app parity: host-leg row check, Rust self-skip rule, checklist 2026-10-06 revision (#1629 follow-ups)

[GoKubar]

Lane P5-APP-PARITY-FINISH (kabir-claude). These are the Commander's follow-ups to #1629, now that #1627, #1628, #1630, #1632 and #1633 have landed.

Changes

1. Host-leg row check. mac-v3-host (ci: run the engine-v3 host tests for real (mac-v3-host); A19 and B3 host tests #1633) now runs app_parity_rows.py check-log --leg host on its swift test log. A host-driven row test that is missing, skipped or failed now fails that job. A change to the rows file or its check now triggers mac-app and mac-v3-host on a pull request (the apple and v3_app path filters).
2. Rust row tests that self-skip fail the row check. A Rust test that returns early still reports ok. check now fails any named Rust test whose body can return early, unless FLASHTEX_REQUIRE_TEXLIVE=1 turns that skip into a failure (checked in the test file or its common module). The function body is found by brace matching.
3. Rows file regenerated after the sibling PRs. A13, A14, A19, B3, C9, C20, C21, C25, D2 and E4 now name their new tests. host_driven gains the A19 and B3 host tests, bringing it to 49. C21, C25, D5 and B9 are now done.
4. Checklist docs/evidence/app-parity-2026-10-05 revised for 2026-10-06.
   • The rows above, plus a counts line: 60 done, 0 partial, 1 missing (A9), 8 different.
   • A rewritten remaining-work section.
   • The symlink bug B3 found (the fix is a follow-up PR).
   • Plan §4.4 records the build.

Verified (on the PC, flashtex.slice)

• python3 -m unittest discover -s tools/parity: 255 OK, 10 skipped.
• app_parity_rows.py check: 69 rows, 118 named tests, all found, no Rust self-skips.
• check-log --leg host against ci: run the engine-v3 host tests for real (mac-v3-host); A19 and B3 host tests #1633's green host log (run 37513534865, artifact mac-host-swift-test-log, 233 executed, 2 skipped benches): 49 of 49 host-driven row tests passed, not skipped.
• The self-skip rule run against cli.rs as it was before release: ship flashtex-host in the app and flashtex-v3 + host in the CLI tarball (row D5) #1632 (d36a00f) flags all three D5 tests. Against main it is clean.
• retirement_refs.py: 0 problems.

Where the gate stands (app_parity_rows.py gate S5)

S5 app parity: not met. The only reason left is owner retirement not in writing: A6, A7, A18, C5, C23. All 5 are "different (intended)" rows, and they need the owner's rulings (link each in owner_retirement). Every other S5 row has named tests that CI now requires to pass, not skip.

S6 also waits on the A9/A17 fallbacks (M3, the bundle route) and E2, E3, E5.

🤖 Generated with Claude Code

https://claude.ai/code/session_01SDAiBvi7bfkb19jLNPTtjp

[GoKubar]

CI on head 78ea218, run 37535981831:

• mac-app (job 112516995764): the hosted-leg row check passed.
• mac-v3-host (rerun job 112542774922): app-parity-rows (host leg): 49 of 49 row tests passed, not skipped. Swift reported 242 executed, 2 skipped (the benches) and 0 failures.
• CI required (job 112542778993): success.

The first attempt of mac-v3-host (job 112516995727) failed in a test that is not a row test: EngineV3ExportTests.testEditsDuringTheExportAreHeldThenSentAndThePreviewStaysIntact failed with "timeout waiting for the export run to start" after 91 s. It passed on the rerun, and on #1633's and #1637's runs. I read it as a timing flake on the hosted runner. That job files main-red after merge, so this test may need a longer bound.