fix(gc): the statepoint bridge must refuse an invoke it cannot root (#7327)

[proggeramlug]

Fixes #7327.

The defect

The bridge's line classifier matched only call:

let is_call = trimmed.starts_with("call ") || trimmed.contains(" = call ")
           || trimmed.starts_with("tail call ") || trimmed.contains(" = tail call ");
if !is_call { continue; }

An invoke therefore skipped both the statepoint conversion and the fail-closed panic! below it, and passed through as an ordinary line.

Since #7302 moved exception lowering to invoke/landingpad, that is every call inside a try. Measured on one program: 58 invokes, 0 carrying gc.statepoint, with allocating callees among them — js_object_alloc_class_inline_keys, js_array_push_f64, js_native_call_method_by_id. Those frames had no roots at all.

The instrument was blind to it too

--statepoint-report counts only lines it recognises, so "0 parser fallbacks" said nothing about any call inside a try — including in #7314's headline census (23,301 safepoints, 0 fallbacks). Invokes are now counted, so the census stops overstating its own coverage.

Why refuse rather than implement

Forming a statepoint from an invoke is real work: the statepoint must itself become an invoke, with gc.result and the relocates in the normal successor. RS4GC already does this correctly.

Until the bridge does, refusing is the same fail-closed rule the plain stack-map fallback was deleted for in #7314 — "a loud compile failure beats silent heap corruption." The refusal is scoped: it fires only when the invoke actually has live roots across it and the callee is neither an llvm.* intrinsic nor in the audited cannot-collect table.

Verified

• Bridge on a try-carrying program now refuses, naming the function and 7 live roots, instead of silently emitting a rootless frame.
• A program with no try still compiles under the bridge — the refusal does not over-fire.
• 11 precise_roots tests pass; file-size, GC store-site, addr-class gates green; cargo fmt clean.

What this exposes — worth reading

There is currently no working statepoint path for try-carrying code on a default toolchain. RS4GC handles invokes, but on this machine it fails with Apple clang 21:

error: unterminated attribute group

and needs PERRY_LLVM_CLANG pointed at a version-matched LLVM 22. Verified pre-existing — it fails identically with an unmodified binary built before this change. The diagnostic says so rather than sending people to a backend that will not build for them.

That gap matters for the shadow-stack removal plan: exceptions are not an edge case, so either the bridge learns invokes or RS4GC becomes reliably usable.

Not verified

Whether this clears any of the 13 gap-suite regressions the statepoint soak found — it converts them from silent wrong answers into compile refusals, which is an improvement in kind but not a pass. Measuring that needs the bridge to actually support invokes.

Summary by CodeRabbit

• Bug Fixes

  • Improved handling and reporting of invoke instructions during statepoint processing.
  • The compiler now explicitly rejects unsupported potentially collecting calls instead of silently omitting them from garbage-collection tracking.
  • Calls without live roots and known non-collecting calls continue to be handled safely.

• Documentation

  • Added guidance on the current limitation and the supported RS4GC workflow.

[coderabbitai]

Caution

Review failed

The pull request is closed.

ℹ️ Recent review info

⚙️ Run configuration

Configuration used: defaults

Review profile: CHILL

Plan: Pro Plus

Run ID: 567e82c3-b52e-45b2-9e53-658c48d79902

📥 Commits

Reviewing files that changed from the base of the PR and between b06b7c7 and febd3c8.

📒 Files selected for processing (2)

• changelog.d/7330-bridge-refuses-invoke.md
• crates/perry-codegen/src/function/precise_roots.rs

---

📝 Walkthrough

Walkthrough

The explicit statepoint bridge now detects invoke instructions, analyzes their live roots, reports them, and refuses unsupported potentially collecting invokes. The changelog documents the refusal behavior and the LLVM 22 requirement for RS4GC.

Changes

Invoke statepoint handling

Layer / File(s)	Summary
Invoke analysis and refusal crates/perry-codegen/src/function/precise_roots.rs, changelog.d/7330-bridge-refuses-invoke.md	Statepoint lowering records invoke instructions and their live roots. It excludes LLVM intrinsics and audited non-collecting calls, then panics for potentially collecting invokes with live roots because conversion is unsupported. The changelog documents this behavior and the RS4GC LLVM 22 requirement.

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

Possibly related PRs

• PerryTS/perry#7305: Adds the related invoke-based exception-handling changes that this PR extends with root analysis and refusal behavior.

✨ Finishing Touches

📝 Generate docstrings

• Create stacked PR
• Commit on current branch

🧪 Generate unit tests (beta)

• Create PR with unit tests
• Commit unit tests in branch fix/7327-bridge-refuses-invoke

---

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.}