Repository navigation
Wait for completed Draft typing in the presentation journey - #1047
Conversation
taras
added
the
ci-main-red-fix
PR #1047: Wait for completed Draft typing in the presentation journey1 files, +27 / -1 Scope✅ PR scope looks good. Structural✅ No structural bloat detected. Slop✅ Slop indicators look low. Static Analysis✅ Oxlint found no issues. CorrectnessNo extraneous code patterns detected. |
taras
left a comment
taras
left a comment
There was a problem hiding this comment.
Independent review by Codex gpt-6-astra (high reasoning): PASS at e6f097dfffbe3b00bb6567b44627ea0eb353d576, against base 2b783c3a11754cb46b5e42c6d73495d4a41c19b8. No actionable findings.
The forced abc session-header collision demonstrates a sufficient cause of the original draw-count failure: the whole-screen predicate can accept an already completed reading before typing is presented. The historical failing session ID was not retained, so this does not claim to recover that ID. The replacement predicate checks the completed Draft reading. Holding the next presentation verifies that typing remains pending and both the draw count and Draft reading stay unchanged until release.
The existing differential-write, identity-row, size-profile and write-lifetime assertions remain intact. The change stays within one test and uses the existing scoped host override, presentation hold and page-reading APIs; I found no unnecessary production change or abstraction.
I inspected the retained original-failure witness and focused Deno results: 22 tests and 82 steps pass for the final candidate, whose patch is byte-identical to this signed commit. I also verified CI run 41265614803 is successful for this exact head. I did not rerun broad suites or perform a timing benchmark: no production behavior changes, and this review makes no FPS, latency or memory assurance.
The PR is open and unmerged at the pre-post check. This is a substantive passing review submitted as COMMENT because the authenticated account is the PR author.
The completed-presentation test waited for
abcanywhere on the screen. A generated session ID could already contain that text, allowing the wait to return before Draft typing had been presented and failing the next draw-count assertion.The test now forces that header collision and waits for text specifically in the completed Draft reading. Holding the next terminal write proves the wait stays pending until presentation finishes. Existing diff, profile, frame-count and write-lifetime assertions remain. Only the test changes.
Validation: the forced collision reproduces the original assertion failure before the repair. The corrected REPL journey and page tests pass under Deno: 22 tests, 82 steps. Scoped formatting and lint pass; the feedback commit is signed and matches the tested candidate. The independent Codex Planner accepts signed feedback commit
e6f097df. Required CI 41265614803, including composability for this labelled repair, passes one6f097dfffbe3b00bb6567b44627ea0eb353d576. The one-time independent Codex Astra review passes with no actionable findings: #1047 (review). Astra replaces Claude under the approved quota fallback; its canonical result and public review are verified through XMD.Repairs the current failing boundary tracked in #1043. That automatic health issue remains open until main CI succeeds; this PR does not close it by keyword.