Add prompt-io entries for gh #470 latency work

Log the AI-assisted session per the NLNet generative-AI
policy: prompt, profiling findings, per-file diff pointers,
measured results and the unimplemented `pdbp`/`platformdirs`
deferral follow-ups.

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

ai/prompt-io/claude/20260702T155626Z_65bf9df5_prompt_io.md

@@ -0,0 +1,83 @@
+---
+model: claude-fable-5
+service: claude
+session: b6b42e23-0454-4d9b-be1c-0a89d65aed1b
+timestamp: 2026-07-02T15:56:26Z
+git_ref: 65bf9df5
+scope: code
+substantive: true
+raw_file: 20260702T155626Z_65bf9df5_prompt_io.raw.md
+---
+
+## Prompt
+
+From `ai/prompt-io/prompts/issue_470.md`:
+
+> attempt to resolve
+> https://github.com/goodboy/tractor/issues/470
+> do it with /open-wkt, ensure you /run-tests on all
+> changes
+
+(plus a hard pause-for-human-review deadline of
+12:50PM EST.)
+
+Issue #470 asks to trim the ~0.42s `import tractor`
+cost — which dominates per-actor spawn latency on the
+`trio` backend — by lazy-importing heavy/optional deps
+(`pdbp`, `stackscope`, `multiaddr`, + audit of
+`colorlog`/`bidict`/`wrapt`/`setproctitle`).
+
+## Response summary
+
+Profiling showed the issue's dep-list only accounted
+for ~20ms; the dominant cost (~244ms) was
+`log.get_logger()`'s `get_caller_mod()` calling
+`inspect.stack()` at module level in ~39 modules —
+each call walks every stack frame (deep during nested
+imports) and scans `sys.modules` per frame via
+`inspect.getmodule()`.
+
+Changes, in impact order:
+
+1. `get_caller_mod()` -> `sys._getframe()` +
+   `f_globals['__name__']` `sys.modules` lookup
+   (~240ms saved).
+2. Issue's lazy-import checklist: `bidict`,
+   `multiaddr`, `colorlog`, `wrapt` moved to
+   `TYPE_CHECKING`/function-local imports;
+   `platformdirs` function-local; `asyncio` +
+   `.to_asyncio` deferred out of the `devx.debug` +
+   `spawn._entry` eager paths (~15ms saved).
+3. PEP 562 `__getattr__` on `tractor/__init__.py`
+   preserving public `tractor.to_asyncio` attr access.
+
+Results: `import tractor` 0.42s -> ~0.145s (~65%);
+sequential `start_actor` latency 0.40-0.44s ->
+~0.179s/actor. `pdbp` (needs `_repl.py` class-base
+restructure) + `platformdirs` (needs
+`UDSAddress.def_bindspace` protocol rework) documented
+as follow-ups.
+
+## Files changed
+
+- `tractor/log.py` — `get_caller_mod()` perf fix +
+  lazy `colorlog`
+- `tractor/__init__.py` — PEP 562 lazy `to_asyncio`
+- `tractor/discovery/_addr.py` — `bidict` ->
+  `TYPE_CHECKING`
+- `tractor/discovery/_multiaddr.py` — lazy `multiaddr`
+- `tractor/ipc/_tcp.py`, `tractor/ipc/_uds.py` —
+  `Multiaddr` -> `TYPE_CHECKING`
+- `tractor/runtime/_state.py` — lazy `platformdirs`
+- `tractor/devx/_frame_stack.py` — lazy `pdbp` +
+  `wrapt`
+- `tractor/devx/debug/_trace.py`,
+  `tractor/devx/debug/_tty_lock.py` — lazy `asyncio` +
+  `.to_asyncio`
+- `tractor/spawn/_entry.py` — lazy
+  `run_as_asyncio_guest`
+
+## Human edits
+
+None yet — pending user review at the 12:50PM EST
+pause gate (test-suite results reported in-session).

ai/prompt-io/claude/20260702T155626Z_65bf9df5_prompt_io.raw.md

@@ -0,0 +1,122 @@
+---
+model: claude-fable-5
+service: claude
+timestamp: 2026-07-02T15:56:26Z
+git_ref: 65bf9df5
+diff_cmd: git diff main..wkt/boot_latency_470
+---
+
+# Raw AI output — gh #470 `import tractor` latency trim
+
+All generated code is committed on the
+`wkt/boot_latency_470` branch; per diff-ref mode each
+file's content is referenced via its diff instead of
+copied verbatim.
+
+## Profiling findings (verbatim analysis output)
+
+Baseline: `import tractor` ~0.39-0.42s wall.
+
+`python -X importtime` + `cProfile` traced the cost NOT
+primarily to third-party deps (the issue's hypothesis)
+but to `tractor/log.py:get_logger()` calling
+`get_caller_mod()` -> `inspect.stack()` at module level
+in ~39 tractor modules:
+
+- `inspect.stack()` builds `FrameInfo` (incl. src-file
+  and line-context resolution) for EVERY frame on the
+  stack; during nested imports the stack is dozens of
+  importlib frames deep.
+- each `FrameInfo` resolution calls
+  `inspect.getmodule()` which scans all of
+  `sys.modules` per frame (1.4M `ismodule()` calls in
+  one profiled import).
+- aggregate: ~244ms of tractor-own module "self" time
+  vs ~20ms for ALL the issue-listed third-party deps
+  (`pdbp` ~10ms, `bidict` ~4.5ms, `multiaddr` ~3.5ms,
+  `wrapt`/`colorlog` ~1ms each); `trio` itself is
+  ~70-100ms and unavoidable.
+
+## Generated changes
+
+> `git diff main..wkt/boot_latency_470 -- tractor/log.py`
+
+`get_caller_mod()` rewritten from `inspect.stack()` +
+`inspect.getmodule()` to `sys._getframe(frames_up)` +
+`frame.f_globals['__name__']` -> `sys.modules` lookup
+(O(1) vs O(stack x sys.modules)). Unused `inspect`
+imports dropped; `FrameType` imported from `types`.
+Also `colorlog` lazy-imported inside
+`get_console_log()`.
+
+> `git diff main..wkt/boot_latency_470 -- tractor/discovery/_addr.py`
+
+`bidict` import moved under `TYPE_CHECKING`
+(annotation-only use; `_address_types` is a plain dict
+literal).
+
+> `git diff main..wkt/boot_latency_470 -- tractor/discovery/_multiaddr.py`
+
+`from __future__ import annotations` added; `multiaddr`
+import moved under `TYPE_CHECKING` + function-local
+imports in `mk_maddr()`/`parse_maddr()`.
+
+> `git diff main..wkt/boot_latency_470 -- tractor/ipc/_tcp.py tractor/ipc/_uds.py`
+
+`Multiaddr` imports moved under `TYPE_CHECKING`
+(annotation-only in both transports).
+
+> `git diff main..wkt/boot_latency_470 -- tractor/runtime/_state.py`
+
+`platformdirs` lazy-imported inside `get_rt_dir()`
+(NOTE: still imported eagerly via
+`UDSAddress.def_bindspace` class-var eval; see
+follow-ups).
+
+> `git diff main..wkt/boot_latency_470 -- tractor/devx/_frame_stack.py`
+
+`pdbp` + `wrapt` lazy-imported inside
+`hide_runtime_frames()` / `api_frame()` respectively.
+
+> `git diff main..wkt/boot_latency_470 -- tractor/devx/debug/_trace.py tractor/devx/debug/_tty_lock.py`
+
+`asyncio` moved to `TYPE_CHECKING` + call-site local
+imports (`asyncio.current_task()` sites);
+`tractor.to_asyncio.run_trio_task_in_future` imports
+moved into the infected-aio runtime branches.
+
+> `git diff main..wkt/boot_latency_470 -- tractor/spawn/_entry.py`
+
+`run_as_asyncio_guest` import moved into the
+`infect_asyncio=True` branches of `_mp_main()` /
+`_trio_main()`.
+
+> `git diff main..wkt/boot_latency_470 -- tractor/__init__.py`
+
+PEP 562 module `__getattr__` added so
+`tractor.to_asyncio` attr-access still works (required
+by `tests/test_child_manages_service_nursery.py` and
+any downstream user) while keeping `asyncio` off the
+eager import path.
+
+## Measured results (verbatim)
+
+- `import tractor`: 0.39-0.42s -> ~0.145s (~65% cut)
+- `start_actor` spawn+boot+reg+cancel: ~0.40-0.44s ->
+  ~0.179s/actor (n=5 sequential, warm parent)
+- post-change eager-module check: only `pdbp` +
+  `platformdirs` of the issue's list remain eager.
+
+## Known follow-ups (not implemented, deadline-bound)
+
+- `pdbp` (~10ms): still eager via
+  `devx/debug/_repl.py` class bases
+  (`class PdbREPL(pdbp.Pdb)`) + `_tty_lock.py`
+  module-level `@pdbp.hideframe`; needs `_repl`
+  restructure + PEP 562 in `devx.debug.__init__`.
+- `platformdirs` (~1.5ms): eager via
+  `UDSAddress.def_bindspace: ClassVar = get_rt_dir()`
+  class-body call; needs `Address`-protocol rework of
+  `def_bindspace` to a lazy accessor.
+- `stackscope` + `setproctitle`: already lazy/absent —
+  no change needed.

ai/prompt-io/claude/20260702T161624Z_65bf9df5_prompt_io.md

@@ -0,0 +1,80 @@
+---
+model: claude-fable-5
+service: claude
+session: f4da6654-63f9-4641-b567-f200da46d697
+timestamp: 2026-07-02T16:16:24Z
+git_ref: 65bf9df5
+scope: tests
+substantive: true
+raw_file: 20260702T161624Z_65bf9df5_prompt_io.raw.md
+---
+
+## Prompt
+
+> attempt to resolve
+> https://github.com/goodboy/tractor/issues/474
+> do it with /open-wkt.
+
+Issue #474: "We need test suite for
+`.trionics.start_or_cancel()`" — dig out the use case
+from `modden` as a unit test (and provide motivation
+for upstream `trio` to possibly provide better
+porcelain around this case). The wrapper landed via
+PR #464 with no tests.
+
+## Response summary
+
+New `tests/trionics/test_taskc.py` (9 tests) covering
+`start_or_cancel()`, authored in wkt
+`start_or_cancel_tests_474`.
+
+Behaviour was first probed empirically (trio 0.29):
+the lossy `RuntimeError('child exited without calling
+task_status.started()')` only fires when the child
+exits pre-`.started()` WITHOUT propagating the ambient
+`Cancelled` — i.e. when the child (or lib code it
+calls) absorbs the cancel in a graceful-teardown
+pattern; a well-behaved child surfaces `Cancelled`
+straight out of `.start()`. The `modden`
+`progman.open_wks()` use case was reconstructed from
+`modden/runtime/progman.py` accordingly.
+
+Tests (each `use_start_or_cancel` parametrization also
+pins upstream trio's current lossy behaviour as
+wart-documentation):
+
+- `test_sibling_err_not_masked_by_startup_rte` — the
+  `modden` case: sibling error OOB-cancels the shared
+  nursery scope; with the wrapper ONLY the root-cause
+  `ValueError` escapes; bare `.start()` adds the lossy
+  RTE alongside.
+- `test_pure_oob_cancel_not_morphed_to_rte` — plain
+  ancestor `cs.cancel()`: wrapper → clean exit; bare
+  → eg-wrapped RTE.
+- `test_genuine_startup_rte_still_raised` — no
+  cancellation → protocol-bug RTE re-raised same as
+  bare.
+- `test_childs_own_rte_never_demoted_to_cancel` — a
+  child's own `RuntimeError('never got started!')` /
+  `RuntimeError(1234)` under ambient cancel is never
+  demoted to `Cancelled` (exact-msg-match + str-guard
+  regression cover).
+- `test_started_value_and_args_passthru` — happy path:
+  positional args, `name=`, `.started()` value.
+
+Verified: 9/9 pass; 0 flakes across 50 hammer runs;
+two impl mutations (checkpoint removed; guard relaxed
+to substring match) each caught by exactly the
+targeted tests; `tests/trionics/` +
+`tests/test_trioisms.py` subset green (23 passed,
+5 xfailed); ruff clean; 69-col style.
+
+## Files changed
+
+- `tests/trionics/test_taskc.py` — new
+  `start_or_cancel()` unit-test suite (gh #474).
+
+## Human edits
+
+Pending review — session paused pre-commit per user
+deadline; nothing committed as of this entry.

ai/prompt-io/claude/20260702T161624Z_65bf9df5_prompt_io.raw.md

@@ -0,0 +1,107 @@
+---
+model: claude-fable-5
+service: claude
+timestamp: 2026-07-02T16:16:24Z
+git_ref: 65bf9df5
+diff_cmd: git diff main..wkt/start_or_cancel_tests_474
+---
+
+# Raw output — gh #474 `start_or_cancel()` test suite
+
+## Generated test code
+
+> `git diff main..wkt/start_or_cancel_tests_474 -- tests/trionics/test_taskc.py`
+
+Prose summary of the generated module
+(`tests/trionics/test_taskc.py`):
+
+- module docstring framing the `trio.Nursery.start()`
+  startup-cancellation wart, the wrapper's repair, and
+  the intent that `use_start_or_cancel=False` params
+  double as upstream-trio wart-documentation (break on
+  a trio upgrade → upstream may have shipped porcelain,
+  re-audit the wrapper); cites gh #474 / PR #464 and
+  `modden`'s `progman.open_wks()` as the source use
+  case.
+- shared children: `absorbs_cancel_pre_started()` (the
+  graceful-teardown cancel-absorber which triggers the
+  lossy RTE path) + `raise_val_err()` (fast-erroring
+  sibling).
+- `test_sibling_err_not_masked_by_startup_rte`
+  (parametrized `use_start_or_cancel`): asserts eg
+  contains exactly one `ValueError` and, wrapper-case,
+  NO residual RTE (`eg.split(ValueError)` remainder is
+  `None`); bare-case, the residual RTE carries trio's
+  exact "child exited without calling" wording.
+- `test_pure_oob_cancel_not_morphed_to_rte`
+  (parametrized): wrapper-case runs clean and asserts
+  `cs.cancelled_caught`; bare-case asserts the
+  eg-wrapped RTE.
+- `test_genuine_startup_rte_still_raised`
+  (parametrized): no-cancel protocol bug → RTE with
+  trio's wording from both call forms.
+- `test_childs_own_rte_never_demoted_to_cancel`
+  (parametrized `rte_arg` in `'never got started!'`,
+  `1234`): child cancels the ambient scope then raises
+  its own RTE synchronously (no checkpoint between →
+  deterministically under-cancellation at catch time);
+  asserts the RTE survives with `args[0]` intact.
+- `test_started_value_and_args_passthru`: `.started()`
+  value, positional args and the `name=` kwarg (via
+  `trio.lowlevel.current_task().name`) all forward.
+
+## Non-code output (verbatim highlights)
+
+Behaviour probe (trio 0.29, scratchpad scripts) — the
+decision basis for the test shapes:
+
+```
+== B-sibling-err use_soc=False
+  start raised: RuntimeError('child exited without
+    calling task_status.started()')
+  top-level: ExceptionGroup([ValueError('sibling blew
+    up!'), RuntimeError('child exited without calling
+    task_status.started()')])
+== B-cs-cancel use_soc=False
+  top-level: ExceptionGroup([RuntimeError('child
+    exited without calling task_status.started()')])
+== B-sibling-err use_soc=True
+  start raised: Cancelled()
+  top-level: ExceptionGroup([ValueError('sibling blew
+    up!')])
+== B-cs-cancel use_soc=True
+  start raised: Cancelled()
+  top-level: clean return
+== own-rte-under-cancel (both) -> RTE('never got
+  started!') propagates unchanged
+```
+
+Key finding: with a WELL-BEHAVED (non-absorbing) child
+an OOB ancestor cancel surfaces `Cancelled` directly
+from `.start()` on trio 0.29 — the lossy RTE requires
+the child to absorb its cancel pre-`.started()`, which
+is what `modden`'s `open_from_wks` teardown did. Trio's
+nursery-exit wait defers cancel delivery to children,
+so all tested shapes are deterministic (0 flakes / 50
+runs).
+
+Mutation verification:
+
+```
+mutation 1 (checkpoint_if_cancelled removed):
+  FAILED test_sibling_err_not_masked_by_startup_rte[True]
+  FAILED test_pure_oob_cancel_not_morphed_to_rte[True]
+mutation 2 (guard relaxed to 'started' substring,
+            isinstance dropped):
+  FAILED test_childs_own_rte_never_demoted_to_cancel[never got started!]
+  FAILED test_childs_own_rte_never_demoted_to_cancel[1234]
+```
+
+Final runs:
+
+```
+tests/trionics/test_taskc.py: 9 passed in 0.03s
+hammer: 0/50 runs failed
+tests/trionics/ + tests/test_trioisms.py:
+  23 passed, 5 xfailed in 3.02s
+```

ai/prompt-io/opencode/20260804T030309Z_65bf9df5_prompt_io.md

@@ -0,0 +1,39 @@
+---
+model: openai/gpt-5.6-sol
+service: opencode
+session: moc-teardown-completion-20260804
+timestamp: 2026-08-04T03:03:09Z
+git_ref: 65bf9df5
+scope: code
+substantive: true
+raw_file: 20260804T030309Z_65bf9df5_prompt_io.raw.md
+---
+
+## Prompt
+
+Patch Tractor's `maybe_open_context()` so the final consumer waits for
+resource `__aexit__()` completion and receives cleanup errors. Reuse
+`outcome.Outcome` for the exit result; use the smaller mutable
+`_CtxExit` holder if that makes the implementation simpler. Add and run
+the relevant existing unit tests, but do not commit or push the patch.
+After reviewing the result, simplify `_CtxExit` back to an optional
+exception because the success outcome carries no useful value.
+
+## Response summary
+
+Added an exception-backed completion handshake between
+`_Cache.run_ctx()` and the final `maybe_open_context()` consumer.
+Serialized consumer registration and final teardown under the per-key
+lock, preserving that lock for queued entrants. Added deterministic
+regressions for normal exit, cleanup errors, cancellation interactions,
+service-nursery cancellation, and teardown re-entry.
+
+## Files changed
+
+- `tractor/trionics/_mngrs.py` - publish and unwrap cached exit outcomes.
+- `tests/test_resource_cache.py` - cover completion and cancellation.
+
+## Human edits
+
+The user directed the final simplification from `outcome.Outcome` to an
+optional exception field. The patch remains uncommitted.

ai/prompt-io/opencode/20260804T030309Z_65bf9df5_prompt_io.raw.md

@@ -0,0 +1,36 @@
+---
+model: openai/gpt-5.6-sol
+service: opencode
+timestamp: 2026-08-04T03:03:09Z
+git_ref: 65bf9df5
+diff_cmd: git diff HEAD~1..HEAD
+---
+
+Implemented cached-context exit completion in
+`tractor.trionics.maybe_open_context()`.
+
+> `git diff HEAD~1..HEAD -- tractor/trionics/_mngrs.py`
+
+The generated implementation adds `_CtxExit`, whose `done` event
+publishes an `outcome.Outcome[None]`. `_Cache.run_ctx()` records either
+`Value(None)` or `Error(exc)` after the resource exit attempt. The final
+MOC consumer signals `no_more_users`, waits for completion under a
+shielded cancel scope, removes the per-key lock, and unwraps the outcome
+so ordinary cleanup failures are raised at the consumer boundary.
+`trio.Cancelled`, `KeyboardInterrupt`, and `SystemExit` continue through
+the service task rather than being converted into regular cleanup
+errors.
+
+> `git diff HEAD~1..HEAD -- tests/test_resource_cache.py`
+
+The generated regressions cover successful exit blocking, cleanup-error
+delivery, final-user cancellation, cancellation combined with a cleanup
+error, and service-nursery cancellation. The existing teardown re-entry
+test now uses explicit events instead of a ten-second cleanup sleep and
+asserts that the replacement resource is a fresh cache miss.
+
+Verification:
+
+`env PYTHONPATH="$PWD" /home/goodboy/repos/tractor/py313/bin/python -m pytest tests/test_resource_cache.py`
+
+Result: `16 passed in 8.37s`.

ai/prompt-io/opencode/README.md

@@ -0,0 +1,22 @@
+# AI Prompt I/O Log - OpenCode
+
+This directory tracks prompt inputs and model outputs for AI-assisted
+development using `opencode`.
+
+## Policy
+
+Prompt logging follows the [NLNet generative AI policy][nlnet-ai]. All
+substantive AI contributions are logged with:
+
+- Model name and version
+- Timestamps
+- The prompts that produced the output
+- Unedited model output (`.raw.md` files)
+
+[nlnet-ai]: https://nlnet.nl/foundation/policies/generativeAI/
+
+## Usage
+
+Entries are created by the prompt-io workflow. Human contributors remain
+accountable for all decisions. AI-generated content is never presented as
+human-authored work.

tests/test_log_sys.py

@@ -2,16 +2,19 @@
 `tractor.log`-wrapping unit tests.
 
 '''
+import logging
 from pathlib import Path
 import shutil
 from types import ModuleType
 
 import pytest
 import tractor
+import trio
 from tractor import (
     _code_load,
     log,
 )
+from tractor.ipc import _chan
 
 
 def test_root_pkg_not_duplicated_in_logger_name():
@@ -222,6 +225,88 @@ def test_add_log_level_pluggable():
             delattr(log.StackLevelAdapter, name.lower())
 
 
+@pytest.mark.parametrize(
+    'suppression',
+    [
+        'level',
+        'logger',
+        'global',
+    ],
+)
+def test_log_guard_skips_payload_formatting(
+    monkeypatch: pytest.MonkeyPatch,
+    suppression: str,
+):
+    '''
+    Suppressed transport logs must not render payloads.
+
+    The original hot-path guard compared only the effective logger
+    level. A logger disabled through its `Logger.disabled` flag or
+    the global `logging.disable()` threshold could therefore still
+    call `pformat()` before `Logger.isEnabledFor()` discarded the
+    record.
+
+    Exercise effective-level, per-logger, and global suppression
+    independently. A poisoned `_chan.pformat()` proves rendering is
+    skipped, while the fake transport proves `Channel.send()` still
+    transmits the original payload and traceback-hiding flag.
+
+    '''
+    sent: list[tuple[object, bool]] = []
+
+    class FakeTransport:
+        async def send(
+            self,
+            payload: object,
+            hide_tb: bool = False,
+        ) -> None:
+            sent.append((payload, hide_tb))
+
+    def fail_pformat(payload: object) -> str:
+        raise AssertionError(
+            f'suppressed log rendered payload: {payload!r}'
+        )
+
+    chan_log = log.get_logger(
+        name=f'guard_test.{suppression}',
+    )
+    std_log = chan_log.logger
+    orig_level: int = std_log.level
+    orig_disable: int = logging.root.manager.disable
+    transport_level: int = log.CUSTOM_LEVELS['TRANSPORT']
+
+    monkeypatch.setattr(_chan, 'log', chan_log)
+    monkeypatch.setattr(_chan, 'pformat', fail_pformat)
+    try:
+        logging.disable(logging.NOTSET)
+        std_log.setLevel(transport_level)
+
+        if suppression == 'level':
+            std_log.setLevel(logging.INFO)
+        elif suppression == 'logger':
+            monkeypatch.setattr(std_log, 'disabled', True)
+        else:
+            logging.disable(logging.CRITICAL)
+
+        assert not chan_log.isEnabledFor(transport_level)
+
+        transport = FakeTransport()
+        chan = _chan.Channel(transport=transport)
+        payload = object()
+
+        async def send_payload() -> None:
+            await chan.send(
+                payload,
+                hide_tb=True,
+            )
+
+        trio.run(send_payload)
+        assert sent == [(payload, True)]
+    finally:
+        std_log.setLevel(orig_level)
+        logging.disable(orig_disable)
+
+
 # TODO, moar tests against existing feats:
 # ------ - ------
 # - [ ] color settings?

tests/test_resource_cache.py

@@ -9,6 +9,7 @@ from typing import Awaitable
 
 import pytest
 import trio
+from trio.testing import wait_all_tasks_blocked
 import tractor
 from tractor.trionics import (
     maybe_open_context,
@@ -94,6 +95,232 @@ def test_resource_only_entered_once(key_on):
     trio.run(main)
 
 
+def test_last_moc_user_waits_for_resource_exit():
+    '''
+    Verify the final user cannot return before resource teardown.
+
+    Previously the final `maybe_open_context()` user only signalled
+    `_Cache.run_ctx()` through its `no_more_users` event. The user
+    then returned while the service task was still running the
+    resource's `__aexit__()`, so callers could observe stale external
+    state immediately after their `async with` block.
+
+    The resource sets `exit_started` before blocking on
+    `allow_exit`. The user task must remain inside MOC until the test
+    releases that deterministic checkpoint and `__aexit__()` sets
+    `exit_finished`.
+
+    '''
+    async def main():
+        exit_started = trio.Event()
+        allow_exit = trio.Event()
+        exit_finished = trio.Event()
+        user_returned = trio.Event()
+
+        @acm
+        async def open_resource():
+            try:
+                yield
+            finally:
+                exit_started.set()
+                await allow_exit.wait()
+                exit_finished.set()
+
+        async def use_resource():
+            async with maybe_open_context(open_resource):
+                pass
+
+            assert exit_finished.is_set()
+            user_returned.set()
+
+        async with (
+            tractor.open_root_actor(),
+            trio.open_nursery() as tn,
+        ):
+            tn.start_soon(use_resource)
+            await exit_started.wait()
+            assert not user_returned.is_set()
+            allow_exit.set()
+            await user_returned.wait()
+
+    trio.run(main)
+
+
+def test_moc_delivers_resource_exit_error():
+    '''
+    Verify a resource exit error reaches the final MOC user.
+
+    Previously `_Cache.run_ctx()` executed the cached resource's
+    `__aexit__()` after the final user had returned. An exit failure
+    therefore surfaced later through the actor service nursery rather
+    than at the user's `async with maybe_open_context()` boundary.
+
+    This resource raises a unique `ResourceExitError` during exit.
+    Catching that exact instance around MOC proves the service task
+    delivered the failure to the final user without replacing it.
+
+    '''
+    class ResourceExitError(Exception):
+        pass
+
+    exit_error = ResourceExitError('resource exit failed')
+
+    async def main():
+        @acm
+        async def open_resource():
+            yield
+            raise exit_error
+
+        async with tractor.open_root_actor():
+            with pytest.raises(ResourceExitError) as exc_info:
+                async with maybe_open_context(open_resource):
+                    pass
+
+            assert exc_info.value is exit_error
+
+    trio.run(main)
+
+
+def test_moc_final_user_cancellation_waits_for_exit():
+    '''
+    Verify final-user cancellation still waits for successful exit.
+
+    Previously cancellation escaped the final MOC user immediately
+    after it signalled `_Cache.run_ctx()`, leaving resource exit to
+    finish later in the actor service task. This violated the context
+    manager boundary even when cleanup itself succeeded.
+
+    The consumer cancels its own scope while holding the sole cached
+    resource. The resource sets `exit_finished` from its `finally`
+    block, and the consumer checks that event immediately after its
+    cancel scope catches `trio.Cancelled`. This proves MOC's
+    completion wait is shielded without suppressing the original
+    cancellation.
+
+    '''
+    async def main():
+        exit_finished = trio.Event()
+
+        @acm
+        async def open_resource():
+            try:
+                yield
+            finally:
+                exit_finished.set()
+
+        async with tractor.open_root_actor():
+            with trio.CancelScope() as cs:
+                async with maybe_open_context(open_resource):
+                    cs.cancel()
+                    await trio.sleep_forever()
+
+            assert cs.cancelled_caught
+            assert exit_finished.is_set()
+
+    trio.run(main)
+
+
+def test_moc_exit_error_masks_final_user_cancellation():
+    '''
+    Verify cleanup errors survive final-user cancellation.
+
+    A cancelled final user previously signalled `no_more_users` and
+    propagated `trio.Cancelled` before `_Cache.run_ctx()` completed
+    resource exit. If `__aexit__()` then failed, its error was
+    detached from the API call which caused teardown.
+
+    The consumer cancels its own scope at a deterministic checkpoint
+    inside MOC. Resource exit raises `ResourceExitError`; observing
+    that exact error outside the cancel scope proves MOC shields the
+    completion wait and applies normal context-manager masking, where
+    a cleanup failure replaces the active cancellation.
+
+    '''
+    class ResourceExitError(Exception):
+        pass
+
+    exit_error = ResourceExitError('resource exit failed')
+
+    async def main():
+        @acm
+        async def open_resource():
+            yield
+            raise exit_error
+
+        async with tractor.open_root_actor():
+            with pytest.raises(ResourceExitError) as exc_info:
+                with trio.CancelScope() as cs:
+                    async with maybe_open_context(open_resource):
+                        cs.cancel()
+                        await trio.sleep_forever()
+
+            assert exc_info.value is exit_error
+
+    trio.run(main)
+
+
+def test_moc_service_nursery_cancellation_completes_exit():
+    '''
+    Verify service-nursery cancellation cannot strand a final user.
+
+    `_Cache.run_ctx()` and an MOC consumer may share a
+    caller-provided service nursery. Cancelling that nursery
+    interrupts the service task's `no_more_users` wait and the
+    consumer body together. A shielded final-user wait would deadlock
+    if `run_ctx()` failed to publish completion while propagating its
+    own `trio.Cancelled`.
+
+    The outer task waits for resource entry, then cancels the exact
+    nursery containing both tasks. The resource shields one cleanup
+    checkpoint and sets `exit_finished`; observing both it and
+    `service_finished` proves cancellation propagated normally while
+    MOC's completion handshake terminated deterministically.
+
+    '''
+    async def main():
+        resource_entered = trio.Event()
+        exit_finished = trio.Event()
+        service_finished = trio.Event()
+        service_tn: trio.Nursery|None = None
+
+        @acm
+        async def open_resource():
+            try:
+                resource_entered.set()
+                yield
+            finally:
+                with trio.CancelScope(shield=True):
+                    await trio.lowlevel.checkpoint()
+                    exit_finished.set()
+
+        async def use_resource(tn: trio.Nursery):
+            async with maybe_open_context(
+                open_resource,
+                tn=tn,
+            ):
+                await trio.sleep_forever()
+
+        async def run_service():
+            nonlocal service_tn
+
+            async with trio.open_nursery() as tn:
+                service_tn = tn
+                tn.start_soon(use_resource, tn)
+
+            service_finished.set()
+
+        async with trio.open_nursery() as outer_tn:
+            outer_tn.start_soon(run_service)
+            await resource_entered.wait()
+            assert service_tn is not None
+            service_tn.cancel_scope.cancel()
+            await service_finished.wait()
+
+        assert exit_finished.is_set()
+
+    trio.run(main)
+
+
 @tractor.context
 async def streamer(
     ctx: tractor.Context,
@@ -548,43 +775,52 @@ def test_moc_reentry_during_teardown(
     loglevel: str,
 ):
     '''
-    Reproduce the piker `open_cached_client('kraken')` race:
+    Reproduce re-entry while an identical cached context exits.
 
-    - same `acm_func`, NO kwargs (identical `ctx_key`)
+    - multiple tasks use the same `acm_func` with no kwargs,
-    - multiple tasks share the cached resource
+      producing an identical `ctx_key`;
-    - all users exit -> teardown starts
+    - all users leave and the final user starts resource teardown;
-    - a NEW task enters during `_Cache.run_ctx.__aexit__`
+    - `_Cache.run_ctx()` removes the cached value and resource entry
-    - `values[ctx_key]` is gone (popped in inner finally)
+      before entering the resource's blocking `__aexit__()` body;
-      but `resources[ctx_key]` still exists (outer finally
+    - a new task attempts to enter that same `ctx_key` during exit;
-      hasn't run yet bc the acm cleanup has checkpoints)
+    - the per-key lock keeps that entrant queued until exit
-    - old code: `assert not resources.get(ctx_key)` FIRES
+      completes;
+    - the entrant then receives a fresh cache miss and resource.
 
-    This models the real-world scenario where `brokerd.kraken`
+    Without teardown sharing the registration lock, re-entry could
-    tasks concurrently call `open_cached_client('kraken')`
+    race resource replacement while the prior generation was still
-    (same `acm_func`, empty kwargs, shared `ctx_key`) and
+    exiting. The final user could also return before that exit
-    the teardown/re-entry race triggers intermittently.
+    completed.
+
+    The first resource generation signals `in_aexit` and waits on
+    `allow_aexit`. The re-entry task signals `reentry_started` and
+    blocks inside MOC; only after `wait_all_tasks_blocked()` confirms
+    that ordering does the coordinator release cleanup. The entrant
+    must then receive a fresh cache miss. `first_done` additionally
+    proves the first MOC user observed completed teardown before
+    returning.
 
     '''
     async def main():
         in_aexit = trio.Event()
+        allow_aexit = trio.Event()
+        reentry_started = trio.Event()
+        generation: int = 0
 
         @acm
         async def cached_client():
             '''
-            Simulates `kraken.api.get_client()`:
+            Simulate a no-argument `kraken.api.get_client()`.
-            - no params (all callers share one `ctx_key`)
-            - slow-ish cleanup to widen the race window
-              between `values.pop()` and `resources.pop()`
-              inside `_Cache.run_ctx`.
 
             '''
+            nonlocal generation
+
+            generation += 1
+            resource_generation: int = generation
             yield 'the-client'
-            # Signal that we're in __aexit__ — at this
+            if resource_generation == 1:
-            # point `values` has already been popped by
+                in_aexit.set()
-            # `run_ctx`'s inner finally, but `resources`
+                await allow_aexit.wait()
-            # is still alive (outer finally hasn't run).
-            in_aexit.set()
-            await trio.sleep(10)
 
         first_done = trio.Event()
 
@@ -598,16 +834,25 @@ def test_moc_reentry_during_teardown(
         async def reenter_during_teardown():
             '''
             Wait for the acm's `__aexit__` to start (meaning
-            `values` is popped but `resources` still exists),
+            the cached value is no longer available), then re-enter.
-            then re-enter — triggering the assert.
 
             '''
             await in_aexit.wait()
+
+            # Tell the coordinator this task is about to enter MOC.
+            # `Event.set()` is not a checkpoint. Though `async with`
+            # awaits MOC's `__aenter__()`, its async generator runs
+            # synchronously until the held per-key `lock.acquire()`
+            # actually suspends this task.
+            reentry_started.set()
             async with maybe_open_context(
                 cached_client,
             ) as (cache_hit, value):
+                assert not cache_hit
                 assert value == 'the-client'
 
+            await first_done.wait()
+
         with trio.fail_after(5):
             async with (
                 tractor.open_root_actor(
@@ -619,5 +864,15 @@ def test_moc_reentry_during_teardown(
             ):
                 tn.start_soon(use_and_exit)
                 tn.start_soon(reenter_during_teardown)
+                await reentry_started.wait()
+
+                # Wait until the re-entry task is queued on MOC's
+                # per-key lock while `_Cache.run_ctx()` remains
+                # blocked in the first generation's `__aexit__()`.
+                # Only then release cleanup, making the intended
+                # enter-during-sibling-exit ordering deterministic.
+                await wait_all_tasks_blocked()
+                assert not first_done.is_set()
+                allow_aexit.set()
 
     trio.run(main)

tests/trionics/test_taskc.py

@@ -0,0 +1,333 @@
+'''
+`tractor.trionics._taskc.start_or_cancel()` unit tests.
+
+`trio.Nursery.start()` collapses an out-of-band (ancestor)
+cancellation into a lossy,
+
+  `RuntimeError('child exited without calling
+  task_status.started()')`
+
+whenever the started child exits pre-`.started()` WITHOUT
+propagating the ambient `trio.Cancelled`; a common outcome
+when the child (or any lib code it calls) runs a graceful
+teardown which absorbs the cancel and returns early. Our
+`start_or_cancel()` wrapper re-surfaces the real in-flight
+cancellation in that case so the true root error/cancel
+propagates to the `.start()` caller instead.
+
+These tests verify both that repair AND document upstream
+`trio`'s current lossy behaviour via the
+`use_start_or_cancel=False` parametrizations; if a `trio`
+upgrade breaks one of THOSE cases it likely means upstream
+shipped better startup-cancellation porcelain and our
+wrapper deserves a re-audit!
+
+The core use case was dug out of `modden`'s
+`progman.open_wks()` program-spawn machinery as per gh
+issue #474; the wrapper landed originally via gh PR #464.
+
+'''
+import pytest
+import trio
+from trio import TaskStatus
+
+from tractor.trionics import start_or_cancel
+
+
+async def absorbs_cancel_pre_started(
+    task_status: TaskStatus[None] = trio.TASK_STATUS_IGNORED,
+):
+    '''
+    Swallow the ambient (ancestor-scope) cancel and return
+    early, a naughty-but-realistic graceful-teardown pattern
+    and the exact shape which causes `trio.Nursery.start()`
+    to raise its lossy startup `RuntimeError` in place of
+    the real `trio.Cancelled`.
+
+    '''
+    try:
+        await trio.sleep_forever()
+    except trio.Cancelled:
+        return
+
+
+async def raise_val_err():
+    '''
+    Sibling task which blows up (fast) thus OOB-cancelling
+    the shared parent-nursery's cancel-scope.
+
+    '''
+    await trio.lowlevel.checkpoint()
+    raise ValueError('sibling blew up!')
+
+
+@pytest.mark.parametrize(
+    'use_start_or_cancel',
+    [
+        True,
+        False,
+    ],
+)
+def test_sibling_err_not_masked_by_startup_rte(
+    use_start_or_cancel: bool,
+):
+    '''
+    The `modden.runtime.progman` use case: a sibling task
+    errors while the `.start()`-ed child is still
+    pre-`.started()`, OOB-cancelling the shared nursery
+    scope; the child absorbs its cancel (graceful teardown)
+    and exits early.
+
+    - with `start_or_cancel()` the in-flight cancellation
+      is re-surfaced as the real `trio.Cancelled` (then
+      absorbed by the cancelled nursery scope) so ONLY the
+      root-cause sibling error escapes the nursery.
+
+    - with a bare `.start()`, upstream `trio` (currently)
+      also delivers its lossy startup `RuntimeError`
+      alongside, obscuring that the child was in fact
+      cancelled due to the sibling's error.
+
+    `cancelled_at_start` records that the wrapper's own await
+    raises `Cancelled`, rather than merely relying on the
+    nursery's eventual exception-group shape.
+
+    '''
+    cancelled_at_start: list[bool] = []
+
+    async def main():
+        async with trio.open_nursery() as tn:
+            tn.start_soon(raise_val_err)
+            if use_start_or_cancel:
+                try:
+                    await start_or_cancel(
+                        tn,
+                        absorbs_cancel_pre_started,
+                    )
+                except trio.Cancelled:
+                    cancelled_at_start.append(True)
+                    raise
+            else:
+                await tn.start(absorbs_cancel_pre_started)
+
+    with pytest.raises(ExceptionGroup) as excinfo:
+        trio.run(main)
+
+    eg: ExceptionGroup = excinfo.value
+    val_eg, rest_eg = eg.split(ValueError)
+    assert len(val_eg.exceptions) == 1
+
+    if use_start_or_cancel:
+        assert cancelled_at_start == [True]
+        # the re-surfaced `Cancelled` is absorbed by the
+        # (sibling-error cancelled) nursery scope leaving
+        # NO startup-noise, just the root cause.
+        assert rest_eg is None
+    else:
+        # the `trio` wart: a lossy startup RTE rides along
+        # with (and distracts from) the root cause.
+        rte = rest_eg.exceptions[0]
+        assert isinstance(rte, RuntimeError)
+        assert 'child exited without calling' in rte.args[0]
+
+
+@pytest.mark.parametrize(
+    'use_start_or_cancel',
+    [
+        True,
+        False,
+    ],
+)
+def test_pure_oob_cancel_not_morphed_to_rte(
+    use_start_or_cancel: bool,
+):
+    '''
+    A plain (error-free) ancestor `CancelScope.cancel()`
+    fired while the (cancel-absorbing) child is still
+    pre-`.started()`:
+
+    - `start_or_cancel()` re-surfaces the `Cancelled` so
+      the cancelled scope exits CLEAN, no error at all.
+
+    - a bare `.start()` (currently) morphs the plain
+      cancel into an (eg-wrapped) startup `RuntimeError`.
+
+    `cancelled_at_start` proves cancellation interrupts the
+    wrapper call itself before the cancelled scope exits.
+
+    '''
+    cancelled_at_start: list[bool] = []
+
+    async def main():
+        with trio.CancelScope() as cs:
+            async with trio.open_nursery() as tn:
+
+                async def canceller():
+                    await trio.lowlevel.checkpoint()
+                    cs.cancel()
+
+                tn.start_soon(canceller)
+                if use_start_or_cancel:
+                    try:
+                        await start_or_cancel(
+                            tn,
+                            absorbs_cancel_pre_started,
+                        )
+                    except trio.Cancelled:
+                        cancelled_at_start.append(True)
+                        raise
+                else:
+                    await tn.start(
+                        absorbs_cancel_pre_started,
+                    )
+
+        assert cs.cancelled_caught
+
+    if use_start_or_cancel:
+        trio.run(main)
+        assert cancelled_at_start == [True]
+    else:
+        with pytest.raises(ExceptionGroup) as excinfo:
+            trio.run(main)
+
+        rte = excinfo.value.exceptions[0]
+        assert isinstance(rte, RuntimeError)
+        assert 'child exited without calling' in rte.args[0]
+
+
+@pytest.mark.parametrize(
+    'use_start_or_cancel',
+    [
+        True,
+        False,
+    ],
+)
+def test_genuine_startup_rte_still_raised(
+    use_start_or_cancel: bool,
+):
+    '''
+    Absent ANY in-flight cancellation, a child exiting
+    cleanly without calling `task_status.started()` is a
+    genuine startup-protocol bug; `start_or_cancel()` must
+    re-raise the resulting `RuntimeError` exactly like a
+    bare `.start()` does.
+
+    '''
+    async def exits_wo_started(
+        task_status: TaskStatus[None] = (
+            trio.TASK_STATUS_IGNORED
+        ),
+    ):
+        await trio.lowlevel.checkpoint()
+
+    async def main():
+        async with trio.open_nursery() as tn:
+            with pytest.raises(RuntimeError) as excinfo:
+                if use_start_or_cancel:
+                    await start_or_cancel(
+                        tn,
+                        exits_wo_started,
+                    )
+                else:
+                    await tn.start(exits_wo_started)
+
+            rte = excinfo.value
+            assert (
+                'child exited without calling'
+                in
+                rte.args[0]
+            )
+
+    trio.run(main)
+
+
+@pytest.mark.parametrize(
+    'rte_arg',
+    [
+        # Broad substring matches would wrongly demote either
+        # child-owned error to `Cancelled` under cancellation.
+        'never got started!',
+        'child exited without calling user hook',
+        # non-`str` first-arg edge; must not `TypeError`
+        # inside the wrapper's msg-match guard.
+        1234,
+    ],
+)
+def test_childs_own_rte_never_demoted_to_cancel(
+    rte_arg: str|int,
+):
+    '''
+    A child's OWN `RuntimeError`, one which merely smells
+    like `trio`'s startup wording (or carries a non-`str`
+    first arg), raised under ambient cancellation must NOT
+    be demoted to a `trio.Cancelled` by the exact-msg-match
+    guard inside `start_or_cancel()`; the real error must
+    always propagate to the caller as the sole exception-group
+    leaf, preserving object identity.
+
+    '''
+    child_rte = RuntimeError(rte_arg)
+
+    async def cancels_cs_then_raises(
+        task_status: TaskStatus[None] = (
+            trio.TASK_STATUS_IGNORED
+        ),
+    ):
+        # cancel the ambient (ancestor) scope then raise
+        # sync-ly, no checkpoint between, so the child
+        # deterministically dies with ITS error while the
+        # caller is under effective cancellation.
+        cs.cancel()
+        raise child_rte
+
+    cs = trio.CancelScope()
+
+    async def main():
+        with cs:
+            async with trio.open_nursery() as tn:
+                await start_or_cancel(
+                    tn,
+                    cancels_cs_then_raises,
+                )
+
+    with pytest.raises(ExceptionGroup) as excinfo:
+        trio.run(main)
+
+    assert excinfo.value.exceptions == (child_rte,)
+
+
+def test_started_value_and_args_passthru():
+    '''
+    Happy path: positional args, the `name=` kwarg and the
+    `.started(value)`-delivered value all pass through
+    `start_or_cancel()` identically to a bare `.start()`.
+
+    '''
+    async def echo_started(
+        *args,
+        task_status: TaskStatus[tuple] = (
+            trio.TASK_STATUS_IGNORED
+        ),
+    ):
+        task_name: str = trio.lowlevel.current_task().name
+        task_status.started((
+            args,
+            task_name,
+        ))
+
+    async def main():
+        async with trio.open_nursery() as tn:
+            (
+                args,
+                task_name,
+            ) = await start_or_cancel(
+                tn,
+                echo_started,
+                'chillin',
+                10,
+                name='doggy',
+            )
+            assert args == ('chillin', 10)
+            assert task_name == 'doggy'
+
+    trio.run(main)

tractor/__init__.py

@@ -74,3 +74,26 @@ from .discovery._registry import (
     Arbiter as Arbiter,
 )
 # from . import hilevel as hilevel
+
+
+def __getattr__(name: str):
+    '''
+    PEP 562 lazy sub-module loading, presently only for
+    `.to_asyncio` which (transitively) imports `asyncio`
+    itself: a non-trivial multi-ms chunk of the eager
+    `import tractor` cost (gh #470) unneeded by
+    `trio`-only apps.
+
+    Any `tractor.to_asyncio.<attr>` access (or a
+    `from tractor import to_asyncio`) still works, the
+    sub-mod is simply imported on first-access instead
+    of at pkg-import time.
+
+    '''
+    if name == 'to_asyncio':
+        from importlib import import_module
+        return import_module('.to_asyncio', __name__)
+
+    raise AttributeError(
+        f'module {__name__!r} has no attribute {name!r}'
+    )

tractor/devx/_frame_stack.py

@@ -39,14 +39,15 @@ from typing import (
     Type,
 )
 
-import pdbp
+# NOTE, `pdbp` + `wrapt` are lazy-imported at their
+# single use-sites below to keep them off the eager
+# `import tractor` path (gh #470).
 from tractor.log import get_logger
 import trio
 from tractor.msg import (
     pretty_struct,
     NamespacePath,
 )
-import wrapt
 
 
 log = get_logger()
@@ -257,6 +258,7 @@ def api_frame(
     caller_frames_up: int = 1,
 
 ) -> Callable:
+    import wrapt
 
     # handle the decorator called WITHOUT () case,
     # i.e. just @api_frame, NOT @api_frame(extra=<blah>)
@@ -320,6 +322,8 @@ def hide_runtime_frames() -> dict[FunctionType, CodeType]:
     as possible, particularly from inside a `PdbREPL`.
 
     '''
+    import pdbp
+
     # XXX HACKZONE XXX
     #  hide exit stack frames on nurseries and cancel-scopes!
     # |_ so avoid seeing it when the `pdbp` REPL is first engaged from

tractor/devx/debug/_trace.py

@@ -24,7 +24,6 @@ mult-process support within a single actor tree.
 
 '''
 from __future__ import annotations
-import asyncio
 import bdb
 from contextlib import (
     AbstractContextManager,
@@ -53,7 +52,6 @@ from trio import (
 )
 import tractor
 from tractor.log import get_logger
-from tractor.to_asyncio import run_trio_task_in_future
 from tractor._context import Context
 from tractor.runtime import _state
 from tractor._exceptions import (
@@ -85,6 +83,10 @@ from ..pformat import (
 )
 
 if TYPE_CHECKING:
+    # NOTE, `asyncio` (and `.to_asyncio`) are
+    # lazy-imported at their use-sites to keep them off
+    # the eager `import tractor` path (gh #470).
+    import asyncio
     from trio.lowlevel import Task
     from threading import Thread
     from tractor.runtime._runtime import (
@@ -164,6 +166,7 @@ async def _pause(
                 'An `asyncio` task should not be calling this!?'
             ) from rte
         else:
+            import asyncio
             task = asyncio.current_task()
 
     if debug_func is not None:
@@ -946,6 +949,7 @@ def pause_from_sync(
 
         asyncio_task: asyncio.Task|None = None
         if is_infected_aio:
+            import asyncio
             asyncio_task = asyncio.current_task()
 
         # TODO: we could also check for a non-`.to_thread` context
@@ -1059,6 +1063,9 @@ def pause_from_sync(
             greenback: ModuleType = maybe_import_greenback()
 
             if greenback.has_portal():
+                from tractor.to_asyncio import (
+                    run_trio_task_in_future,
+                )
                 DebugStatus.shield_sigint()
                 fute: asyncio.Future = run_trio_task_in_future(
                     partial(

tractor/devx/debug/_tty_lock.py

@@ -20,7 +20,6 @@ Root-actor TTY mutex-locking machinery.
 
 '''
 from __future__ import annotations
-import asyncio
 from contextlib import (
     AbstractContextManager,
     asynccontextmanager as acm,
@@ -52,7 +51,6 @@ from trio import (
     TaskStatus,
 )
 import tractor
-from tractor.to_asyncio import run_trio_task_in_future
 from tractor.log import get_logger
 from tractor._context import Context
 from tractor.runtime import _state
@@ -66,6 +64,10 @@ from tractor.runtime._state import (
 )
 
 if TYPE_CHECKING:
+    # NOTE, `asyncio` (and `.to_asyncio`) are
+    # lazy-imported at their use-sites to keep them off
+    # the eager `import tractor` path (gh #470).
+    import asyncio
     from trio.lowlevel import Task
     from threading import Thread
     from tractor.ipc import (
@@ -910,6 +912,9 @@ class DebugStatus:
                     async def _set_repl_release():
                         repl_release.set()
 
+                    from tractor.to_asyncio import (
+                        run_trio_task_in_future,
+                    )
                     fute: asyncio.Future = run_trio_task_in_future(
                         _set_repl_release
                     )

tractor/discovery/_addr.py

@@ -22,7 +22,6 @@ from typing import (
     TYPE_CHECKING,
 )
 
-from bidict import bidict
 from trio import (
     SocketListener,
 )
@@ -35,6 +34,9 @@ from ..ipc._tcp import TCPAddress
 from ..ipc._uds import UDSAddress
 
 if TYPE_CHECKING:
+    # ONLY type-annots, the eager import costs ~4.5ms
+    # of `import tractor` wall-time (gh #470).
+    from bidict import bidict
     from ..runtime._runtime import Actor
 
 log = get_logger()

tractor/discovery/_multiaddr.py

@@ -24,13 +24,16 @@ Multiaddress support using the upstream `py-multiaddr` lib
 - https://github.com/multiformats/multiaddr/blob/master/protocols/unix.md
 
 '''
+from __future__ import annotations
 import ipaddress
 from pathlib import Path
 from typing import TYPE_CHECKING
 
-from multiaddr import Multiaddr
-
 if TYPE_CHECKING:
+    # NOTE, `multiaddr` is lazy-imported at first use
+    # (in the fns below) to keep it off the eager
+    # `import tractor` path (gh #470).
+    from multiaddr import Multiaddr
     from tractor.discovery._addr import Address
 
 # map from tractor-internal `proto_key` identifiers
@@ -56,6 +59,8 @@ def mk_maddr(
     multiaddr-spec-compliant protocol path.
 
     '''
+    from multiaddr import Multiaddr
+
     proto_key: str = addr.proto_key
     maddr_proto: str|None = _tpt_proto_to_maddr.get(proto_key)
     if maddr_proto is None:
@@ -98,6 +103,7 @@ def parse_maddr(
 
     '''
     # lazy imports to avoid circular deps
+    from multiaddr import Multiaddr
     from tractor.ipc._tcp import TCPAddress
     from tractor.ipc._uds import UDSAddress
 

tractor/ipc/_chan.py

@@ -198,9 +198,6 @@ class Channel:
         # assert transport.raddr == addr
         chan = Channel(transport=transport)
 
-        # ?TODO, compact this into adapter level-methods?
-        # -[ ] would avoid extra repr-calcs if level not active?
-        #   |_ how would the `calc_if_level` look though? func?
         if log.at_least_level('runtime'):
             from tractor.devx import (
                 pformat as _pformat,
@@ -325,10 +322,12 @@ class Channel:
         '''
         __tracebackhide__: bool = hide_tb
         try:
-            log.transport(
+            if log.at_least_level('transport'):
-                '=> send IPC msg:\n\n'
+                # don't materialize the payload repr if not necessary
-                f'{pformat(payload)}\n'
+                log.transport(
-            )
+                    '=> send IPC msg:\n\n'
+                    f'{pformat(payload)}\n'
+                )
             # assert self._transport  # but why typing?
             await self._transport.send(
                 payload,

tractor/ipc/_tcp.py

@@ -21,6 +21,7 @@ from __future__ import annotations
 import ipaddress
 from typing import (
     ClassVar,
+    TYPE_CHECKING,
 )
 # from contextlib import (
 #     asynccontextmanager as acm,
@@ -33,7 +34,6 @@ from trio import (
     open_tcp_listeners,
 )
 
-from multiaddr import Multiaddr
 from tractor.msg import MsgCodec
 from tractor.log import get_logger
 from tractor.discovery._multiaddr import mk_maddr
@@ -42,6 +42,11 @@ from tractor.ipc._transport import (
     MsgpackTransport,
 )
 
+if TYPE_CHECKING:
+    # ONLY type-annots, the eager import costs
+    # `import tractor` wall-time (gh #470).
+    from multiaddr import Multiaddr
+
 
 log = get_logger()
 

tractor/ipc/_transport.py

@@ -309,7 +309,10 @@ class MsgpackTransport(MsgTransport):
             log.transport(f'received header {size}')  # type: ignore
             msg_bytes: bytes = await self.recv_stream.receive_exactly(size)
 
-            log.transport(f"received {msg_bytes}")  # type: ignore
+            if log.at_least_level('transport'):
+                log.transport(  # type: ignore
+                    f'received {msg_bytes}'
+                )
             try:
                 # NOTE: lookup the `trio.Task.context`'s var for
                 # the current `MsgCodec`.

tractor/ipc/_uds.py

@@ -49,7 +49,6 @@ from trio._highlevel_open_unix_stream import (
     has_unix,
 )
 
-from multiaddr import Multiaddr
 from tractor.msg import MsgCodec
 from tractor.log import get_logger
 from tractor.discovery._multiaddr import mk_maddr
@@ -63,6 +62,9 @@ from tractor.runtime._state import (
 )
 
 if TYPE_CHECKING:
+    # ONLY type-annots, the eager import costs
+    # `import tractor` wall-time (gh #470).
+    from multiaddr import Multiaddr
     from tractor.runtime._runtime import Actor
 
 

tractor/log.py

@@ -26,11 +26,6 @@ built on `tractor`.
 '''
 from collections.abc import Mapping
 from functools import partial
-from inspect import (
-    FrameInfo,
-    getmodule,
-    stack,
-)
 import sys
 import logging
 from logging import (
@@ -38,10 +33,16 @@ from logging import (
     Logger,
     StreamHandler,
 )
-from types import ModuleType
+from types import (
+    FrameType,
+    ModuleType,
+)
 import warnings
 
-import colorlog  # type: ignore
+# NOTE, `colorlog` is lazy-imported in
+# `get_console_log()` to keep it off the eager
+# `import tractor` path (gh #470).
+#
 # ?TODO, some other (modern) alt libs?
 # import coloredlogs
 # import colored_traceback.auto  # ?TODO, need better config?
@@ -111,9 +112,7 @@ def at_least_level(
     if isinstance(level, str):
         level: int = CUSTOM_LEVELS[level.upper()]
 
-    if log.getEffectiveLevel() <= level:
+    return log.isEnabledFor(level)
-        return True
-    return False
 
 
 # TODO, compare with using a "filter" instead?
@@ -438,16 +437,33 @@ def get_logger(
         pkg_name: str =  _root_name
 
     def get_caller_mod(
-        frames_up:int = 2
+        frames_up: int = 2,
-    ):
+    ) -> ModuleType|None:
         '''
-        Attempt to get the module which called `tractor.get_logger()`.
+        Attempt to get the module which called
+        `tractor.get_logger()`.
+
+        Resolve the caller's frame with `sys._getframe()` and
+        map its `__name__` through `sys.modules`; `inspect.stack()`
+        (the previous impl) builds src-file info for EVERY frame
+        on the stack, scanning all of `sys.modules` per frame via
+        `inspect.getmodule()`, which made module-level
+        `get_logger()` calls dominate `import tractor` time
+        (see gh #470).
 
         '''
-        callstack: list[FrameInfo] = stack()
+        try:
-        caller_fi: FrameInfo = callstack[frames_up]
+            caller_frame: FrameType = sys._getframe(frames_up)
-        caller_mod: ModuleType = getmodule(caller_fi.frame)
+        except ValueError:
-        return caller_mod
+            return None
+
+        mod_name: str|None = caller_frame.f_globals.get(
+            '__name__',
+        )
+        if mod_name is None:
+            return None
+
+        return sys.modules.get(mod_name)
 
     # --- Auto--naming-CASE ---
     # -------------------------
@@ -782,6 +798,10 @@ def get_console_log(
             None,
         )
     ):
+        # lazy-imported to keep it off the eager
+        # `import tractor` path (gh #470).
+        import colorlog  # type: ignore
+
         fmt: str = LOG_FORMAT  # always apply our format?
         handler = StreamHandler()
         formatter = colorlog.ColoredFormatter(

tractor/msg/_ops.py

@@ -306,15 +306,15 @@ class PldRx(Struct):
             ):
                 try:
                     pld: PayloadT = self._pld_dec.decode(pld)
-                    log.runtime(
+                    if log.at_least_level('runtime'):
-                        f'Decoded payload for\n'
+                        # don't materialize the payload repr if not necessary
-                        # f'\n'
+                        log.runtime(
-                        f'{msg}\n'
+                            f'Decoded payload for\n'
-                        # ^TODO?, ideally just render with `,
+                            f'\n'
-                        # pld={decode}` in the `msg.pformat()`??
+                            f'{msg}\n'
-                        f'where, '
+                            f'where, '
-                        f'{type(msg).__name__}.pld={pld!r}\n'
+                            f'{type(msg).__name__}.pld={pld!r}\n'
-                    )
+                        )
                     return pld
                 except TypeError as typerr:
                     __tracebackhide__: bool = False

tractor/runtime/_rpc.py

@@ -1003,20 +1003,18 @@ async def process_messages(
             task_status.started(loop_cs)
 
             async for msg in chan:
-                log.transport(   # type: ignore
+                if log.at_least_level('transport'):
-                    f'IPC msg from peer\n'
+                    log.transport(   # type: ignore
-                    f'<= {chan.aid.reprol()}\n\n'
+                        f'IPC msg from peer\n'
+                        f'<= {chan.aid.reprol()}\n\n'
 
-                    # TODO: use of the pprinting of structs is
+                        # TODO: pretty-printing structs is FRAGILE;
-                    # FRAGILE and should prolly not be
+                        # -[ ] add a non-raising log formatter with
-                    #
+                        #      native-repr fallback before using
-                    # avoid fmting depending on loglevel for perf?
+                        #      `.msg.pretty_struct` here.
-                    # -[ ] specifically `pretty_struct.pformat()` sub-call..?
+                        # f'{pretty_struct.pformat(msg)}\n'
-                    #   - how to only log-level-aware actually call this?
+                        f'{msg}\n'
-                    # -[ ] use `.msg.pretty_struct` here now instead!
+                    )
-                    # f'{pretty_struct.pformat(msg)}\n'
-                    f'{msg}\n'
-                )
 
                 match msg:
                     # msg for an ongoing IPC ctx session, deliver msg to
@@ -1262,11 +1260,12 @@ async def process_messages(
                         log.exception(message)
                         raise RuntimeError(message)
 
-                log.transport(
+                if log.at_least_level('transport'):
-                    'Waiting on next IPC msg from\n'
+                    log.transport(
-                    f'peer: {chan.aid.reprol()}\n'
+                        'Waiting on next IPC msg from\n'
-                    f'|_{chan}\n'
+                        f'peer: {chan.aid.reprol()}\n'
-                )
+                        f'|_{chan}\n'
+                    )
 
             # END-OF `async for`:
             # IPC disconnected via `trio.EndOfChannel`, likely

tractor/runtime/_state.py

@@ -30,7 +30,6 @@ from typing import (
     TYPE_CHECKING,
 )
 
-import platformdirs
 from trio.lowlevel import current_task
 
 from msgspec import (
@@ -332,6 +331,10 @@ def get_rt_dir(
     the lovely `platformdirs` lib.
 
     '''
+    # lazy-imported to keep it off the eager
+    # `import tractor` path (gh #470).
+    import platformdirs
+
     rt_dir: Path = Path(
         platformdirs.user_runtime_dir(
             appname=appname,

tractor/spawn/_entry.py

@@ -38,7 +38,11 @@ from ..devx import (
     pformat,
 )
 # from ..msg import pretty_struct
-from ..to_asyncio import run_as_asyncio_guest
+#
+# NOTE, `.to_asyncio` (and thus `asyncio` itself) is
+# lazy-imported at the `infect_asyncio=True` use-sites
+# below to keep it off the eager `import tractor` path
+# (gh #470).
 from ..discovery._addr import UnwrappedAddress
 from ..runtime._runtime import (
     async_main,
@@ -96,6 +100,7 @@ def _mp_main(
     )
     try:
         if infect_asyncio:
+            from ..to_asyncio import run_as_asyncio_guest
             actor._infected_aio = True
             run_as_asyncio_guest(trio_main)
         else:
@@ -156,6 +161,7 @@ def _trio_main(
     )
     try:
         if infect_asyncio:
+            from ..to_asyncio import run_as_asyncio_guest
             actor._infected_aio = True
             run_as_asyncio_guest(trio_main)
         else:

tractor/trionics/_mngrs.py

@@ -198,6 +198,16 @@ async def gather_contexts(
 # Further potential examples of interest:
 # https://gist.github.com/njsmith/cf6fc0a97f53865f2c671659c88c1798#file-cache-py-L8
 
+class _CtxExit:
+    '''
+    Completion state for a cached context's shared exit.
+
+    '''
+    def __init__(self) -> None:
+        self.done = trio.Event()
+        self.error: Exception|None = None
+
+
 class _Cache:
     '''
     Globally (actor-processs scoped) cached, task access to
@@ -213,7 +223,11 @@ class _Cache:
     values: dict[Any,  Any] = {}
     resources: dict[
         Hashable,
-        tuple[trio.Nursery, trio.Event]
+        tuple[
+            trio.Nursery,
+            trio.Event,
+            _CtxExit,
+        ],
     ] = {}
     # nurseries: dict[int, trio.Nursery] = {}
     no_more_users: trio.Event|None = None
@@ -223,18 +237,38 @@ class _Cache:
         cls,
         mng,
         ctx_key: tuple,
+        ctx_exit: _CtxExit,
         task_status: trio.TaskStatus[T] = trio.TASK_STATUS_IGNORED,
 
     ) -> None:
-        async with mng as value:
+        entered: bool = False
-            _, no_more_users = cls.resources[ctx_key]
+        try:
-            cls.values[ctx_key] = value
+            async with mng as value:
-            task_status.started(value)
+                entered = True
-            try:
+                (
-                await no_more_users.wait()
+                    _,
-            finally:
+                    no_more_users,
-                value = cls.values.pop(ctx_key)
+                    _,
-                cls.resources.pop(ctx_key)
+                ) = cls.resources[ctx_key]
+                cls.values[ctx_key] = value
+                task_status.started(value)
+                try:
+                    await no_more_users.wait()
+                finally:
+                    cls.values.pop(ctx_key)
+                    cls.resources.pop(ctx_key)
+
+        except Exception as exc:
+            if not entered:
+                raise
+
+            # Deliver regular `__aexit__()` failures to the final
+            # consumer instead of raising into the service nursery.
+            ctx_exit.error = exc
+
+        finally:
+            if entered:
+                ctx_exit.done.set()
 
 
 class _UnresolvedCtx:
@@ -281,9 +315,10 @@ async def maybe_open_context(
         )
 
     # yielded output
-    # sentinel = object()
     yielded: Any = _UnresolvedCtx
     user_registered: bool = False
+    ctx_exit: _CtxExit|None = None
+    exit_error: Exception|None = None
 
     # Lock resource acquisition around task racing  / ``trio``'s
     # scheduler protocol.
@@ -300,7 +335,6 @@ async def maybe_open_context(
         ] = trio.StrictFIFOLock()
         header: str = 'Allocated NEW lock for @acm_func,\n'
     else:
-        await trio.lowlevel.checkpoint()
         header: str = 'Reusing OLD lock for @acm_func,\n'
 
     log.debug(
@@ -368,7 +402,6 @@ async def maybe_open_context(
             resources = _Cache.resources
             entry: tuple|None = resources.get(ctx_key)
             if entry:
-                service_tn, ev = entry
                 raise RuntimeError(
                     f'Caching resources ALREADY exist?!\n'
                     f'ctx_key={ctx_key!r}\n'
@@ -376,12 +409,18 @@ async def maybe_open_context(
                     f'task: {task}\n'
                 )
 
-            resources[ctx_key] = (service_tn, trio.Event())
+            ctx_exit = _CtxExit()
+            resources[ctx_key] = (
+                service_tn,
+                trio.Event(),
+                ctx_exit,
+            )
             try:
                 yielded: Any = await service_tn.start(
                     _Cache.run_ctx,
                     mngr,
                     ctx_key,
+                    ctx_exit,
                 )
             except BaseException:
                 # If `run_ctx` (wrapping the acm's `__aenter__`)
@@ -427,6 +466,11 @@ async def maybe_open_context(
             raise taskc
     else:
         # XXX, cached-entry-path
+        (
+            _,
+            _,
+            ctx_exit,
+        ) = _Cache.resources[ctx_key]
         _Cache.users[ctx_key] += 1
         user_registered = True
         log.debug(
@@ -445,44 +489,57 @@ async def maybe_open_context(
         )
 
     finally:
-        if lock.locked():
-            stats: trio.LockStatistics = lock.statistics()
-            owner: trio.Task|None = stats.owner
-            log.error(
-                f'Lock never released by last owner={owner!r} !?\n'
-                f'{stats}\n'
-                f'\n'
-                f'task={task!r}\n'
-                f'ctx_key={ctx_key!r}\n'
-                f'acm_func={acm_func}\n'
-
-            )
-
         if user_registered:
-            _Cache.users[ctx_key] -= 1
+            # Serialize user registration and teardown under the same
+            # per-key lock so no entrant can acquire a resource after
+            # its final user has committed to exiting it.
+            with trio.CancelScope(shield=True):
+                await lock.acquire()
+                try:
+                    _Cache.users[ctx_key] -= 1
 
-        if yielded is not _UnresolvedCtx:
+                    # If no consumers remain, keep entrants queued
-            # if no more consumers, teardown the client
+                    # until the cached context has completely exited.
-            if _Cache.users[ctx_key] <= 0:
+                    if _Cache.users[ctx_key] <= 0:
-                log.debug(
+                        log.debug(
-                    f'De-allocating @acm-func entry\n'
+                            f'De-allocating @acm-func entry\n'
-                    f'ctx_key={ctx_key!r}\n'
+                            f'ctx_key={ctx_key!r}\n'
-                    f'acm_func={acm_func!r}\n'
+                            f'acm_func={acm_func!r}\n'
-                )
+                        )
 
-                # XXX: if we're cancelled we the entry may have never
+                        # XXX: if we're cancelled, the entry may
-                # been entered since the nursery task was killed.
+                        # have never been entered since the nursery
-                # _, no_more_users = _Cache.resources[ctx_key]
+                        # task was killed.
-                entry = _Cache.resources.get(ctx_key)
+                        entry = _Cache.resources.get(ctx_key)
-                if entry:
+                        if entry:
-                    _, no_more_users = entry
+                            (
-                    no_more_users.set()
+                                _,
+                                no_more_users,
+                                ctx_exit,
+                            ) = entry
+                            no_more_users.set()
 
-                maybe_lock = _Cache.locks.pop(
+                        assert ctx_exit is not None
-                    ctx_key,
+                        await ctx_exit.done.wait()
-                    None,
+                        exit_error = ctx_exit.error
-                )
+
-                if maybe_lock is None:
+                        # A queued entrant already holds a reference
-                    log.error(
+                        # to this lock. Keep it registered until that
-                        f'Resource lock for {ctx_key} ALREADY POPPED?'
+                        # task has acquired and released it.
-                    )
+                        stats = lock.statistics()
+                        if not stats.tasks_waiting:
+                            maybe_lock = _Cache.locks.get(ctx_key)
+                            if maybe_lock is lock:
+                                _Cache.locks.pop(ctx_key)
+                            else:
+                                log.error(
+                                    f'Resource lock for {ctx_key} '
+                                    f'was replaced before teardown?'
+                                )
+                finally:
+                    lock.release()
+
+        if exit_error is not None:
+            # Always re-raise a regular `__aexit__()` error at the
+            # final consumer's context boundary.
+            raise exit_error

tractor/trionics/_taskc.py

@@ -349,7 +349,10 @@ async def start_or_cancel(
             # demote it to a `Cancelled`, losing the real error. The
             # `isinstance` guard also avoids a `TypeError` when
             # `rte.args[0]` isn't a `str`.
-            'child exited without calling' in rte.args[0]
+            rte.args[0] == (
+                'child exited without calling '
+                'task_status.started()'
+            )
         ):
             # re-raises the in-flight `trio.Cancelled` IFF we're
             # under effective cancellation; else a cheap no-op and

uv.lock

@@ -900,11 +900,11 @@ wheels = [
 
 [[package]]
 name = "setuptools"
-version = "82.0.1"
+version = "83.0.0"
 source = { registry = "https://pypi.org/simple" }
-sdist = { url = "https://files.pythonhosted.org/packages/4f/db/cfac1baf10650ab4d1c111714410d2fbb77ac5a616db26775db562c8fab2/setuptools-82.0.1.tar.gz", hash = "sha256:7d872682c5d01cfde07da7bccc7b65469d3dca203318515ada1de5eda35efbf9", size = 1152316, upload-time = "2026-03-09T12:47:17.221Z" }
+sdist = { url = "https://files.pythonhosted.org/packages/34/26/f5d29e25ffdb535afef2d35cdb55b325298f96debd670da4c325e08d70f4/setuptools-83.0.0.tar.gz", hash = "sha256:025bccbbf0fa05b6192bc64ae1e7b16e001fd6d6d4d5de03c97b1c1ade523bef", size = 1154254, upload-time = "2026-07-04T15:31:22.699Z" }
 wheels = [
-    { url = "https://files.pythonhosted.org/packages/9d/76/f789f7a86709c6b087c5a2f52f911838cad707cc613162401badc665acfe/setuptools-82.0.1-py3-none-any.whl", hash = "sha256:a59e362652f08dcd477c78bb6e7bd9d80a7995bc73ce773050228a348ce2e5bb", size = 1006223, upload-time = "2026-03-09T12:47:15.026Z" },
+    { url = "https://files.pythonhosted.org/packages/5d/40/e1e72872c6354b306daef1703549e8e83b4d43cfea356311bf722a043752/setuptools-83.0.0-py3-none-any.whl", hash = "sha256:29b23c360f22f414dc7336bb39178cc7bcbf6021ed2733cde173f09dba19abb3", size = 1008090, upload-time = "2026-07-04T15:31:20.885Z" },
 ]
 
 [[package]]