                match='Shared broadcast receiver failed',
+            ) as exc_info:
+                await late.receive()
+            late_error.append(exc_info.value)
+        assert late_error[0] is not child_error[0]
+        assert late_error[0] is not root_error[0]
+        assert late_error[0].__cause__ is child_error[0]
+        assert source.calls == 2
+
+    trio.run(main)
+
+
+def test_control_flow_exit_wakes_broadcast_peer() -> None:
+    '''
+    Non-terminal control flow must wake peers without being retained.
+
+    Process-control and cancellation-like `BaseException` values
+    should remain local to the task which receives them, but the old
+    owner still has to wake subscribers blocked on its shared event.
+    Make one child own a controlled `BaseException` receive while the
+    root waits behind it. After release, prove the child gets that
+    exact exit and the root takes ownership of the next underlying
+    receive instead of hanging or replaying the control-flow event.
+
+    '''
+    class ReceiveExit(BaseException):
+        '''
+        Model a non-terminal process-control receive exit.
+
+        '''
+
+    class ControlFlowReceiver:
+        '''
+        Raise one controlled exit, then return a value.
+
+        '''
+        def __init__(self) -> None:
+            self.calls: int = 0
+            self.exit_started = trio.Event()
+            self.release_exit = trio.Event()
+
+        async def receive(self) -> int:
+            '''
+            Drive the scripted control-flow-then-value sequence.
+
+            '''
+            self.calls += 1
+            if self.calls == 1:
+                self.exit_started.set()
+                await self.release_exit.wait()
+                raise ReceiveExit
+
+            return 2
+
+    async def main() -> None:
+        source = ControlFlowReceiver()
+        brx = broadcast_receiver(source, 3)
+        child_exit: list[ReceiveExit] = []
+        root_value: list[int] = []
+
+        async def receive_child() -> None:
+            async with brx.subscribe() as child:
+                try:
+                    await child.receive()
+                except ReceiveExit as exc:
+                    child_exit.append(exc)
+
+        async def receive_root() -> None:
+            root_value.append(await brx.receive())
+
+        with trio.fail_after(1):
+            async with trio.open_nursery() as nursery:
+                nursery.start_soon(receive_child)
+                await source.exit_started.wait()
+                nursery.start_soon(receive_root)
+
+                while True:
+                    _, event = brx._state.recv_ready
+                    if event.statistics().tasks_waiting:
+                        break
+                    await trio.lowlevel.checkpoint()
+
+                source.release_exit.set()
+
+        assert len(child_exit) == 1
+        assert root_value == [2]
+        assert source.calls == 2
+        assert brx._state.receive_exc is None
+
+    trio.run(main)
+
+
+def test_closing_non_owner_preserves_source_wait() -> None:
+    '''
+    Closing one subscriber must not wake another receiver's peers.
+
+    `BroadcastReceiver.aclose()` previously set the one shared
+    `BroadcastState.recv_ready` event even when a different receiver
+    owned the source read. Waiting peers then repeatedly awaited an
+    already-set event until the source produced another value,
+    creating a runnable hot loop on idle streams.
+
+    Block one child in the source receive, then place both the root
+    and a closing child behind its event. Close only that waiting
+    child and prove it gets `ClosedResourceError` without setting the
+    shared event. Both remaining receivers must still get the same
+    value after the source is released.
+
+    '''
+    async def main() -> None:
+        tx, rx = trio.open_memory_channel(1)
+        brx = broadcast_receiver(rx, 3)
+        owner_value: list[int] = []
+        root_value: list[int] = []
+        closing_closed = trio.Event()
+
+        async with (
+            brx.subscribe() as owner,
+            brx.subscribe() as closing,
+        ):
+            async def receive_owner() -> None:
+                owner_value.append(await owner.receive())
+
+            async def receive_root() -> None:
+                root_value.append(await brx.receive())
+
+            async def receive_closing() -> None:
+                with pytest.raises(trio.ClosedResourceError):
+                    await closing.receive()
+                closing_closed.set()
+
+            with trio.fail_after(1):
+                async with trio.open_nursery() as nursery:
+                    nursery.start_soon(receive_owner)
+                    while brx._state.recv_ready is None:
+                        await trio.lowlevel.checkpoint()
+
+                    nursery.start_soon(receive_root)
+                    nursery.start_soon(receive_closing)
+                    _, event = brx._state.recv_ready
+                    while event.statistics().tasks_waiting < 2:
+                        await trio.lowlevel.checkpoint()
+
+                    await closing.aclose()
+                    await closing_closed.wait()
+                    assert not event.is_set()
+                    await tx.send(1)
+
+        assert owner_value == [1]
+        assert root_value == [1]
+
+    trio.run(main)
+
+
+def test_concurrent_receive_raises_busy() -> None:
+    '''
+    Reject concurrent receives on one broadcast handle.
+
+    A receiver stores one private peer-wait cancellation scope. If two
+    tasks receive through the same handle, the second task can replace
+    that scope and prevent `BroadcastReceiver.aclose()` from waking the
+    first task. Block one task in the shared source receive, then prove
+    a second call raises `BusyResourceError` before it can mutate any
+    per-receiver wait state. Releasing the source proves the original
+    receive remains usable.
+
+    '''
+    async def main() -> None:
+        tx, rx = trio.open_memory_channel(1)
+        brx = broadcast_receiver(rx, 1)
+        values: list[int] = []
+
+        async def receive() -> None:
+            values.append(await brx.receive())
+
+        async with trio.open_nursery() as nursery:
+            nursery.start_soon(
+                receive,
+                name='first broadcast consumer',
+            )
+
+            # Synchronize with the background task after it blocks in
+            # the shared source `.receive()`, ensuring the next call is
+            # concurrent with an already-active receive on this handle.
+            while brx._state.recv_ready is None:
+                await trio.lowlevel.checkpoint()
+
+            with pytest.raises(
+                trio.BusyResourceError,
+                match='first broadcast consumer',
+            ):
+                await brx.receive()
+
+            await tx.send(1)
+
+        assert values == [1]
+
+    trio.run(main)
+
+
+@pytest.mark.parametrize(
+    'first_outcome',
+    [
+        1,
+        RuntimeError('discarded source error'),
+        trio.EndOfChannel(),
+    ],
+)
+def test_closing_source_owner_hands_read_to_peer(
+    first_outcome: int|Exception,
+) -> None:
+    '''
+    Closing the source-read owner must transfer ownership to a peer.
+
+    Merely suppressing the old shared-event wake would leave peers
+    blocked behind an externally closed receiver that still owned an
+    idle source read. Script a first receive which blocks until its
+    private scope is cancelled and a second which returns immediately.
+    Close that owner only after the root is waiting behind it. Cover
+    a shielded value, ordinary error and EOC from the cancelled source
+    read. The owner must always get `ClosedResourceError`, while the
+    awakened root takes the second source read without publishing the
+    discarded source outcome.
+
+    '''
+    class HandoffReceiver:
+        '''
+        Block the first source read and satisfy the second.
+
+        '''
+        def __init__(self) -> None:
+            self.calls: int = 0
+            self.first_started = trio.Event()
+            self.release_first = trio.Event()
+
+        async def receive(self) -> int:
+            '''
+            Drive one cancelled read followed by one value.
+
+            '''
+            self.calls += 1
+            if self.calls == 1:
+                self.first_started.set()
+                with trio.CancelScope(shield=True):
+                    await self.release_first.wait()
+                if isinstance(first_outcome, BaseException):
+                    raise first_outcome
+                return first_outcome
+
+            return 2
+
+    async def main() -> None:
+        source = HandoffReceiver()
+        brx = broadcast_receiver(source, 3)
+        owner_closed = trio.Event()
+        root_value: list[int] = []
+
+        async with brx.subscribe() as owner:
+            async def receive_owner() -> None:
+                with pytest.raises(trio.ClosedResourceError):
+                    await owner.receive()
+                owner_closed.set()
+
+            async def receive_root() -> None:
+                root_value.append(await brx.receive())
+
+            with trio.fail_after(1):
+                async with trio.open_nursery() as nursery:
+                    nursery.start_soon(receive_owner)
+                    await source.first_started.wait()
+                    nursery.start_soon(receive_root)
+
+                    _, event = brx._state.recv_ready
+                    while not event.statistics().tasks_waiting:
+                        await trio.lowlevel.checkpoint()
+
+                    await owner.aclose()
+                    source.release_first.set()
+                    await owner_closed.wait()
+
+        assert source.calls == 2
+        assert root_value == [2]
+        assert brx._state.receive_exc is None
+        assert not brx._state.eoc
+
+    trio.run(main)
+
+
 def test_ensure_slow_consumers_lag_out(
     reg_addr,
     start_method,
@@ -448,6 +1293,7 @@ def test_first_recver_is_cancelled():
                     async with brx.subscribe() as bc:
                         async for value in bc:
                             print(value)
+                assert cs.cancelled_caught
 
             async def cancel_and_send():
                 await trio.sleep(0.2)
@@ -519,3 +1365,74 @@ def test_no_raise_on_lag():
 
     with pytest.raises(KeyboardInterrupt):
         trio.run(main)
+
+
+@pytest.mark.parametrize(
+    ('subscribe', 'chan_attr'),
+    [
+        (tractor.MsgStream.subscribe, '_rx_chan'),
+        (LinkedTaskChannel.subscribe, '_from_aio'),
+    ],
+    ids=['msg-stream', 'linked-task-channel'],
+)
+def test_stream_subscribe_forwards_lag_policy(
+    subscribe,
+    chan_attr: str,
+) -> None:
+    '''
+    Stream wrappers must expose per-subscriber lag policy.
+
+    `MsgStream.subscribe()` and `LinkedTaskChannel.subscribe()`
+    previously omitted `BroadcastReceiver.raise_on_lag`, forcing
+    downstream users to mutate a private receiver attribute. Invoke
+    each public wrapper against a minimal receive-compatible handle.
+    Prove the first non-raising subscription configures both the
+    irreversible root broadcaster and its child, while a later strict
+    child selects its own policy without changing that root.
+
+    '''
+    class StreamHandle:
+        '''
+        Provide the wrapper fields needed for local fan-out.
+
+        '''
+        def __init__(self) -> None:
+            self._broadcaster = None
+            setattr(
+                self,
+                chan_attr,
+                SimpleNamespace(
+                    _state=SimpleNamespace(max_buffer_size=1),
+                ),
+            )
+
+        async def receive(self):
+            '''
+            Block if a regression unexpectedly enters source receive.
+
+            '''
+            await trio.sleep_forever()
+
+        async def send(self, value) -> None:
+            '''
+            Satisfy `MsgStream` duplex-handle patching.
+
+            '''
+
+    async def main() -> None:
+        stream = StreamHandle()
+        async with subscribe(
+            stream,
+            raise_on_lag=False,
+        ) as first:
+            assert not stream._broadcaster._raise_on_lag
+            assert not first._raise_on_lag
+
+        async with subscribe(
+            stream,
+            raise_on_lag=True,
+        ) as second:
+            assert not stream._broadcaster._raise_on_lag
+            assert second._raise_on_lag
+
+    trio.run(main)

tractor/_streaming.py

@@ -103,6 +103,12 @@ class MsgStream(trio.abc.Channel):
         self._eoc: bool|trio.EndOfChannel = False
         self._closed: bool|trio.ClosedResourceError = False
 
+        # `MsgStream.receive()` sets this while it calls
+        # `MsgStream.aclose()` after source EOC. That close is
+        # re-entrant from the root `BroadcastReceiver._recv`, so it
+        # must not cancel the same receiver before EOC propagates.
+        self._eoc_close_task: trio.lowlevel.Task|None = None
+
     @property
     def ctx(self) -> Context:
         '''
@@ -256,7 +262,16 @@ class MsgStream(trio.abc.Channel):
 
         # when the send is closed we assume the stream has
         # terminated and signal this local iterator to stop
-        drained: list[Exception|dict] = await self.aclose()
+        #
+        # Preserve virtual dispatch through the public zero-argument
+        # `MsgStream.aclose()` API. The task marker lets the base
+        # implementation distinguish this receive-internal close from
+        # an explicit caller or `MsgStream.__aexit__()` close.
+        self._eoc_close_task = trio.lowlevel.current_task()
+        try:
+            drained: list[Exception|dict] = await self.aclose()
+        finally:
+            self._eoc_close_task = None
         if drained:
         #  ^^^^^^^^TODO? pass these to the `._ctx._drained_msgs:
         #  deque` and then iterate them as part of any
@@ -335,6 +350,20 @@ class MsgStream(trio.abc.Channel):
         # `.__aexit__()` as well!!!
         # => SO ENSURE WE CATCH ALL TERMINATION STATES in this
         # block including the EoC..
+
+        # `MsgStream.subscribe()` stores its hidden root broadcaster
+        # on `self._broadcaster`. Explicit teardown owns that root and
+        # must close it to release its subscriber and cancelled-task
+        # diagnostic. Skip only the receive-internal EOC close above:
+        # cancelling the active root's source-read scope there would
+        # turn graceful EOC into `trio.ClosedResourceError`.
+        if (
+            trio.lowlevel.current_task() is not self._eoc_close_task
+            and
+            (broadcaster := self._broadcaster) is not None
+        ):
+            await broadcaster.aclose()
+
         if self.closed:
             # this stream has already been closed so silently succeed as
             # per ``trio.AsyncResource`` semantics.
@@ -512,6 +541,7 @@ class MsgStream(trio.abc.Channel):
     @acm
     async def subscribe(
         self,
+        raise_on_lag: bool = True,
 
     ) -> AsyncIterator[BroadcastReceiver]:
         '''
@@ -526,6 +556,11 @@ class MsgStream(trio.abc.Channel):
         value from the far end via the internally created broudcast
         receiver wrapper.
 
+        ``raise_on_lag=False`` makes this subscription warn and resume
+        at the oldest retained value after an overrun. The first call
+        also sets that policy for this stream's root receive handle;
+        later child subscriptions choose their policy independently.
+
         '''
         # NOTE: This operation is indempotent and non-reversible, so be
         # sure you can deal with any (theoretical) overhead of the the
@@ -541,6 +576,7 @@ class MsgStream(trio.abc.Channel):
                 # TODO: can remove this kwarg right since
                 # by default behaviour is to do this anyway?
                 receive_afunc=self.receive,
+                raise_on_lag=raise_on_lag,
             )
 
             # NOTE: we override the original stream instance's receive
@@ -552,7 +588,9 @@ class MsgStream(trio.abc.Channel):
             # seems there's no graceful way to type this with ``mypy``?
             # https://github.com/python/mypy/issues/708
 
-        async with self._broadcaster.subscribe() as bstream:
+        async with self._broadcaster.subscribe(
+            raise_on_lag=raise_on_lag,
+        ) as bstream:
             assert bstream.key != self._broadcaster.key
             assert bstream._recv == self._broadcaster._recv
 

tractor/to_asyncio.py

@@ -213,6 +213,14 @@ class LinkedTaskChannel(
     _broadcaster: BroadcastReceiver|None = None
 
     async def aclose(self) -> None:
+        # `LinkedTaskChannel.subscribe()` lazily allocates and retains
+        # this root receiver. Close it first so its receiver-local
+        # source-read scope and cancellation diagnostics are released
+        # before `self._from_aio` becomes inaccessible; child
+        # subscriptions retain their own independent close lifetimes.
+        if (broadcaster := self._broadcaster) is not None:
+            await broadcaster.aclose()
+
         await self._from_aio.aclose()
 
     # ?TODO? async version of this?
@@ -324,6 +332,7 @@ class LinkedTaskChannel(
     @acm
     async def subscribe(
         self,
+        raise_on_lag: bool = True,
 
     ) -> AsyncIterator[BroadcastReceiver]:
         '''
@@ -335,6 +344,11 @@ class LinkedTaskChannel(
 
         See ``tractor._streaming.MsgStream.subscribe()`` for further
         similar details.
+
+        ``raise_on_lag=False`` makes this subscription warn and resume
+        at the oldest retained value after an overrun. The first call
+        also sets that policy for this channel's root receive handle;
+        later child subscriptions choose their policy independently.
         '''
         if self._broadcaster is None:
 
@@ -343,11 +357,14 @@ class LinkedTaskChannel(
                 # use memory channel size by default
                 self._from_aio._state.max_buffer_size,  # type: ignore
                 receive_afunc=self.receive,
+                raise_on_lag=raise_on_lag,
             )
 
             self.receive = bcast.receive  # type: ignore
 
-        async with self._broadcaster.subscribe() as bstream:
+        async with self._broadcaster.subscribe(
+            raise_on_lag=raise_on_lag,
+        ) as bstream:
             assert bstream.key != self._broadcaster.key
             assert bstream._recv == self._broadcaster._recv
             yield bstream

tractor/trionics/__init__.py

@@ -26,6 +26,7 @@ from ._mngrs import (
 from ._broadcast import (
     AsyncReceiver as AsyncReceiver,
     broadcast_receiver as broadcast_receiver,
+    BroadcastReceiveError as BroadcastReceiveError,
     BroadcastReceiver as BroadcastReceiver,
     Lagged as Lagged,
 )

tractor/trionics/_broadcast.py

@@ -100,6 +100,20 @@ class Lagged(trio.TooSlowError):
     '''
 
 
+class BroadcastReceiveError(Exception):
+    '''
+    A shared underlying receiver failed in another subscriber task.
+
+    '''
+
+
+class _BroadcastReceiverClosed(Exception):
+    '''
+    An active receiver was closed while owning the source read.
+
+    '''
+
+
 class BroadcastState(Struct):
     '''
     Common state to all receivers of a broadcast.
@@ -115,6 +129,7 @@ class BroadcastState(Struct):
     # broadcast event to wake up all sleeping consumer tasks
     # on a newly produced value from the sender.
     recv_ready: tuple[int, trio.Event]|None = None
+    recv_scope: trio.CancelScope|None = None
 
     # if a ``trio.EndOfChannel`` is received on any
     # consumer all consumers should be placed in this state
@@ -122,7 +137,13 @@ class BroadcastState(Struct):
     # For now, this is solely for testing/debugging purposes.
     eoc: bool = False
 
-    # If the broadcaster was cancelled, we might as well track it
+    # Any non-EOC failure from the shared underlying receiver is
+    # terminal for every subscriber. Retained values remain readable
+    # before this failure is replayed at each receiver's boundary.
+    receive_exc: Exception | None = None
+
+    # Retain the latest interrupted source-reader task until its
+    # receiver next makes progress or closes.
     cancelled: dict[int, Task] = {}
 
     def statistics(self) -> dict[str, Any]:
@@ -142,13 +163,20 @@ class BroadcastState(Struct):
 
         qlens: dict[int, int] = {}
         for tid, sz in subs.items():
-            qlens[tid] = sz if sz != -1 else 0
+            qlens[tid] = min(
+                sz + 1,
+                len(self.queue),
+            )
 
         return {
             'open_consumers': len(subs),
             'queued_len_by_task': qlens,
             'max_buffer_size': self.maxlen,
-            'tasks_waiting': ev.statistics().tasks_waiting if ev else 0,
+            'tasks_waiting': (
+                ev.statistics().tasks_waiting
+                if ev is not None
+                else 0
+            ),
             'tasks_cancelled': self.cancelled,
             'next_value_receiver_id': key,
         }
@@ -156,12 +184,18 @@ class BroadcastState(Struct):
 
 class BroadcastReceiver(ReceiveChannel):
     '''
-    A memory receive channel broadcaster which is non-lossy for
-    the fastest consumer.
+    One logical subscriber to a shared receive-channel broadcast.
 
-    Additional consumer tasks can receive all produced values by
-    registering with ``.subscribe()`` and receiving from the new
-    instance it delivers.
+    Each instance owns one sequence cursor. Additional consumer tasks
+    must call `.subscribe()` and receive through the new instance it
+    yields. Overlapping `.receive()` calls on the same instance raise
+    `trio.BusyResourceError` rather than racing that cursor or the
+    receiver's close-cancellation state.
+
+    A strict subscriber reads each retained value in sequence. Falling
+    behind the retention window raises `Lagged` instead of silently
+    losing values; `raise_on_lag=False` explicitly opts into dropping
+    displaced values.
 
     '''
     def __init__(
@@ -190,6 +224,8 @@ class BroadcastReceiver(ReceiveChannel):
         self._recv = receive_afunc or rx_chan.receive
         self._closed: bool = False
         self._raise_on_lag = raise_on_lag
+        self._wait_scope: trio.CancelScope|None = None
+        self._receive_task: trio.lowlevel.Task|None = None
 
     def receive_nowait(
         self,
@@ -237,7 +273,10 @@ class BroadcastReceiver(ReceiveChannel):
                 # https://docs.rs/tokio/1.11.0/tokio/sync/broadcast/index.html#lagging
 
                 mxln = state.maxlen
-                lost = seq - mxln
+                # `seq == mxln` is already one past the final
+                # valid deque index, so include that first
+                # displaced value in the loss count.
+                lost = seq - mxln + 1
 
                 # decrement to the last value and expect
                 # consumer to either handle the ``Lagged`` and come back
@@ -255,8 +294,21 @@ class BroadcastReceiver(ReceiveChannel):
                     return self.receive_nowait(_key, _state)
 
             state.subs[key] -= 1
+            state.cancelled.pop(key, None)
             return value
 
+        receive_exc = state.receive_exc
+        if receive_exc is not None:
+            # Re-raising one shared exception mutates its traceback on
+            # every delivery. Give each receiver a stable wrapper while
+            # retaining the original failure as its cause.
+            raise BroadcastReceiveError(
+                'Shared broadcast receiver failed'
+            ) from receive_exc
+
+        if state.eoc:
+            raise trio.EndOfChannel
+
         raise trio.WouldBlock
 
     async def _receive_from_underlying(
@@ -270,14 +322,34 @@ class BroadcastReceiver(ReceiveChannel):
             raise trio.ClosedResourceError
 
         event = trio.Event()
+        recv_scope = trio.CancelScope()
         assert state.recv_ready is None
+        assert state.recv_scope is None
         state.recv_ready = key, event
+        state.recv_scope = recv_scope
 
         try:
             # if we're cancelled here it should be
             # fine to bail without affecting any other consumers
             # right?
-            value = await self._recv()
+            receive_exc: BaseException|None = None
+            with recv_scope:
+                try:
+                    value = await self._recv()
+                except BaseException as exc:
+                    receive_exc = exc
+
+            # Only this receiver's `aclose()` cancels its private
+            # source-read scope, and it marks the receiver closed
+            # first without a checkpoint. Outer task cancellation does
+            # not set `recv_scope.cancel_called`; it remains a real
+            # `trio.Cancelled` and follows the handler below.
+            if recv_scope.cancel_called:
+                assert self._closed
+            if self._closed:
+                raise _BroadcastReceiverClosed
+            if receive_exc is not None:
+                raise receive_exc
 
             # items with lower indices are "newer"
             # NOTE: ``collections.deque`` implicitly takes care of
@@ -303,6 +375,8 @@ class BroadcastReceiver(ReceiveChannel):
             ):
                 state.subs[sub_key] += 1
 
+            state.cancelled.pop(key, None)
+
             # NOTE: this should ONLY be set if the above task was *NOT*
             # cancelled on the `._recv()` call.
             event.set()
@@ -312,11 +386,20 @@ class BroadcastReceiver(ReceiveChannel):
             # if any one consumer gets an EOC from the underlying
             # receiver we need to unblock and send that signal to
             # all other consumers.
+            state.cancelled.clear()
             self._state.eoc = True
             if event.statistics().tasks_waiting:
                 event.set()
             raise
 
+        except _BroadcastReceiverClosed:
+            # `aclose()` cancelled this receiver's source-read scope.
+            # Wake peers so one of them can take ownership after this
+            # task clears `recv_ready` in `finally`.
+            if event.statistics().tasks_waiting:
+                event.set()
+            raise trio.ClosedResourceError
+
         except (
             trio.Cancelled,
         ):
@@ -329,14 +412,59 @@ class BroadcastReceiver(ReceiveChannel):
                 event.set()
             raise
 
+        except Exception as receive_exc:
+            # The underlying receiver is shared by every subscriber,
+            # so any non-EOC failure terminates the entire broadcast.
+            # Publish it before waking peers so they can drain their
+            # retained values and then observe the same failure.
+            state.cancelled.clear()
+            state.receive_exc = receive_exc
+            if event.statistics().tasks_waiting:
+                event.set()
+            raise
+
+        except BaseException:
+            # Process-control and cancellation-like exceptions must
+            # not become durable broadcast state, but peers still
+            # need waking before `recv_ready` is cleared.
+            state.cancelled.pop(key, None)
+            if event.statistics().tasks_waiting:
+                event.set()
+            raise
+
         finally:
             # Reset receiver waiter task event for next blocking condition.
             # this MUST be reset even if the above ``.recv()`` call
             # was cancelled to avoid the next consumer from blocking on
             # an event that won't be set!
             state.recv_ready = None
+            state.recv_scope = None
 
     async def receive(self) -> ReceiveType:
+        '''
+        Receive the next value for this subscriber's sequence cursor.
+
+        Only one task may receive through this instance at a time. Use
+        `.subscribe()` to give each concurrent consumer its own cursor
+        and loss/lag policy. `trio.BusyResourceError` identifies the
+        task which owns an already-active receive.
+
+        '''
+        if receive_task := self._receive_task:
+            raise trio.BusyResourceError(
+                'another task is already receiving from this '
+                '`BroadcastReceiver`\n'
+                f'active receive task: {receive_task.name!r}\n'
+                f'{receive_task!r}'
+            )
+
+        self._receive_task = trio.lowlevel.current_task()
+        try:
+            return await self._receive()
+        finally:
+            self._receive_task = None
+
+    async def _receive(self) -> ReceiveType:
         key = self.key
         state = self._state
 
@@ -362,7 +490,23 @@ class BroadcastReceiver(ReceiveChannel):
                 # seq = state.subs[key]
                 # assert seq == -1  # sanity
                 _, ev = state.recv_ready
-                await ev.wait()
+                wait_scope = trio.CancelScope()
+                self._wait_scope = wait_scope
+                try:
+                    with wait_scope:
+                        await ev.wait()
+
+                    # As with `recv_scope`, only this receiver's
+                    # `aclose()` cancels its private peer-wait scope
+                    # after marking the receiver closed. Outer task
+                    # cancellation remains `trio.Cancelled`.
+                    if wait_scope.cancel_called:
+                        assert self._closed
+                    if self._closed:
+                        raise trio.ClosedResourceError
+                finally:
+                    self._wait_scope = None
+
                 try:
                     return self.receive_nowait(
                         _key=key,
@@ -405,11 +549,12 @@ class BroadcastReceiver(ReceiveChannel):
 
     ) -> AsyncIterator[BroadcastReceiver]:
         '''
-        Subscribe for values from this broadcast receiver.
+        Create a receiver with its own logical subscription cursor.
 
-        Returns a new ``BroadCastReceiver`` which is registered for and
-        pulls data from a clone of the original
-        ``trio.abc.ReceiveChannel`` provided at creation.
+        The new `BroadcastReceiver` is registered against the shared
+        source and receives every retained value in sequence. Give each
+        concurrent consumer task its own receiver instead of sharing
+        one instance across overlapping `.receive()` calls.
 
         '''
         if self._closed:
@@ -440,18 +585,35 @@ class BroadcastReceiver(ReceiveChannel):
         if self._closed:
             return
 
-        # if there are sleeping consumers wake
-        # them on closure.
-        rr = self._state.recv_ready
-        if rr:
-            _, event = rr
-            event.set()
-
         # XXX: leaving it like this consumers can still get values
         # up to the last received that still reside in the queue.
-        self._state.subs.pop(self.key)
+        state = self._state
+        state.subs.pop(self.key)
+        state.cancelled.pop(self.key, None)
         self._closed = True
 
+        # A non-owner close must not wake peers waiting behind some
+        # other receiver's source read. If this receiver owns that
+        # read, cancel only its private scope; the owner task wakes
+        # peers after cancellation is delivered and state is ready
+        # for a clean ownership handoff.
+        rr = state.recv_ready
+        if (
+            rr is not None
+
+            # `recv_ready[0]` identifies the receiver which currently
+            # owns the one shared source read. Only that receiver may
+            # cancel `BroadcastState.recv_scope`; closing any other
+            # subscriber must not disturb the owner or its peer tasks.
+            and
+            rr[0] == self.key
+        ):
+            recv_scope = state.recv_scope
+            assert recv_scope is not None
+            recv_scope.cancel()
+        elif (wait_scope := self._wait_scope) is not None:
+            wait_scope.cancel()
+
 
 def broadcast_receiver(
 
@@ -462,6 +624,11 @@ def broadcast_receiver(
 
 ) -> BroadcastReceiver:
 
+    if max_buffer_size < 1:
+        raise ValueError(
+            '`max_buffer_size` must be greater than zero'
+        )
+
     return BroadcastReceiver(
         recv_chan,
         state=BroadcastState(