Repository navigation
chore(gates): GC store-site verdicts for seven raw writes; bind class_field_set to the _known emitter - #14750
Conversation
|
Warning Review limit reachedYou've used all free OSS reviews for now. Wait for the free limit to reset to keep reviewing this public repository. Next included review available in 11 minutes. View limit details
No actionable comments were generated in the recent review. Walkthrough
Merge Risk:
|
…_field_set to the _known emitter The class-field store calls emit_jsvalue_slot_store_pointer_tested_known directly, which the stem census did not list, so the class_field_set witness read as stale. Register the _known emitter (stem at index 8) and mark the plain emitter as a stem forwarder. Verdicts for seven unaudited raw writes: - async_module.rs store_root: ROOT (fields registered as global roots; the root barrier runs first) - init_guard_tests.rs: POINTER_FREE (integer guard words in a test static) - record_array.rs pop/get: POINTER_FREE (immediate record words copied to the caller's frame) - typedarray/slice_ops.rs fill: POINTER_FREE (element bytes into a typed array data buffer)
722a652 to
49ef8cc
Compare
Summary
GC store-site inventory(python3 scripts/gc_store_site_inventory.py) fails onmain(CI run 43639432681,lint) for two reasons: the stem census lost track of the class-field store, so itsclass_field_setIR witness was reported stale although the emitter call still exists, and seven raw slot writes/copies carry noGC_STORE_AUDITverdict. This PR fixes both. Comments and gate tables only; generated code and runtime behaviour are unchanged.Changes
scripts/gc_store_site_inventory.py: 0b12532 moved the class-field store (expr/property_set.rs) ontoemit_jsvalue_slot_store_pointer_tested_known, a variant the census did not list. Register it inSTEM_EMITTER_ARG_INDEX(stem at index 8, same as the plain emitter) and addemit_jsvalue_slot_store_pointer_testedtoSTEM_FORWARDERS(it now forwards itsstemparameter to_known). The census bindsclass_field_setagain: 25 codegen barrier sites bound to 18 IR-witnessed stems. No registry entry is deleted.async_module.rsstore_root: ROOT — module record fields live in the module globals the collector scans as roots, andruntime_write_barrier_root_nanboxruns first (8999cdb, ES modules: evaluate async siblings while top-level await is suspended #14413).record_array.rsjs_record_array_pop/js_record_array_get: POINTER_FREE — record words are immediates (the module docs: the element buffer is a pointer-free leaf) copied out to the caller's frame; same verdict as the existing push/birth copies in that file (30d4a62, wpfacts, codegen, runtime: an Array of value records is a record array (value records VR-4b) #14487).typedarray/slice_ops.rsfill word loop (covers both word stores): POINTER_FREE — encoded element bytes into a typed array's data buffer (0ad0e85).module_require/init_guard_tests.rs: POINTER_FREE — integer guard words in a test static (8d14c14).Proposed verdicts; please confirm the ROOT classification of the async-module record fields in particular.
Related issue
Refs #14595. Main's CI run: https://github.com/PerryTS/perry/actions/runs/43639432681
Test plan
Measured on macOS arm64:
class_field_setat0b12532106and not at its parent4a2e900fbb(the gate already failed there for other reasons). The other sites' commits above are fromgit log -S(where the line was added), not from a gate probe.upstream/main: FAIL (7 unaudited sites + staleclass_field_setwitness).GC store-site inventory passed (3596 files scanned, 809 audited sites, 160 allowlisted);--self-testpassed.record_array.rsfindings; removing the_knownentry fromSTEM_EMITTER_ARG_INDEXbrings back the staleclass_field_seterror.BASE_SHA=upstream/main scripts/run_lint_gates.sh(full run): FAIL set = main's minusGC store-site inventory; no new FAIL.crates/), so no crate tests were rerun for this branch.Checklist
feat:/fix:/docs:/chore:prefix convention used in the log🤖 Generated with Claude Code
https://claude.ai/code/session_01CUQanVsnitcrmYLAWWPc3p
Summary by CodeRabbit