docs(ci): attribute the native_owned_typed_views control_flow misses to #8457

[proggeramlug]

Records the attribution for #8489 and corrects three claims in the existing rationale that were all false. Comment and changelog only — allowed_missed_reason_kinds is unchanged (#8490 already widened it on 2026-08-20; the release was never actually blocked after that).

Attribution: 40e2ce86d (#8457, "make -Os the default native optimization level")

Found from CI job history rather than guessed: the native-ABI step was green at 526e0b502 (Aug 20 03:31) and red at 40e2ce86d (07:28). That window holds exactly two commits, and the other (f14a9e262, #8418) measures clean.

commit	control_flow	observed
526e0b502 (parent of window)	0	call 1, generic 2, unknown 1
f14a9e262 (#8418)	0	identical
40e2ce86d (#8457)	3	call 4, cf 3, generic 8, unknown 1

40e2ce86d's counts match CI's Linux numbers (call 4, cf 3, unknown 1) exactly.

Verdict: legitimate, not a codegen regression

Perry's emitted IR is byte-identical across the flip — same sha256 at 526e0b502, f14a9e262 and 40e2ce86d. Codegen output did not change at all. What changed is the clang args:

526e0b502: analysis_clang_args = ['-O3', ...]   + -mllvm -inlinehint-threshold=850
40e2ce86d: analysis_clang_args = ['-Os', ...]

The harness's remark pass reuses the compile's own args (_analysis_args_from_metadata), so the analysis moved to -Os too. At -Os LLVM declines the SimplifyCFG speculation and if-conversion that at -O3 flattened those branches into selects before the vectorizer ran — so the loops genuinely carry control flow it cannot if-convert, and vectorizing them would grow code size, which is exactly what -Os asks it not to do. -Os cost nothing measurable: vectorized_buffer_transform still vectorizes 2 loops against a floor of 1.

So the widening in #8490 was the right call. This PR only replaces its rationale, which recorded three now-falsified claims and no cause.

The three corrections

1. Not Linux-only. The gate builds with cargo build -p perry → --perry target/debug/perry (test.yml:1809, :1821) — a debug build. The macOS attempt that "did not reproduce" used --profile perry-dev (opt-level 1), a different profile, not a different OS. It reproduces on macOS with a CI-matching debug build.
2. Not release-configuration-specific, for the same reason.
3. Not the string-codegen work. control_flow=3 already at 5876c9cdd, which precedes refactor(stdlib): centralize string header readers #8448/perf(codegen): fuse String.concat arguments into concat chain #8450/fix(ui): replace borrowed runtime strings with owned copies #8453/perf(codegen): append strings through variable cells #8454. And 40e2ce86d is an ancestor of the 3627657c7 attribution baseline — which is precisely why that baseline looked "pre-existing" and made the original attribution circular.

Validation

• all 30 lint-job gates pass
• benchmarks/compiler_output/workloads.toml parses; allowed_missed_reason_kinds byte-for-byte unchanged
• no version bump; no code change

Summary by CodeRabbit

• Documentation
  • Clarified vectorization notes to explain reproducible control-flow misses across platforms and build configurations.
  • Corrected the changelog to attribute the behavior to optimization-level changes.
  • Added historical context and clarified that the documented expectations help detect future changes.

[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: 1d027b6a-8ef0-4e54-b4db-b2c14078af64

📥 Commits

Reviewing files that changed from the base of the PR and between d9d39a6 and 7286205.

📒 Files selected for processing (2)

• benchmarks/compiler_output/workloads.toml
• changelog.d/8490-vectorization-allow-control-flow.md

---

📝 Walkthrough

Walkthrough

The change corrects documentation for three control_flow vectorization misses. It attributes them to LLVM behavior under -Os, records cross-platform reproducibility, and notes that the behavior predates related string-codegen changes.

Changes

Vectorization documentation

Layer / File(s)	Summary
Document vectorization cause and scope benchmarks/compiler_output/workloads.toml, changelog.d/8490-vectorization-allow-control-flow.md	The comments and changelog now describe the -Os cause, cross-platform reproducibility, unchanged emitted IR, and relevant change history.

Estimated code review effort: 1 (Trivial) | ~3 minutes

Suggested reviewers: jdalton

✨ Finishing Touches

🧪 Generate unit tests (beta)

• Create PR with unit tests
• Commit unit tests in branch fix/8489-attribution

---

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.}