test(typed-array): re-pin fractional-index expectation to spec-correct undefined (unblocks cargo-test on main)

[proggeramlug]

What

Fixes the inline_typed_array_fast_path_matches_slow_path failure that has held cargo-test red on main since the Jul 28 nightly (it replaced the proxy failure that #6907 cleared, and it currently blocks the v0.5.1265 release gate).

This is a stale test expectation, not a code regression — the runtime got more spec-correct and the test was pinned to the old bug.

Root cause

Section (B) asserted g(t, 1.5) === g(t, 1) for t = new Int32Array(3), i.e. that a fractional index truncates to 1. The test's own comment documented this as a deliberate encoding of a known-wrong behavior:

a pre-existing quirk: the untyped dynamic getter truncates a fractional index rather than returning undefined … since the whole point is that inline == slow.

Per §10.4.5.2, a non-integral canonical numeric index is not an own property of an integer-indexed exotic object, so [[Get]] yields undefined. #6903 (representation-selection Phase 1 — canonical unboxed i32 locals) made both the inline and slow paths spec-correct, so the comparison is now false and the pinned string no longer matches.

Attribution is solid from both directions: the Jul 27 nightly (b5727f305, pre-#6903) did not fail this test, the Jul 28 nightly (95960e0df, post-#6903) does, and building b0d789dab (#6903) directly reproduces the new B=true,true,false,true.

Why this is safe to re-pin rather than revert

The invariant the test exists to protect — inline path == slow path — still holds; only the shared value changed. Verified directly with a probe that exercises both paths on the same inputs:

inline t[1.5]=undefined  t[-0.5]=undefined  t[2.7]=undefined
slow   t2[1.5]=undefined t2[-0.5]=undefined t2[2.7]=undefined   # under PERRY_TA_VIEW_GUARD
slow view[0.5]=undefined

node --experimental-strip-types produces exactly the same three lines, and the same B=true,true,false,true for the test's own section (B). So the paths agree with each other and with node.

Changes

• Correct the pinned expectation to B=true,true,false,true and rewrite the (B) comment to explain that the quirk was fixed by perf(codegen): representation-selection Phase 1 — canonical unboxed i32 locals #6903, citing §10.4.5.2, instead of documenting it as permanent.
• Strengthen the test: add a B2 line asserting t[1.5]/t[-0.5]/t[2.7] are each === undefined. The old (B) form only asserted "not equal to t[1]", which a future regression returning some other wrong value (0, garbage, a truncated read of a different slot) would still satisfy. B2 pins the actual value, so this test now guards the corrected behavior rather than merely tolerating it.

Test-only change; no runtime or codegen code is touched.

Summary by CodeRabbit

• Bug Fixes

  • Fixed typed-array access with fractional or NaN-like indices so values such as 1.5 correctly return undefined instead of being treated as truncated integer indices.
  • Ensured consistent results across optimized and fallback execution paths, matching standard runtime behavior.

• Tests

  • Added stronger regression coverage to verify correct results for multiple non-integer index values.

[coderabbitai]

No actionable comments were generated in the recent review. 🎉

ℹ️ Recent review info

⚙️ Run configuration

Configuration used: defaults

Review profile: CHILL

Plan: Pro Plus

Run ID: 05ad220a-b714-47a4-9474-1739d1b7f621

📥 Commits

Reviewing files that changed from the base of the PR and between df68b83 and 9783c40.

📒 Files selected for processing (2)

• changelog.d/6957-ta-fractional-index-expectation.md
• crates/perry/tests/issue_5525_typed_array_untyped_index.rs

---

📝 Walkthrough

Walkthrough

The regression test now verifies that fractional and NaN-like typed-array indices return undefined on inline and slow paths, with updated expected output and changelog documentation.

Changes

Typed-array fractional indexing

Layer / File(s)	Summary
Fractional index regression assertions crates/perry/tests/issue_5525_typed_array_untyped_index.rs, changelog.d/6957-ta-fractional-index-expectation.md	The test explicitly checks undefined for fractional, negative fractional, and NaN-like indices on both paths, adds B2 coverage, updates expected stdout, and documents the corrected behavior.

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

Possibly related PRs

• PerryTS/perry#6850: Changes typed-array reads to handle fractional indices without truncating to an in-bounds element.
• PerryTS/perry#6883: Verifies undefined fallback behavior for fractional typed-array reads.

🚥 Pre-merge checks | ✅ 5

✅ Passed checks (5 passed)

Check name	Status	Explanation
Title check	✅ Passed	The title clearly names the typed-array fractional-index test expectation fix and matches the change set.
Description check	✅ Passed	The description covers the fix, root cause, changes, and validation, though it doesn't use the template's exact section headings.
Docstring Coverage	✅ Passed	Docstring coverage is 100.00% which is sufficient. The required threshold is 80.00%.
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 docstrings

• Create stacked PR
• Commit on current branch

🧪 Generate unit tests (beta)

• Create PR with unit tests
• Commit unit tests in branch fix/ta-fractional-index-expectation

---

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.}