chore: merge train 243 (v0.5.1622)

[proggeramlug]

Merge train 243 — the last compiler-output-regression blocker, released as v0.5.1622.

Contents

source	Change
#10804	fix(codegen): point the module-global Ptr<Shape> denial at #10803, not closed #7109
#10782	ci: give compiler-output-regression's cold runtime builds their own budget

What was still red, and why it is the last one

compiler-output-regression is a main-gate input and has been red for weeks. As of v0.5.1618 its structural verdicts are all fixed — failed_workloads: [] — and the only remaining failure was 2 × TimeoutExpired on h1_native_rep_equivalence, which is the residual recorded when #10782 was closed: that fix suppressed auto-optimize on the HIR probe but deliberately not on the linking compile, so the 343 s cold build simply moved into a different 300 s budget.

The exclusion had to stand. runtime_budgets' allocations_traced / gc_collections_traced / write_barriers_traced are derived solely from the executed binary's PERRY_GC_TRACE stderr, and optimized_libs/freshness.rs:288-293 adds perry-runtime/diagnostics to the rebuild because the compiler sees that variable — a derivation that only runs on the auto-optimize branch. Those budgets are maxima, so a runtime emitting no trace scores 0 and passes all three vacuously.

So the fix pays the cold build in a step of its own, before the gates.

Two cold builds, not one

This is the part worth reading. The obvious design warms once. It is wrong:

The perry-auto-<hash> directory name excludes the cross-feature set, but the build stamp inside it includes it. The harness sets PERRY_GC_TRACE=1 only for workloads carrying *_traced budgets, and that variable adds perry-runtime/diagnostics. loop_bound_semantics is the one gated workload with no *_traced budget — so it wants the other feature set, hits a stamp mismatch, and re-runs cargo. A single-warm design would have left exactly one gate step to absorb a cold build.

The step therefore warms both, then verifies.

It asserts it actually warmed — twice

A pre-warm that silently no-ops is worse than none: the gates then fail exactly as before while looking warmed. Two independent assertions:

1. Presence — a target/perry-auto-*/ holding both a build stamp and a non-empty libperry_runtime.a, or ::error::pre-warm warmed nothing and exit 1.
2. Effect — a third verification compile whose duration must be under half the 300 s budget every gate step runs under, or an explicit error saying auto-optimize is still rebuilding.

The second is the one that discriminates. Presence can pass on a stale or partial directory; "a compile after warming still took 150 s" measures the property directly.

A gate caught the new step, which is the gate working

The first validation aborted at lint:

post-build subject inventory drifted; expected [9 steps] found [10 steps]

scripts/compiler_output_step_liveness.py pins that job's post-build step inventory. Registered in COMPILER_SUBJECTS rather than exempted — it depends on the build and already carries the matching if: guard, so a sibling failure cannot hide it. Gate now passes at 10 subjects, and its own self-test still passes at 8 cases, which was checked separately: a registry edit that broke the checker would be worse than the drift.

Listing it matters beyond satisfying the check. The pre-warm carries its own liveness assertions and can fail, so it is a subject; were it ever silently removed, every gate step below would quietly inherit the timeout it exists to prevent, and the failure would look exactly like the red it just fixed.

Provenance

The #10782 work was produced by an agent that died on an authentication error after committing but before pushing. The commit was recovered locally, rebased twice as main moved beneath it, and pushed. Because it never reached lint, the workflow change was verified here instead: test.yml parses at 28 jobs, the pre-warm sits at step 10 with all seven gates at 11–17, and ci_plan.py --self-test passes.

Validation

Assembled on e2a0839074; both source heads asserted fresh; PR fully represented — zero missing insertions; no attribution trailers. Ten cheap gates, cargo check --workspace --all-targets under -D warnings, all five pinned artifacts byte-identical before and after, six unit suites with an empty failing set, and lint complete at 6-of-6 with nothing outside the known-red public-baseline step on the re-run.

Both compiler-output suites at failed_workloads=[]; repsel_census rc=0 wasted_promotion=False.

Gap sweep at PERRY_RUN_TIMEOUT=30, six areas weighted to #10804's Ptr<Shape> surface, 197 fixtures, every area asserted live, zero unexplained regressions:

class 84   gc_ 54   shape 23   property 18   module 14   const_ 4

[coderabbitai]

Note

Currently processing new changes in this PR. This may take a few minutes, please wait...

⚙️ Run configuration

Configuration used: defaults

Review profile: CHILL

Plan: Advanced

Run ID: 21b35959-6900-4b0d-a92c-58c8c34b9bb8

📥 Commits

Reviewing files that changed from the base of the PR and between e2a0839 and 473ea9b.

⛔ Files ignored due to path filters (1)

• Cargo.lock is excluded by !**/*.lock

📒 Files selected for processing (8)

• .github/workflows/test.yml
• CLAUDE.md
• Cargo.toml
• changelog.d/10782-prewarm-compiler-output.md
• changelog.d/10782-step-liveness-registry.md
• changelog.d/10804-stale-denial-issue-pointer.md
• crates/perry-codegen/src/expr/slot_rep.rs
• scripts/compiler_output_step_liveness.py

 _______________________________________________________
< Once upon a dream, I found all the bugs in your code. >
 -------------------------------------------------------
  \
   \   \
        \ /\
        ( )
      .( o ).

✨ Finishing Touches

📝 Generate docstrings

• Commit to this branch
• Create a new PR

🧪 Generate unit tests (beta)

• Commit to this branch
• Create a new PR

---

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.}