jit(bridge): read the traced frame's vable pointer when patching a retraced loop

[fregataa]

Summary

python3 pyre/check.py --backend dynasm failed on synth/str_search_index_bounds with:

rust panic: compile.py:458 assert i == len(inputargs) failed (16 != 26)

A loop compiled from a bridge (trace_and_compile_from_bridge) reads the virtualizable's array lengths through orig_vable_ptr_from_trace_ctx, which preferred the MetaInterp's cached vable_ptr. That cache is clobbered when a residual call re-enters compiled code and deopts nested frames, and the residual-call epilogue only restores TraceCtx::virtualizable_heap_ptr — so the compile-time length read hit a nested callee frame whose locals_cells_stack_w (8 slots) is shorter than the traced frame's (18 slots), tripping the field-load preamble's inputargs accounting (16 = 2 reds + 6 statics + 8 vs the trace's 26 expanded inputargs).

Fix

Prefer the trace-local TraceCtx::virtualizable_heap_ptr in the bridge fallback — RPython's orig_inpargs is the trace's own history, so the trace-local pointer is the parity-correct source — and keep the MetaInterp cache only as a last resort.

Testing

• synth/str_search_index_bounds completes and its output matches CPython exactly
• cargo test -p majit-metainterp --release: 1510 passed
• python3 pyre/check.py --backend dynasm: 342/342 ALL PASSED

🤖 Generated with Claude Code

Summary by CodeRabbit

• Bug Fixes
  • Improved bridge-trace handling during re-entrant deoptimization.
  • Corrected virtualizable state selection to use the most accurate available runtime pointer, improving stability in affected execution paths.

[coderabbitai]

No actionable comments were generated in the recent review. 🎉

ℹ️ Recent review info

⚙️ Run configuration

Configuration used: Organization UI

Review profile: ASSERTIVE

Plan: Pro Plus

Run ID: 27607024-9374-45f3-9b7f-3ff9b3a52d66

📥 Commits

Reviewing files that changed from the base of the PR and between 83fe239 and 00cbd33.

📒 Files selected for processing (1)

• majit/majit-metainterp/src/pyjitpl.rs

---

Walkthrough

Bridge-trace virtualizable heap pointer selection now checks the context-provided pointer first, ignoring null values, and falls back to self.vable_ptr when unavailable.

Changes

Bridge pointer selection

Layer / File(s)	Summary
Update bridge pointer fallback majit/majit-metainterp/src/pyjitpl.rs	Bridge tracing prioritizes a non-null ctx.virtualizable_heap_ptr() and otherwise returns self.vable_ptr; comments describe the updated pointer sources during re-entrant deoptimization.

Estimated code review effort: 2 (Simple) | ~10 minutes

Possibly related PRs

• youknowone/pyre#501: Reworks virtualizable identity and resume decoding used during deoptimization.
• youknowone/pyre#834: Updates virtualizable identity, box, and bridge pointer initialization logic.

Suggested reviewers: youknowone

Poem

A bunny hops through bridge-trace air,
Checking context pointers with care.
If nullness blocks the way,
The frame pointer saves the day—
Deoptimization finds its hare!

🚥 Pre-merge checks | ✅ 5

✅ Passed checks (5 passed)

Check name	Status	Explanation
Description Check	✅ Passed	Check skipped - CodeRabbit’s high-level summary is enabled.
Title check	✅ Passed	The title accurately reflects the main fix: preferring the trace-local vable pointer when patching a retraced bridge loop.
Docstring Coverage	✅ Passed	No functions found in the changed files to evaluate docstring coverage. Skipping docstring coverage check.
Linked Issues check	✅ Passed	Check skipped because no linked issues were found for this pull request.
Out of Scope Changes check	✅ Passed	Check skipped because no linked issues were found for this pull request.

✨ Finishing Touches

🧪 Generate unit tests (beta)

• Create PR with unit tests

---

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.

❤️ Share

• X
• Mastodon
• Reddit
• LinkedIn

_{Comment @coderabbitai help to get the list of available commands.}

[fregataa]

Closed since #876 has already resolved the same issue

[youknowone]

sorry! I missed this patch