Expose the last normalize failure on /tasks/.../status - #84
Merged
Conversation
The normalize gate is the most common reason a dispatched task opens no PR, but on the error path the terminal `error` string names only the last symptom. The 2026-07-29 nightly reported the longcat_flash group as "LLM returned unparseable output"; what actually doomed it was the normalizer failing on check_docstrings, three corrections earlier. That detail only existed in the job history, so the dispatcher rendering the triage issue could not explain the outcome. Add `normalizer_error` to the status payload, derived from the most recent persisted `normalize_error` event (tail-bounded — checker output is tail-informative). Derived from history rather than a new column so it works for jobs already on disk. Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Why
On the 2026-07-29 nightly, the
longcat_flashintegration-failure group was reported in transformers#47631 as⚠️ task failed/ "LLM returned unparseable output". That is the last symptom, not the cause.What actually happened (job
34d4b1f3…): the model produced a patch, the normalizer rejected it with1 failed: docstrings, both corrections hit the same unfixable failure, and the tool-free re-asks then came back empty — so the run surfaced the empty-answer error and the normalizer failure was never mentioned. It only existed in the job history, which the dispatcher rendering the triage issue cannot see.The normalize gate is the single most common reason a dispatched task opens no PR, so the caller should be able to explain the outcome without a dashboard round trip.
What
GET /tasks/{owner}/{repo}/{job_id}/statusnow returnsnormalizer_error: the most recent persistednormalize_errorevent, tail-bounded to 8k chars (checker output is tail-informative — traceback, then theN failed: <names>summary).Derived from the persisted history rather than a new column, so it works for jobs already on disk and needs no migration.
nullwhen the normalizer never rejected a patch.Consumed by a companion transformers-ci PR that renders it in the triage issue's outcome recap.
Test
tests/test_webapp_tasks.py: last-failure-wins when several are recorded,nullwhen the gate never failed, tail-preserving bound, and tolerance of missing/blank history. Full suite: 618 passed.🤖 Generated with Claude Code