check.py: take the bench directory off sys.path; direction-labelled jit-stats gate; re-record 43 baselines - #1034
Conversation
…at do not reproduce, and resolve a per-platform baseline Three changes to the jit-stats gate. Direction. `_jit_stats_change` now returns (regressions, improvements) instead of one string, classified per counter: a rise in the badness fields or in guard_failures is a regression, a fall in loops_compiled is a regression, and the opposite moves are gains. bridges_compiled is in neither list, so both its directions report as regressions — a fall to 0 is the dead-bridge case when guards still fail and a gain when they stopped (list_length_hint_validate fell 4 -> 0 as guard_failures fell 828 -> 1), and a rise is either wider coverage or a guard storm. Both outcomes still fail; the labels are REGRESSED and IMPROVED, and each carries the --snapshot command that re-records it. Reproducibility. A fixture that disagrees with its baseline is now re-run JITSTATS_STABILITY_RUNS times. If a repeat of the same binary reports different counters, the comparison is reported as UNSTABLE and not gated. Measured on this tree: two runs of one pyre-dynasm binary disagreed by loops_compiled +2 on five unrelated fixtures. Instability is measured per invocation, not annotated per fixture. Per-platform baselines. `_jitstats_baseline_path` prefers <name>.<backend>.<sys.platform>.jitstats when it exists. windows-latest cranelift reports closure_per_call guard_failures 416 and recursive_call_frame_relocation 637 where macos-latest and ubuntu-24.04 both report 415 and 638, on two independent runs, with every other counter equal and that host's dynasm leg passing 371/371. Those two overlays are added here; the shared files still gate the other two platforms exactly. Assisted-by: Claude
The overlay recorded guard_failures=637, while the windows cranelift run produces 638 — the value the shared baseline already carries. The overlay only made the fixture fail on windows. Assisted-by: Claude
…-stats baselines
`sys.path[0]` is the script's own directory, and `pyre/bench/synth` holds over a
thousand entries. The import machinery's scan of that entry is itself a Python
loop, and at that size it crosses the 1039 compile threshold — so a fixture with
no loop anywhere in its body still recorded `loops_compiled=1`, produced by a
single `import` statement. Varying only the directory a fixture runs from, the
fixture itself unchanged: 400 files compile nothing, 800 compile that loop. It
is invariant to PYPY_GC_NURSERY from 1MB to 16MB.
Path handling differs per platform, so that ambient loop landed on either side of
the threshold depending on the host: every jit-stats diff CI reported was a
uniform +-1 on `loops_compiled`, 80 of 80 on windows and 82 of 86 on ubuntu, with
macOS — where the baselines were recorded — reporting none.
`pyre_env` now sets PYTHONSAFEPATH, and an explicit value in the environment
still wins, so passing it empty reaches the old behaviour. `-P` alone rather than
`-I`, which also implies `-E` and would change how the child resolves its stdio
encoding while its stdout is being diffed against an oracle.
The wasm guest has no environment, so the variable is resolved host-side and
handed over through a new `pyre_set_safe_path` export, alongside the existing
`pyre_set_script_path`. A module without the export is left on its previous
behaviour.
40 of the 754 native baselines move — the same 20 fixtures per backend CI listed
— on `loops_compiled` and `guard_failures` only, all downwards. No wasm baseline
moves under the flag.
Three wasm baselines are re-recorded for an unrelated reason: they carry a drift
that is already red on main at this base sha, and two of the three include a
`loops_aborted` rise.
list_append_write_barrier_gc loops_aborted 0->1, loops_compiled 1->12,
bridges_compiled 1->5, guard_failures 200->1335
pickle_terminal_raise_resume loops_compiled 74->73
recursion_memo_branch loops_aborted 1->2, loops_compiled 2->3,
bridges_compiled 11->14, guard_failures 2227->3083
dynasm 377/377, cranelift 377/377 and wasm 373/373 pass locally.
Assisted-by: Claude
The overlay was added for a windows-only `guard_failures` 415 -> 416 on cranelift, before the bench directory was taken off `sys.path`. Whether that delta survives the flag cannot be measured here, and an overlay that is no longer needed does not fail quietly: it shadows the shared baseline permanently, so a stale value becomes a failure main does not have. Removing it leaves at worst the delta main already reports, which is diagnosable from the windows log; the overlay can be restored against a value read back from it. Assisted-by: Claude
WalkthroughThe change adds safe-path propagation for Wasm execution, expands JIT-stat validation with stability and directional classifications, and updates synthetic benchmark baselines for new JIT results. ChangesJIT validation and safe-path execution
Estimated code review effort: 4 (Complex) | ~45 minutes Sequence Diagram(s)sequenceDiagram
participant Check
participant WasmRunner
participant WasmHost
Check->>WasmRunner: Set PYTHONSAFEPATH=1
WasmRunner->>WasmHost: Call pyre_set_safe_path(1)
WasmHost->>WasmHost: Suppress script-directory sys.path entry
sequenceDiagram
participant Check
participant Backend
participant JITStats
Check->>Backend: Execute changed fixture
Backend->>JITStats: Produce counters
Check->>Backend: Repeat fixture
Backend->>JITStats: Produce repeat counters
Check->>Check: Classify the result
Possibly related PRs
Poem
🚥 Pre-merge checks | ✅ 5✅ Passed checks (5 passed)
✨ Finishing Touches📝 Generate docstrings
🧪 Generate unit tests (beta)
Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out. Comment |
![chatgpt-codex-connector[bot]](https://avatars.githubusercontent.com/in/1144995?s=60&v=4){kind=small}
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: c6f768e2cb
ℹ️ About Codex in GitHub
Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you
- Open a pull request for review
- Mark a draft as ready
- Comment "@codex review".
If Codex has suggestions, it will comment; otherwise it will react with 👍.
Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".
| repeats = self._jitstats_repeats(backend, script, timeout) | ||
| drifted = next( | ||
| (s for s in repeats or () if s != jitstats), None | ||
| ) | ||
| if drifted is not None: |
There was a problem hiding this comment.
If one counter has a repeatable regression while any unrelated counter varies between the initial run and a repeat, this branch marks the entire fixture unstable and records it as a pass. For example, a stable loops_aborted 0 -> 1 can be hidden by ordinary guard_failures jitter, even when every repeat retains the new abort. Compare stability per changed field and continue failing fields whose baseline delta persists rather than discarding the whole snapshot.
AGENTS.md reference: AGENTS.md:L238-L246
Useful? React with 👍 / 👎.
| pub extern "C" fn pyre_set_safe_path(enabled: u32) { | ||
| super::SAFE_PATH.with(|f| f.set(enabled != 0)); |
There was a problem hiding this comment.
When the wasm runner receives nonempty PYTHONSAFEPATH, this setter updates only the new TLS cell used by run_python_impl; it never updates importing::SYS_SAFE_PATH, which is what safe_path_flag() and sys.flags.safe_path read. Consequently the script directory is suppressed while sys.flags.safe_path still reports False and other safe-path-dependent interpreter behavior remains disabled, diverging from the native launcher. Route the setting through the existing interpreter-owned runtime flag instead of maintaining a second TLS value.
AGENTS.md reference: AGENTS.md:L148-L162
Useful? React with 👍 / 👎.
| internal_compile_panics=0 | ||
| loops_aborted=0 | ||
| loops_compiled=1 | ||
| loops_aborted=1 |
There was a problem hiding this comment.
This baseline update makes the gate accept loops_aborted rising from 0 to 1, alongside guard_failures rising from 200 to 1335, even though this commit contains no corresponding wasm/JIT change or root-cause explanation; the commit message explicitly calls the drift unrelated. The same masking occurs in recursion_memo_branch.wasm.jitstats (loops_aborted 1 -> 2). Leave these baselines unchanged until the regression is explained or fixed, otherwise future checks treat the degraded behavior as healthy.
AGENTS.md reference: AGENTS.md:L238-L246
Useful? React with 👍 / 👎.
![coderabbitai[bot]](https://avatars.githubusercontent.com/in/347564?s=60&v=4){kind=small}
There was a problem hiding this comment.
Actionable comments posted: 3
🤖 Prompt for all review comments with AI agents
Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.
Inline comments:
In `@pyre/check.py`:
- Around line 1284-1290: Filter out absent repeat snapshots before selecting a
drifted result in the regression/improvement handling around _jitstats_repeats,
so None entries do not terminate next() and later differing snapshots are
detected. Add test coverage for repeats containing [None, changed_snapshot],
ensuring the changed snapshot is reported as drift.
In `@pyre/pyre-wasm-runner/src/main.rs`:
- Around line 379-391: The PYTHONSAFEPATH handling must be enforced on every
Wasm execution path, not silently skipped when the module lacks
pyre_set_safe_path. Update the logic around pyre_set_safe_path and both engine
branches to require the setter when safe path is enabled, propagate an error if
unavailable or ineffective, and ensure this validation occurs before invoking
pyre_run_python.
In `@pyre/pyre-wasm/src/lib.rs`:
- Around line 519-525: Move SAFE_PATH out of thread-local storage and store the
startup setting on the Wasm execution owner or pass it through
PyExecutionContext. Update the public pyre_set_safe_path ABI path and
run_python_impl to read and propagate this owned execution state, preserving the
existing safe-path behavior for wasm-host builds.
🪄 Autofix
Fix all unresolved CodeRabbit comments on this PR:
- Push a commit to this branch (recommended)
- Create a new PR with the fixes
ℹ️ Review info
⚙️ Run configuration
Configuration used: Organization UI
Review profile: ASSERTIVE
Plan: Pro Plus
Run ID: 1cc6ff24-2dc1-4617-bf9a-dad9b89b0393
📒 Files selected for processing (46)
pyre/bench/synth/arith_int_bool.cranelift.jitstatspyre/bench/synth/arith_int_bool.dynasm.jitstatspyre/bench/synth/ast_compile_roundtrip.cranelift.jitstatspyre/bench/synth/ast_compile_roundtrip.dynasm.jitstatspyre/bench/synth/gc_deque_backing_list.cranelift.jitstatspyre/bench/synth/gc_deque_backing_list.dynasm.jitstatspyre/bench/synth/imp_lock_rlock_semantics.cranelift.jitstatspyre/bench/synth/imp_lock_rlock_semantics.dynasm.jitstatspyre/bench/synth/import_from_name_path.cranelift.jitstatspyre/bench/synth/import_from_name_path.dynasm.jitstatspyre/bench/synth/inline_freevar_after_mayforce.cranelift.jitstatspyre/bench/synth/inline_freevar_after_mayforce.dynasm.jitstatspyre/bench/synth/list_append_write_barrier_gc.wasm.jitstatspyre/bench/synth/module_getattr_descr_error.cranelift.jitstatspyre/bench/synth/module_getattr_descr_error.dynasm.jitstatspyre/bench/synth/module_getattr_surrogate_cls.cranelift.jitstatspyre/bench/synth/module_getattr_surrogate_cls.dynasm.jitstatspyre/bench/synth/operator_set_inplace_ops.cranelift.jitstatspyre/bench/synth/operator_set_inplace_ops.dynasm.jitstatspyre/bench/synth/pickle_ctor_args.cranelift.jitstatspyre/bench/synth/pickle_ctor_args.dynasm.jitstatspyre/bench/synth/pickle_terminal_raise_resume.wasm.jitstatspyre/bench/synth/print_stdout_redirect.cranelift.jitstatspyre/bench/synth/print_stdout_redirect.dynasm.jitstatspyre/bench/synth/pypy_dict_primitives_nonbinding.cranelift.jitstatspyre/bench/synth/pypy_dict_primitives_nonbinding.dynasm.jitstatspyre/bench/synth/recursion_memo_branch.wasm.jitstatspyre/bench/synth/seqiter_tuple_error_parity.cranelift.jitstatspyre/bench/synth/seqiter_tuple_error_parity.dynasm.jitstatspyre/bench/synth/simple_namespace_type.cranelift.jitstatspyre/bench/synth/simple_namespace_type.dynasm.jitstatspyre/bench/synth/sre_pattern_methods.cranelift.jitstatspyre/bench/synth/sre_pattern_methods.dynasm.jitstatspyre/bench/synth/sre_wasm_min.cranelift.jitstatspyre/bench/synth/sre_wasm_min.dynasm.jitstatspyre/bench/synth/sre_wasm_min1.cranelift.jitstatspyre/bench/synth/sre_wasm_min1.dynasm.jitstatspyre/bench/synth/str_encode_text_codec.cranelift.jitstatspyre/bench/synth/str_encode_text_codec.dynasm.jitstatspyre/bench/synth/struct_pack_unpack.cranelift.jitstatspyre/bench/synth/struct_pack_unpack.dynasm.jitstatspyre/bench/synth/type_dotted_name.cranelift.jitstatspyre/bench/synth/type_dotted_name.dynasm.jitstatspyre/check.pypyre/pyre-wasm-runner/src/main.rspyre/pyre-wasm/src/lib.rs
| if regressions or improvements: | ||
| repeats = self._jitstats_repeats(backend, script, timeout) | ||
| drifted = next( | ||
| (s for s in repeats or () if s != jitstats), None | ||
| ) | ||
| if drifted is not None: | ||
| moved = _jit_stats_change(jitstats, drifted) |
There was a problem hiding this comment.
🎯 Functional Correctness | 🟡 Minor | ⚡ Quick win
Skip absent repeat snapshots before selecting drift.
Line 1286 uses None both as the no-drift sentinel and as a possible repeat snapshot. If the first repeat has no [jit-stats] line, it ends next() and prevents a later differing repeat from being checked. The fixture can then be gated instead of reported unstable.
Filter missing repeat snapshots before selecting drift. Add coverage for [None, changed_snapshot].
Proposed fix
drifted = next(
- (s for s in repeats or () if s != jitstats), None
+ (
+ snapshot
+ for snapshot in repeats or ()
+ if snapshot is not None and snapshot != jitstats
+ ),
+ None,
)📝 Committable suggestion
‼️ IMPORTANT
Carefully review the code before committing. Ensure that it accurately replaces the highlighted code, contains no missing lines, and has no issues with indentation. Thoroughly test & benchmark the code to ensure it meets the requirements.
| if regressions or improvements: | |
| repeats = self._jitstats_repeats(backend, script, timeout) | |
| drifted = next( | |
| (s for s in repeats or () if s != jitstats), None | |
| ) | |
| if drifted is not None: | |
| moved = _jit_stats_change(jitstats, drifted) | |
| if regressions or improvements: | |
| repeats = self._jitstats_repeats(backend, script, timeout) | |
| drifted = next( | |
| ( | |
| snapshot | |
| for snapshot in repeats or () | |
| if snapshot is not None and snapshot != jitstats | |
| ), | |
| None, | |
| ) | |
| if drifted is not None: | |
| moved = _jit_stats_change(jitstats, drifted) |
🤖 Prompt for AI Agents
Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.
In `@pyre/check.py` around lines 1284 - 1290, Filter out absent repeat snapshots
before selecting a drifted result in the regression/improvement handling around
_jitstats_repeats, so None entries do not terminate next() and later differing
snapshots are detected. Add test coverage for repeats containing [None,
changed_snapshot], ensuring the changed snapshot is reported as drift.
| // `-P` / PYTHONSAFEPATH, resolved host-side: the guest's environment is | ||
| // permanently empty, so the variable has to be handed over explicitly or it | ||
| // would read as unset there while working natively. A presence flag, matching | ||
| // `pyrex::resolve_safe_path` — any non-empty value enables it, `"0"` | ||
| // included, and an empty value counts as unset. Absent on a module predating | ||
| // the export, leaving the guest on its default of seeding `sys.path[0]`. | ||
| if std::env::var_os("PYTHONSAFEPATH").is_some_and(|value| !value.is_empty()) { | ||
| if let Ok(set_safe_path) = | ||
| instance.get_typed_func::<u32, ()>(&mut store, "pyre_set_safe_path") | ||
| { | ||
| set_safe_path.call(&mut store, 1)?; | ||
| } | ||
| } |
There was a problem hiding this comment.
🔒 Security & Privacy | 🟠 Major | ⚡ Quick win
🧩 Analysis chain
🏁 Script executed:
#!/bin/bash
set -euo pipefail
rg -n -C 8 'WasmEngine::Wasmi|wasmi_host::run|PYTHONSAFEPATH|pyre_set_safe_path' pyre/pyre-wasm-runner/src/main.rs
rg -n -C 6 'SAFE_PATH|SCRIPT_PATH|add_sys_path|pyre_set_safe_path' pyre/pyre-wasm/src/lib.rsRepository: youknowone/pyre
Length of output: 7647
🏁 Script executed:
#!/bin/bash
set -euo pipefail
echo "== wasmi_host outline =="
ast-grep outline pyre/pyre-wasm-runner/src/wasmi_host.rs --view compact || true
echo "== wasmi_host relevant symbols =="
rg -n -C 8 'run\(|inspect\(|pyre_set_safe_path|pyre_set_script_path|wasmi|WasmModule|Memory|Table|Function' pyre/pyre-wasm-runner/src/wasmi_host.rs
echo "== main safe path/script path call context =="
sed -n '340,420p' pyre/pyre-wasm-runner/src/main.rsRepository: youknowone/pyre
Length of output: 12823
Security And Privacy (CWE-426): Untrusted Search Path
Reachability: External · Exploitability: Moderate
Reachability path
● Entry
pyre/pyre-wasm/src/lib.rs:818
pyre_set_safe_path
│
▼
● Hop
pyre/check.py:1019
Check: Measure each timed interpreter/backend's empty-program user-CPU cost. Runs an empty script STARTUP_SAMPLES times per interpreter and reco…
│
▼
● Sink
pyre/pyre-wasm-runner/src/main.rs
Enforce PYTHONSAFEPATH before running either Wasm engine.
Both engine paths currently continue when a module lacks the new pyre_set_safe_path export. When the caller enables safe path, sys.path[0] is still seeded for modules that do not export the setter; call it on every execution path and fail before pyre_run_python if the selected module cannot enforce the requested policy.
🤖 Prompt for AI Agents
Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.
In `@pyre/pyre-wasm-runner/src/main.rs` around lines 379 - 391, The PYTHONSAFEPATH
handling must be enforced on every Wasm execution path, not silently skipped
when the module lacks pyre_set_safe_path. Update the logic around
pyre_set_safe_path and both engine branches to require the setter when safe path
is enabled, propagate an error if unavailable or ineffective, and ensure this
validation occurs before invoking pyre_run_python.
| /// `-P` / PYTHONSAFEPATH, which suppresses the `sys.path[0]` entry. The | ||
| /// guest has no environment, so the embedder passes the resolved flag in | ||
| /// through `pyre_set_safe_path` rather than the variable being read here. | ||
| /// Only the native-host binding seeds that entry — the browser build has no | ||
| /// filesystem to seed it from. | ||
| #[cfg(feature = "wasm-host")] | ||
| static SAFE_PATH: std::cell::Cell<bool> = const { std::cell::Cell::new(false) }; |
There was a problem hiding this comment.
📐 Maintainability & Code Quality | 🟠 Major | 🏗️ Heavy lift
🧩 Analysis chain
🏁 Script executed:
#!/bin/bash
set -euo pipefail
rg -n -C 6 'thread_local!|SCRIPT_PATH|SAFE_PATH|pyre_set_script_path|pyre_set_safe_path|run_python_impl' pyre/pyre-wasm/src/lib.rs
rg -n -C 4 'resolve_safe_path|PYTHONSAFEPATH|pymain_sys_path_add_path0' .Repository: youknowone/pyre
Length of output: 29272
🏁 Script executed:
#!/bin/bash
set -euo pipefail
echo "== pyre/pyre-wasm/src/lib.rs relevant outline =="
ast-grep outline pyre/pyre-wasm/src/lib.rs --view compact 2>/dev/null | sed -n '1,220p' || true
echo "== pyre/pyre-interpreter/src/importing.rs SYS_SAFE_PATH and safe_path_flag =="
sed -n '1,80p;380,435p;1848,1872p' pyre/pyre-interpreter/src/importing.rs
echo "== pyre/pyrex/src/lib.rs LaunchFlags / resolve_safe_path =="
sed -n '1,260p' pyre/pyrex/src/lib.rs
echo "== safe_path usages in pyre-interpreter/src/importing.rs =="
rg -n -C 3 'safe_path|safe_path_flag|SYS_SAFE_PATH|add_sys_path_0|add_sys_path' pyre/pyre-interpreter/src/importing.rsRepository: youknowone/pyre
Length of output: 22135
Move SAFE_PATH out of TLS.
SAFE_PATH is interpreter startup state selected by the public WASM host ABI and used by run_python_impl. The upstream Python implementation maps -P / PYTHONSAFEPATH into interpreter configuration; store it on the Wasm execution owner or stage it via PyExecutionContext instead of thread_local! with no upstream citation or exceptional-TLS justification.
🤖 Prompt for AI Agents
Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.
In `@pyre/pyre-wasm/src/lib.rs` around lines 519 - 525, Move SAFE_PATH out of
thread-local storage and store the startup setting on the Wasm execution owner
or pass it through PyExecutionContext. Update the public pyre_set_safe_path ABI
path and run_python_impl to read and propagate this owned execution state,
preserving the existing safe-path behavior for wasm-host builds.
Source: Coding guidelines
🤖 Codex parity reviewStatic analysis of this diff vs the local RPython/PyPy sources (commit c6f768e). Files in the reviewed diff1. Regressions to PyPy parity introduced by this patchNone. 2. Other mismatches introduced by this patch
3. Pre-existing mismatches (already present before this patch)None. 4. Structural adaptations
|
1 participant
What the cross-platform jit-stats red actually was
sys.path[0]is the script's own directory, andpyre/bench/synthholds over athousand entries. The import machinery's scan of that entry is itself a Python
loop, and at that size it crosses the 1039 compile threshold — so a fixture with
no loop anywhere in its body still recorded
loops_compiled=1, produced by asingle
importstatement.Decomposed in the real bench directory:
loops_compiledpassfrom types import SimpleNamespaceimport re, collections, struct, jsonsimple_namespace_type.pyfixtureAnd varying only the directory a fixture runs from, the fixture itself unchanged:
loops_compiledsynth/It is invariant to
PYPY_GC_NURSERYacross 1MB..16MB, so it is not thecollection schedule.
Path handling differs per platform, so that ambient loop landed on either side of
the threshold depending on the host. Every jit-stats diff CI reported was a
uniform ±1 on
loops_compiled— 80 of 80 on windows, 82 of 86 on ubuntu, andnone at all on macOS, where the baselines were recorded. Setting the flag on
macOS reproduces the Linux numbers exactly:
arith_int_bool8 → 7 withguard_failures2213 → 2211,gc_deque_backing_list6 → 5 / 204 → 203,struct_pack_unpack2 → 1 / 2 → 1.The change
pyre_envsetsPYTHONSAFEPATH. An explicit value in the environment stillwins, and pyre reads it as a presence flag, so passing it empty reaches the old
behaviour for an A/B.
-Pand not-I: the latter also implies-E, which dropsPYTHONIOENCODINGand so changes how the child resolves its stdio encoding — not something to fold
into a run whose stdout is being diffed against an oracle.
The wasm guest has no environment at all, so the variable is resolved host-side
and handed over through a new
pyre_set_safe_pathexport, alongside the existingpyre_set_script_path. A module built without the export is left on its previousbehaviour. No wasm baseline moves under the flag — that backend never had the
sensitivity — but the plumbing keeps the three backends on one contract.
All 423 bench
.pyfiles were checked for sibling and relative imports; thereare none, so nothing depends on the entry being there.
Baselines
40 of the 754 native baselines move — the same 20 fixtures per backend CI listed
— on
loops_compiledandguard_failuresonly, all downwards.Three wasm baselines are re-recorded for an unrelated reason: they carry a drift
already red on main at this base sha, and two of the three include a
loops_abortedrise, so they are worth a look on their own.Also here
The jit-stats gate now labels a move by direction —
regressedfor a changetoward more aborts or fewer compiled loops,
improvedfor the other way, andanything ambiguous as
regressed. Both still fail; the split is for reading alog, not for deciding one. A fixture whose counters do not reproduce across a
re-run is reported
unstableand not gated on, so a genuinely variable countercannot be laundered into a baseline.
_jitstats_baseline_pathresolves a<name>.<backend>.<platform>.jitstatsoverlay when one exists. Both overlays this branch had are removed: one froze
guard_failures=637while the shared baseline already carried 638 and windowsproduced 638, which made a fixture fail that passed on main. The mechanism is
kept, the two uses are not — an overlay shadows the shared file permanently, so
it has to be justified against the current value, not a remembered one.
Verification
dynasm 377/377,cranelift 377/377,wasm 373/373locally, on LLBCre-extracted at this base.
— authored by Claude
Summary by CodeRabbit
Chores
Improvements