Repository navigation
fix(*): end leftover ACP agents, trim the heap, and bound the session cache - #942
Conversation
… cache A long-running host kept memory it no longer used: ACP agents of a host that exited without closing them, native memory glibc keeps in its arenas, and every session SessionManager had ever loaded. On Linux an ACP agent now starts through an exec wrapper that arms PR_SET_PDEATHSIG, and every host records the process groups it launched under acp-launches/ beside the config file. The first ACP launch in any process ends the groups of hosts that have since died: only in the same boot and PID namespace, only for a record of this version, and only where a member still carries the dead host's RAVEN_ACP_HOST mark. The gateway, raven serve, the TUI's RPC host, raven acp and raven a2a serve run malloc_trim(0) in a worker thread, at most every ten minutes, once anonymous memory stands 32 MiB above its lowest since the last trim. SessionManager's cache becomes an LRU with a soft limit of 64. Only a session with nothing unsaved leaves it, one a caller still holds comes back as the same object, and a reload keeps the saved updated_at.
gloryfromca
left a comment
gloryfromca
left a comment
There was a problem hiding this comment.
No blockers; this can merge as far as I am concerned.
I reviewed the full github/main...HEAD diff and the relevant surrounding code and history. I covered the repository rules in AGENTS.md/CLAUDE.md and CONTEXT-MAP.md, the ACP launch/close callers and Linux process-identity safeguards, the heap-trimmer host integrations, session cache/dirty-state behavior and metadata-patch interactions, backward compatibility, architecture import boundaries, and the tests for signs of weakened coverage. I found no concrete failure worth raising.
Verification:
uv sync --all-packages(the first test attempt could not enter tests because the workspace pluginraven_everoswas not installed)uv run pytest tests/test_acp_leftovers.py tests/test_utils_heap.py tests/test_session_manager.py tests/test_cli_a2a_commands.py tests/test_cli_acp_commands.py tests/test_cli_gateway_commands.py tests/test_cli_serve_commands.py tests/test_cli_tui_commands.py tests/test_subagent_host_env.py -q- 620 passeduv run ruff check ...over the changed Python files - passeduv run lint-imports- 10 contracts kept- source-language and large-file checks over
github/main...HEAD- passed
|
Windows compatibility review: head No Windows compatibility regression found in the inspected changes. Reviewed all 19 changed files, ACP launch/close callers, all five host integrations, session persistence/locking and timestamp consumers, tests, and the existing review. Native Windows / Python 3.13.12 verification:
Broader file-level runs were not green. Each failure below was reproduced on the exact base as well as this head:
The initial broader run also stalled in Verification commandsRun from isolated detached head/base checkouts. Setup succeeded with $run = 'C:/Users/livxu/AppData/Local/Temp/raven-windows-review/f3a5f15497914585b2ca031de778aff2'
$cases = @(
'tests/test_acp_leftovers.py',
'tests/test_utils_heap.py',
'tests/test_session_manager.py',
'tests/test_cli_a2a_commands.py::test_serve_starts_the_heap_trimmer',
'tests/test_cli_acp_commands.py::TestServe::test_the_heap_trimmer_starts_before_the_channel_is_served',
'tests/test_cli_gateway_commands.py::test_run_starts_the_heap_trimmer_at_boot',
'tests/test_cli_serve_commands.py::test_serve_starts_the_heap_trimmer_at_boot',
'tests/test_cli_tui_commands.py::test_the_terminal_host_starts_the_heap_trimmer',
'tests/test_subagent_host_env.py::test_acp_launch_preserves_stored_windows_path_quotes',
'tests/test_subagent_host_env.py::test_on_windows_a_bare_program_starts_from_the_childs_own_path',
'tests/test_subagent_host_env.py::test_on_windows_an_acp_server_that_is_a_batch_file_starts_from_the_childs_path'
)
$suite = '942-changed'
uv run pytest @cases -q -n 0 -x --tb=short --basetemp "$run/checks/$suite/pytest" --junitxml "$run/checks/$suite/junit.xml"
# 170 passed, 72 skipped
$suite = '942-native'
uv run pytest "$run/probes/test_windows_pr942.py" -q -n 0 -x --tb=short --basetemp "$run/checks/$suite/pytest" --junitxml "$run/checks/$suite/junit.xml"
# 2 passedEach baseline control used Coverage limits: no full-suite pass, Linux parent-death/real heap-trim execution, real ACP product launch, or live UNC-share test. The Linux-only scope and remaining Windows work in #928 remain relevant. |
|
Addressed the inherited Windows test-fixture limitations from the native review:
Native Windows 10.0.26300 / Python 3.13.12, The original discovery test and both path assertions failed before the corrections. The final regression command (JUnit output path omitted) was: uv run --frozen --no-sync pytest tests/test_acp_leftovers.py tests/test_utils_heap.py tests/test_session_manager.py tests/test_session_project_grouping.py tests/test_subagent_host_env.py tests/test_cli_a2a_commands.py::test_serve_starts_the_heap_trimmer tests/test_cli_acp_commands.py::TestServe::test_the_heap_trimmer_starts_before_the_channel_is_served tests/test_cli_gateway_commands.py::test_run_starts_the_heap_trimmer_at_boot tests/test_cli_serve_commands.py::test_serve_starts_the_heap_trimmer_at_boot tests/test_cli_tui_commands.py::test_the_terminal_host_starts_the_heap_trimmer -k 'not test_acp_child_inherits_custom_raven_home and not test_acp_child_is_told_it_serves_as_a_subagent' -n 0 --idle-ceiling-strict --tb=short229 passed, 73 skipped, 2 deselected; exit 0. Skips: 72 existing Linux/glibc cases and one measured Windows symlink-privilege limitation. The two real ACP child cases remain dependent on #908; the broader CLI mode/signal cases are covered separately by #939. No full Windows suite, privileged symlink run, ACL audit, real ACP lifecycle, or new Linux execution is claimed. The Linux-only cleanup scope remains unchanged. Ruff, 10 import contracts, changed-production type checks, full-PR pre-commit, commitlint, commit/source-language/asset gates and the complete-diff review passed. The description now separates the original Linux measurements from this Windows follow-up. The exact tested commits were published through a guarded fast-forward ref update and read back from GitHub; the new CI run is pending. |
gloryfromca
left a comment
gloryfromca
left a comment
There was a problem hiding this comment.
No blockers; this can merge as far as I am concerned.
I reviewed the three test-only commits since the previously reviewed head. The native-path expectations match the production str(Path(...)) behavior, the .exe fixtures preserve the boot-PATH and refreshed-registry-PATH assertions on both simulated and native Windows, and the symlink test skips only for Windows privilege error 1314 while re-raising unrelated failures. I also checked that the changed tests still exercise their original behavior rather than weakening it.
Verification:
uv run pytest tests/test_session_project_grouping.py tests/test_subagent_host_env.py -q- 66 passeduv run ruff check tests/test_session_project_grouping.py tests/test_subagent_host_env.py- passed- source-language check over
github/main...HEAD- passed
|
Windows compatibility re-review at Rechecked the complete 20-file PR scope and its callers against the previously reviewed tree. Only the two test files changed since Native Windows / Python 3.13.12, an exact detached checkout, fresh isolated uv environment and separate home/temp/token directories; imports verified at this head. Setup $R = 'C:/Users/livxu/AppData/Local/Temp/raven-windows-review/c281ddf96c3446c98e4f4127722b59c6'
uv run pytest tests/test_acp_leftovers.py tests/test_utils_heap.py tests/test_session_manager.py tests/test_session_project_grouping.py tests/test_subagent_host_env.py tests/test_cli_a2a_commands.py::test_serve_starts_the_heap_trimmer tests/test_cli_acp_commands.py::TestServe::test_the_heap_trimmer_starts_before_the_channel_is_served tests/test_cli_gateway_commands.py::test_run_starts_the_heap_trimmer_at_boot tests/test_cli_serve_commands.py::test_serve_starts_the_heap_trimmer_at_boot tests/test_cli_tui_commands.py::test_the_terminal_host_starts_the_heap_trimmer -k 'not test_acp_child_inherits_custom_raven_home and not test_acp_child_is_told_it_serves_as_a_subagent' --idle-ceiling-strict -q -n 0 -x --tb=short --basetemp "$R/checks/942-regression/pytest" --junitxml "$R/checks/942-regression/junit.xml"
# 229 passed, 73 skipped, 2 deselected; exit 0The 73 skips are 72 Linux/glibc cases and one unavailable symlink privilege. The two real ACP-child tests remain excluded because the unchanged Windows close limitation was reproduced at both head and base in the previous review. Its broader CLI pipe/mode/signal limits also remain outside this passing selection. This verifies the fixture corrections and selected host behavior; it does not establish Windows parent-death cleanup, Linux heap-trim behavior, real ACP product lifecycle, privileged symlinks, ACL enforcement, or a full-suite pass. |
…iled (#944) ## Summary `SessionManager.save()` lost the turns that follow an undo or a clear whose own save failed, and brought the cut turn back. `save()` appends `messages[_persisted_count:]`, and rewrites the file only while the list is shorter than `_persisted_count`. `Session.undo_last_turn()` and `Session.clear()` cut the list in memory and leave the write to their callers. When that write fails, the file keeps the cut messages and `_persisted_count` keeps its old value. Once new messages take the list back to that count, the next save appends past it: the file keeps the cut messages, and the ones that replaced them are never written. Past the old count, the file ends up holding the cut messages followed by the tail of the new ones. Where such a failed save comes from: - `session.undo` (the TUI's `/undo` and `/retry`) and `session.clear` save right after the cut and only log a failure. - `/new` saves after `clear()` and raises on a failure before it invalidates the session, so the cleared copy stays cached. So a `/retry` after an undo whose save failed loses the turn it sent again on the next reload, and the undone turn comes back. The bounded session cache from #942 read that state as saved too, since its clean check compares the message count and a fingerprint of the other saved fields, so such a session could leave the cache and be reloaded from the stale file. `Session.truncations` already counts every cut: both methods bump it, and the episodic annotator reads it for the same reason, since a cut is invisible to indices once the list has grown back. `Session._persisted_truncations` holds that count as of the last time this copy read the file or wrote it successfully. Three readers treat a difference as unsaved: - `save()` rewrites the file atomically, however far the list has grown back. - The cache's clean check keeps the session cached. - `flush()`, which `session.close` calls, saves it. `session.close`'s docstring names the new case. Nothing on the wire or on disk changes shape. ## Type - [x] Fix - [ ] Feature - [ ] Docs - [ ] CI / tooling - [ ] Refactor - [ ] Other ## Verification - **New tests, against the unfixed code:** six of the seven new cases failed there, each reading back the file as the cut left it, for example `['q1', 'a1', 'q2', 'a2']` where the retried answer belonged. The seventh, `test_once_a_cut_is_on_disk_a_save_keeps_what_another_writer_appended`, passes there by design: it holds the append path for concurrent writers once a cut is on disk. All seven pass with the fix. - **Mutation:** each construct was removed or moved one at a time, with the backup outside the tree, and each mutant failed the tests written for it: `save()` ignoring the cut (4 failed), the counter never recorded (1, the concurrent-writer case), the clean check ignoring the cut (1), `flush()` ignoring it (1), and the counter recorded before the write instead of after it (6). - **Directly affected and neighbouring files:** `uv run --frozen --python 3.12 --all-extras pytest -q -p no:cacheprovider --idle-ceiling-strict` over `tests/test_session_manager.py`, `tests/test_rpc_session.py`, `tests/test_rpc_schema_match.py`, `tests/test_rpc_session_init_bundle.py`, `tests/test_rpc_session_naming.py`, `tests/test_session_{export,project_grouping,title}.py`, `tests/test_defer_manager.py` and `tests/test_agent_loop_{episodic_annotate,empty_recovery}.py`. Result at this head: 1003 passed. - **Full suite at this head:** `uv run --frozen --python 3.12 --all-extras pytest -q -p no:cacheprovider` (repo addopts: `-n 4`, integration and e2e deselected). Result: 8 failed, 31091 passed, 164 skipped, 2 xfailed. The failures are this box's environment, none introduced: five in `tests/test_config_update_providers.py` follow the shell's proxy variables (that file passes 193 of 193 with them unset), and `test_install_script.py::test_resolve_node_dir_answers_each_case_it_exists_for`, `test_knowledge_parser_excel.py::test_a_legacy_workbook_is_converted_and_read` (no LibreOffice) and `test_subagent_node_runtime.py::test_what_cannot_be_read_names_nothing` (run as root) fail the same way on an exported copy of the base commit. - **Diff coverage:** the seven executable changed lines in `raven/session/manager.py` are all executed by `tests/test_session_manager.py` and `tests/test_rpc_session.py`. - **Lint and gates:** `make lint-python lint-imports lint-deps lint-types check-large-files check-source-language check-commits`. Result: all seven exit 0; import contracts 10 kept, 0 broken. - [x] Relevant tests pass locally - [x] Relevant lint / type checks pass locally - [ ] User-facing docs or screenshots are updated when needed No user-facing docs change, because no command, flag, setting or file format does. ## Risk User-visible behaviour changes: - **A save after a cut whose own save failed now rewrites the file.** A rewrite writes this copy's messages only, so turns another process appended to the same session since this copy last read it are dropped, as they already are when a cut is saved on the first try. Before this change that save appended: it kept those turns, and left the cut messages in place of the new ones. - **`session.close` writes a pending cut.** `flush()` used to skip a session whose list was not longer than the saved count. - **A session with a pending cut stays in the bounded cache** until a save succeeds. The limit is soft, as it already is for any session with unsaved changes. Left as today: - `session.undo` and `session.clear` still return success when their own save fails, and only log it. The cut now reaches the disk with the next successful save, and is lost if the process ends first. - [x] Security impact considered - [x] Backward compatibility considered - [x] Rollback path is clear for risky changes Rollback is a revert. Session files are written in the same format as before, and the new counter is never written to disk. ## Related Issues #942 (its Risk section disclosed the cache half of this)
… dead host left (#953) ## Summary On macOS an ACP agent outlives a Raven host that dies without closing it (killed, crashed, OOM-killed). The agent runs in a session of its own, so it gets none of the host's signals, and only `AcpClient.close` ends it. #942 closed this on Linux with a parent-death wrapper and a launch-record sweep. On macOS both were no-ops. This PR brings both layers to macOS. **1. The agent ends with its host (`raven/acp_client/leftovers.py`, `_DARWIN_BINDING`).** macOS has no `PR_SET_PDEATHSIG`. Instead, the exec wrapper double-forks a small watcher before it execs the agent, then reaps the middle child. - The watcher is adopted by launchd, so the agent has no child it did not start. - It stays in the agent's process group, closes every descriptor it inherited (so it holds none of the agent's pipes), and changes to `/`. - It waits on the host's exit with kqueue (`EVFILT_PROC` / `NOTE_EXIT`), then ends the whole group. An `npx` launcher's grandchild goes with it, which the Linux signal does not reach. - If the host is already gone when the watch is set, registration fails with `ESRCH`, and that counts as an exit. - It ends the group only while the agent leads its own group, which every launch does. Any other failure leaves the agent as it would have been. - `close()` ends the watcher along with the rest of the group. - Cost: about 4 MiB RSS per launched agent. **2. The sweep at the next launch runs on macOS.** This is the backstop for when the watcher itself is gone. The decision logic is unchanged; each `/proc` read has a macOS equivalent: - process start time: `sysctl KERN_PROC_PID` (`kinfo_proc.p_starttime`, in microseconds). It reads any process, including launchd and zombies; `proc_pidinfo` refuses launchd with EPERM. - group members: `proc_listpids(PROC_PGRP_ONLY)` - launch mark: `sysctl KERN_PROCARGS2`. EINVAL (gone, zombie, or another user's process) counts as no mark. - boot identity: `kern.bootsessionuuid` and `kern.boottime`. macOS has no PID namespaces, so the namespace is a fixed value. macOS hides the environment of its own platform binaries (`/bin/sh`, `/bin/sleep`). Only a group member that is not one can show the mark, such as the agent's node or python process, or the watcher. What cannot be read still leaves a group alone. **Linux is unchanged.** The Linux wrapper string is byte-identical to main's `_WRAPPER`. `_start_ticks` is renamed to `_start_time`, which now dispatches by platform. `CONTEXT.md` (Leftover agent, Launch mark) is updated to match. **Windows lifetime handling is provided by #951.** This PR adds the macOS wrapper and sweep; these helpers remain a no-op on Windows. ## Type - [x] Fix - [ ] Feature - [ ] Docs - [ ] CI / tooling - [ ] Refactor - [ ] Other ## Verification Before rebasing onto the merged Windows lifetime change, on macOS 26.6.2, arm64, Python 3.12 (project venv): - `uv run pytest tests/test_acp_leftovers.py tests/test_subagent_host_env.py tests/test_acp_client_teardown.py tests/test_acp_client_launch.py -q -p no:cacheprovider -m "slow or not slow"` -> 117 passed, 9 skipped (Linux-only mechanisms). - Every other `tests/test_*acp*.py` and `tests/test_*subagent*.py` file, run one file at a time -> all pass except `tests/test_subagent_dag_runner.py`. On this Mac that file hangs at the same two points with this branch and with main's `leftovers.py`: - after `TestBackgroundRun::test_a_run_that_ignores_cancellation_fails_named_instead_of_hanging_loop_close` fails; - with that test deselected, after `TestBackgroundRun::test_the_failure_path_reaps_a_subagent_child_left_running` (21 passed in both runs). It is unrelated to this change; Linux CI covers that file. - Red/green with a stand-in host (`AcpClient.launch`, then `kill -9` of the host). The stand-in agent ignores SIGTERM, never reads stdin, and starts a child of its own. - main: 3 s later the agent is alive with ppid 1, and its child is alive. - this branch: 3 s later both are gone. - Sweep backstop: kill the watcher, then `kill -9` the host. The agent stays (ppid 1). The next host's launch ends the agent and its child. - `uv run ruff format --check` and `uv run ruff check` on the changed files -> clean. `make check-source-language` and `make check-large-files` -> clean. New macOS tests cover: - a killed host ending the agent and what it started; - the watcher holding none of the agent's pipes and being the only member left after the agent exits; - the watcher ending nothing when the agent does not lead its group; - the exec keeping the agent's pid and argv; - SIGPIPE and SIGXFSZ reaching the agent at their defaults; - the start-time, member, environment and boot readers (including an empty argv entry and a platform binary). Most existing Linux tests now run on macOS too. After rebasing onto the merged Windows lifetime change, on Linux / Python 3.12: - The only manual conflict resolution is in `CONTEXT.md`: retain both platform descriptions, scope the record sweep to Linux and macOS, and remove the stale sentence that macOS only has explicit teardown. - The macOS implementation and its tests are byte-identical to the approved revision. The Windows implementation and its tests are byte-identical to main; the shared client and environment-test logic also match main, with only the macOS comments/docstring added. - `uv run --no-sync pytest tests/test_acp_leftovers.py tests/test_acp_windows.py tests/test_acp_client_launch.py tests/test_acp_client_teardown.py tests/test_subagent_host_env.py tests/test_subagent_acp.py tests/test_living_docs.py -q -x` -> 468 passed, 10 macOS-only tests skipped. - `uv run --no-sync ty check raven/acp_client` and `uv run --no-sync lint-imports` -> passed; all 10 import contracts kept. - `uv run --no-sync pre-commit run --from-ref origin/main --to-ref HEAD` -> passed for every applicable hook. - `make check-large-files check-source-language check-commits` with the existing environment and `UV_NO_SYNC=1` -> passed. Native macOS and Windows tests were not repeated after this documentation-only conflict resolution. The current-head Windows re-review before the rebase reported 65 passed and 79 platform-specific skips. - [x] Relevant tests pass locally - [x] Relevant lint / type checks pass locally - [x] User-facing docs or screenshots are updated when needed ## Risk - User-visible: on macOS an ACP agent is now SIGKILLed with its whole process group when its host exits without closing it. On Linux the parent-death signal ends the direct child; the launch-record sweep handles surviving marked descendants. An agent that relied on outliving its host no longer does. - Each launched agent on macOS carries one extra idle process (about 4 MiB) in its group until the group is closed or the host exits. - The sweep only acts on groups recorded by a dead host in the same boot, whose leader start time matches and which still hold a member carrying that host's mark. A recycled pid or group number is left alone. - Rollback: revert this commit. Record files a macOS host left under `acp-launches/` are then inert, because the sweep is a no-op on macOS again. - [x] Security impact considered - [x] Backward compatibility considered - [x] Rollback path is clear for risky changes ## Related Issues #928 (macOS); #951 provides the Windows lifetime handling.
Summary
A long-running Raven host held on to three kinds of memory it no longer used. Observed on one dev box on 2026-10-12:
SessionManagerkept every session it had ever loaded.This PR cleans up all three automatically.
1. ACP agents that outlive their host (Linux). An ACP agent starts in a session of its own, so it gets none of the signals sent to its host's process group or terminal, and while the host lives only
AcpClient.closeends it. A host that exited without closing its agents, or was killed, crashed or OOM-killed before its teardown, left them running.The agent now starts through a one-line exec wrapper (
raven/acp_client/leftovers.py). It runs the host's own interpreter with-I -S, armsPR_SET_PDEATHSIG(SIGKILL), checks that its parent is still the host, restores SIGPIPE and SIGXFSZ to their defaults and execs the agent in place, with the same pid and group and the agent's own argv. The agent's process then ends with the host.The signal reaches only the direct child, and an
npxlauncher's real server is a grandchild. So every host also records the process groups it launched, one file per host underacp-launches/in the config file's directory. The first ACP launch in any process ends the groups recorded by hosts that have since died. That sweep runs once per process on a daemon thread. Every launch waits for it for at most 5 s, and once one launch has waited that long the rest stop waiting for it.A group is ended only when all of the following hold:
RAVEN_ACP_HOST=<host pid>-<host start ticks>, which every launched agent gets last in its environment.A read that fails, of a record or of
/proc, keeps the record and ends nothing; a record that does not parse is deleted.2. Native heap trimming (Linux with glibc). glibc keeps the memory native code frees, such as PyMuPDF rasters and tokenizer tables, in its arenas.
raven/utils/heap.pychecks every ten minutes and, once anonymous memory stands 32 MiB above its lowest reading since the last trim, runsmalloc_trim(0)in a worker thread. The five long-running hosts start it:raven gateway,raven serve, the TUI's RPC host,raven acpandraven a2a serve.3. A bounded session cache.
SessionManager._cachebecomes an LRU with a soft limit of 64 sessions.updated_atit was saved with.CONTEXT.mdgains two terms, Leftover agent and Launch mark.Measured on this box on 2026-10-13:
store_shrink(100), and 93.5 and 84.6 MiB afterheap.trim().agents/, which that config cannot start, and eleven presets, eight of which completed the handshake (Claude Code, Codex, OpenCode, Hermes Agent, OpenClaw, CodeBuddy, Qoder and Pi). Afterwards no marked process and no launch record remained.AcpClient. On main, a direct agent, annpx-shaped launcher and its grandchild all survive with ppid 1. On this branch, the direct agent and the launcher end with the host, and the next host's first launch ends the grandchild.agents/was verified twice in a home holding only dummy keys (a handshake calls no model): Raven-Design completed the handshake both ways, and the other four stopped on the same missing key both ways.The Windows regression fixtures use explicit executable filenames and native project-path strings. The unreadable-transcript symlink fixture skips only when Windows reports missing symlink privilege (WinError 1314); other creation errors still fail. These test corrections do not extend the Linux-only cleanup mechanisms.
Type
Verification
The original author's Linux measurements below precede the Windows test-fixture follow-up. The native Windows results that follow were run against the final follow-up tree.
uv run --frozen --python 3.12 --all-extras pytest -q -p no:cacheprovider(repo addopts:-n 4, integration and e2e deselected). Result: 3 failed, 30261 passed, 158 skipped, 2 xfailed. The failures are the ones this box gives on main too:test_install_script.py::test_resolve_node_dir_answers_each_case_it_exists_for,test_knowledge_parser_excel.py::test_a_legacy_workbook_is_converted_and_read(no LibreOffice) andtest_subagent_node_runtime.py::test_what_cannot_be_read_names_nothing(run as root).tests/test_acp_leftovers.py,tests/test_utils_heap.py,tests/test_session_manager.py,tests/test_cli_{a2a,acp,gateway,serve,tui}_commands.pyandtests/test_subagent_host_env.py, under--idle-ceiling-strictas CI runs them. Result: 620 passed.tests/test_rpc_session.py,tests/test_session_{export,title,project_grouping}.py,tests/test_defer_manager.py,tests/test_subagent_acp.pyandtests/test_acp_{ask_user,elicitation,remote}.py. Result: 683 passed.make lint-python lint-imports lint-deps lint-types check-large-files check-source-language check-commits. Result: all seven exit 0; import contracts 10 kept, 0 broken./proc/<pid>/statparse; the trimmer's reader and threshold; and the cache's eviction order, fingerprint,updated_atrestore and patch rule./procread or akillpgthat fails, and the import insideraven a2a serve, which only a running server reaches.Native Windows test-fixture follow-up (2026-10-14): Windows 10.0.26300, Python 3.13.12,
PYTHONUTF8=1, a dedicated worktree anduv sync --frozen --python 3.13 --all-packages --extra dev. The imported Raven package was verified to come from that worktree. Pytest's isolated homes and a separate callerRAVEN_HOMEwere used. JUnit output paths are omitted from the command below.Result: 229 passed, 73 skipped, 2 deselected, exit 0 with the idle ceiling enforced. Of the skips, 72 are existing Linux/glibc cases and one is the measured lack of Windows symlink privilege. The three corrected discovery/path cases failed before the corrections and pass in this selection. Before the symlink guard, the same broader selection had 229 passes, 72 skips, two deselections and one WinError 1314 failure.
The two deselected real ACP child tests still depend on the native close fix in #908. The broader CLI permission/signal limitations reported in the Windows review are covered separately by #939. This follow-up claims no full Windows suite pass, real ACP lifecycle validation, privileged symlink run, Windows ACL audit, or new Linux run.
uv run --frozen --no-sync ruff check raven evolver agents plugins-dist tests scripts docs-siteand the matchingruff format --check: passed (2,272 files formatted); the final symlink edit also passed both file-scoped checks.uv run --frozen --no-sync lint-imports: 10 contracts kept.uv run --frozen --no-sync ty check raven/acp_client/client.py raven/acp_client/leftovers.py raven/cli/a2a_commands.py raven/cli/acp_commands.py raven/cli/gateway_commands.py raven/cli/serve_commands.py raven/cli/tui_commands.py raven/session/manager.py raven/utils/heap.py: passed.uv run --frozen --no-sync pre-commit run --from-ref origin/main --to-ref HEAD: all applicable hooks passed.node node_modules/@commitlint/cli/cli.js --from origin/main --to HEAD --config commitlint.config.cjs,uv run --frozen --no-sync python scripts/check_commit_messages.py origin/main..HEAD,uv run --frozen --no-sync python scripts/check_large_files.py origin/main...HEAD,uv run --frozen --no-sync python scripts/check_source_language.py origin/main...HEAD, andgit diff --check origin/main...HEAD: passed.The relevant-tests checkbox remains unchecked because the broader Windows files still contain the disclosed limitations above.
No user-facing docs change, because no command, flag or setting does. The two new terms are glossary entries in
CONTEXT.md.Risk
User-visible behaviour changes:
raven: cannot start <path>: <error>to the agent's stderr and exits 127, so the failure surfaces as a connection that ends with that line in its stderr, whereAcpClient.launchused to raisecannot startat once. A missing program still fails at launch, as before.RAVEN_ACP_HOST. It is an identifier, not a secret; its parts are world-readable in/proc/<pid>/stat.LC_CTYPE=C.UTF-8, because the wrapper interpreter's PEP 538 coercion carries through the exec.acp-launches/beside the config file, synchronously and with an fsync.updated_atis the one it was saved with, where it used to be the time of the load, so a save that adds no message no longer moves it. Sentinel's defer timing and the export header read it in memory, and the session list orders by the saved value.Left as today:
--configsweeps that directory's records only.Edge cases of the records directory:
Edge cases of the session cache:
save()already mishandles that state, so the residue predates this PR.A group is signalled only when it passes all five checks above. A crafted record cannot point a sweep at a live host's agents, because the mark names the identity the liveness check reads.
Rollback is a revert. Nothing persisted changes shape: session files are written exactly as before, and
acp-launches/only holds what the next sweep reads, so it can be deleted at any time.Related Issues
#928 (macOS and Windows; stays open)