Repository navigation
function-length-check pre-commit hook blocks any commit to lifespan.py: cleanup_services (267 lines) and _init_graph_rag_service (68 lines) exceed the 65-line ceiling #16250
Description
The local function-length-check pre-commit hook (autobot-infrastructure/shared/scripts/hooks/pre-commit-function-length) scans every staged .py file in full (no diff-scoping, no baseline/grandfather list, unlike scripts/python_file_size_known_large.py's file-size ratchet). Two pre-existing functions in autobot-backend/initialization/lifespan.py already exceed its 65-line ERROR threshold, and now block any local commit that stages this file at all, regardless of what the commit actually changes.
Discovered while implementing #16231 (Dev_new_gui, worktree issue-16231-pricing-install): a one-line addition to initialize_background_services (net zero line change to the file) could not be committed locally because git commit re-scans the whole file and reports:
ERROR autobot-backend/initialization/lifespan.py:1167
Function '_init_graph_rag_service' has 68 body lines
Threshold: >65 lines
ERROR autobot-backend/initialization/lifespan.py:2251
Function 'cleanup_services' has 267 body lines
Threshold: >65 lines
COMMIT BLOCKED: 2 function(s) exceed 65 lines
Neither function was touched by the #16231 change (confirmed via git diff hunks).
Notes:
cleanup_servicesgrew to its current shape in bug(lifespan): shutdown handler joins its own thread, and its except swallows every later cleanup error #13585 (shutdown-teardown rewrite that guards each step independently — see its docstring)._init_graph_rag_servicepredates that.- CI's
enforce-precommit.ymlruns pre-commit diff-scoped (--from-ref HEAD^1 --to-ref HEAD^2) and treats hook findings as informational (::warning::), not blocking, for every hook except the one behavioural-gating hook (ssot-config-lib-guard). So this specific gap is local-only: it does not currently gate merges, but it does block a correctly-scoped localgit commiton these two files for anyone who has pre-commit installed, indefinitely, until someone refactors code unrelated to their own change.
Fix options:
- Refactor
cleanup_services(extract per-resource teardown helpers) and_init_graph_rag_service(extract the LlamaIndex construction block) under 65 body lines each, following the hook's own "Extract Method" guidance. - Or give the local hook the same diff-scoping / baseline mechanism
python_file_size_known_large.pyhas for file size, so a legacy function does not block an unrelated change to the same file.
Either closes the gap between "local pre-commit blocks" and "CI does not," which is itself worth resolving on its own (a required local gate that is silently advisory in CI is confusing and invites --no-verify).
Activity
mrveiss commentedon Sep 11, 2026
This is an instance of #16191, not a separate defect. It's the same hook,
function-length-check, and the same whole-file read. #16191's first acceptance criterion covers this case word for word: "A one-line change to a file containing a >65-line function it does not touch lands." This issue has no acceptance criteria of its own. Tracking it separately risks the one hook being fixed twice, in two different ways.Recommendation: fold this issue's facts into #16191 and close it as a duplicate. The facts are two new offenders,
lifespan.py'scleanup_services(267 body lines) and_init_graph_rag_service(68), and the CI finding below. Or keep it only as a sub-issue of #16191. Both are the filer's or coordinator's call, not mine. It's a sub-issue of umbrella #16238 as asked.One fact here is worth keeping, and I've checked it.
enforce-precommit.ymlruns every pre-commit hook with--from-ref HEAD^1 --to-ref HEAD^2(:202) and turns findings into::warning::(:207-209). Only the gating hook,ssot-config-lib-guard, blocks (:250). So the block really is local-only. Separately, any fix that scopes a hook to the staged diff has to handle this job: under--from-ref, pre-commit stages nothing, so a staged-mode scope sees an empty change. pre-commit exports the range asPRE_COMMIT_FROM_REF(pre-commit 4.6.0,commands/run.py:391). #16241 hit exactly this forno-local-schemasand uses that variable. #16191's fix should do the same.fix(guards): stay honest when pre-commit stages nothing, under --from…
fix(hooks): judge only the functions a change touches in function-len…
fix(guards): scope no-local-schemas to the lines a change added, and …
mrveiss commentedon Sep 12, 2026
Stale-issue sweep (never auto-closed — Dev_new_gui wasn't GitHub's default branch at merge time). Leaving open — UNVERIFIABLE, not a guessed disposition.
This issue has no acceptance-criteria checklist, only prose describing two possible fix directions. The delivering PR (#16241) says so itself: "#16250 (
lifespan.py) is a third instance ... with no criteria of its own." Since there's nothing to tick against, I'm flagging this rather than closing on my own read.That said, strong evidence exists that the underlying scenario is resolved:
autobot-infrastructure/shared/scripts/hooks/function_length_checker.pyis now diff-scoped (added_lines/resolve_base/staged_pathsfromtools/lint/_scan_helpers.py) rather than whole-file — the literal complaint (a one-line unrelated edit tolifespan.pyblocked by unrelated long functions) should no longer reproduce. Worth a quick manual check before closing, since the issue itself gives nothing to verify against formally.mrveiss commentedon Sep 12, 2026
Proposed acceptance criteria (derived from the report; the owner may amend). Related to #16186, where whole-file hooks block changes to lines nobody touched.
function-length-checkhook doesn't block a commit for functions the commit didn't change. It's diff-scoped, or it has a shrink-only baseline of pre-existing violations.cleanup_servicesand_init_graph_rag_serviceininitialization/lifespan.pyare brought under the limit, or recorded in that baseline with a removal issue.fix(lifespan): write the startup error file off the event loop, and l…
refactor(lifespan): split cleanup_services and _init_graph_rag_servic…
test(repo): drop the vanished lifespan.py:1353 citation from the line…
refactor(lifespan): split cleanup_services and _init_graph_rag_servic…
mrveiss commentedon Sep 13, 2026
Closure evidence (batch 2c)
Verified against
origin/main(PR #16509, 20e7e6a). No checkboxes, so the Fix is taken as the criteria. All met. Body lines were counted with the checker's own rule (_count_body_linesinfunction_length_checker.py) applied to theorigin/maintext.cleanup_servicesunder the 65-line ERROR thresholdautobot-backend/initialization/lifespan.py); it calls_stop_request_and_scheduled_work(11),_drain_background_work(8),_release_shared_resources(9)_init_graph_rag_serviceunder 65autobot-backend/initialization/neural_mesh_wiring.py(longest function 19)lifespan.py,lifespan_shutdown.pyandneural_mesh_wiring.py; the longest isinitialize_background_servicesat 63 (WARNING band)autobot-infrastructure/shared/scripts/hooks/function_length_checker.py_scope_to_change; testspre-commit-function-length_test.py::test_a_one_line_change_beside_a_legacy_long_function_lands,test_editing_the_long_function_itself_still_fails,test_adding_a_new_long_function_still_fails. Shutdown order pinned bylifespan_shutdown_order_test.py::test_cleanup_runs_every_step_once_in_the_pinned_order