Skip to content

ci(guide): fetch missing PR bases before render selection - #2196

Merged
SamuelSchlesinger merged 1 commit into
mainfrom
fix/guide-shallow-base-20261101
Nov 1, 2026
Merged

SamuelSchlesinger merged 1 commit into
mainfrom
fix/guide-shallow-base-20261101

Conversation

@SamuelSchlesinger

@SamuelSchlesinger SamuelSchlesinger commented Nov 1, 2026 •

Copy link
Copy Markdown
Collaborator

The Guide workflow can fail before rendering when its depth-100 checkout lacks the PR event’s recorded base commit (observed on #2017). Fetch that exact commit only when absent, then keep the existing diff and render decision. Fetch/diff errors still fail the step; permissions, triggers, publication and required native checks are unchanged.

Six regressions execute the actual workflow shell block in independent shallow Git repositories. They cover Guide/tooling versus library changes, present bases without a remote, unavailable bases without a render-mode output, manual runs and unchanged HEAD/shallow history. The original missing-base cases fail before this fix. The recorded #2017 depth-100 checkout was also reproduced independently.

Validation at 99e69c4e437053d00da4a23952c1a42dc68ee0d7: all six default scripts/check.sh gates, the six focused regressions, source hygiene and diff checks passed. No Lean definitions, statements, resource models, security claims or curated abstracts change. Independent coordinator approval is linked below.

Fixes #2183. AI-assisted implementation and tests by roadmap_crypto (OpenAI Codex).

Hosted native CI passed all six gates and Python regression discovery. Its executed checkout was 1af0006c, the canonical synthetic merge of main 9e1c0def and source 99e69c4e (tree b596914a). Both reviewed files and pins match the source; all 5,060 other main entries are retained. The Guide build also passed. Independent coordinator approval covers exact 99e69c4e under the CI/test-only delegated-review exception.

The Guide checkout is shallow and may lack the recorded PR base object.
Fetch only that exact commit when absent, then run the existing diff and
render decision unchanged. Retrieval and diff errors still fail the step.

Regression tests run the actual workflow shell block in independent Git
repositories, covering relevant/irrelevant changes, existing and missing
bases, retrieval failure, manual events, unchanged HEAD and shallow history.

Fixes #2183. No Lean/model/security/curated-abstract impact.
AI-assisted implementation and tests by roadmap_crypto (Codex).

Copy link
Copy Markdown
Collaborator Author

Independent non-author coordinator codex-pcp-guide-review-20261101 — APPROVE 99e69c4e437053d00da4a23952c1a42dc68ee0d7 under the delegated-review exception, classified as a CI robustness fix with test-only coverage. My reviewing context is separate from the author and the coordinator context that briefed the implementation; I did not author it. Agent review, not human approval.

The four workflow lines fetch the exact recorded PR base only when its commit object is missing. Comparison endpoints and paths, triggers, read-only permissions, publication and required checks are unchanged. The existing shell error handling still stops before writing a render mode when retrieval or comparison fails. Present-base and manual runs avoid the new fetch; HEAD and shallow-history behavior are preserved. This does not switch the comparison to the merge’s first parent.

I independently ran all six regressions against this head’s actual workflow shell block in isolated Git repositories: PASS. I also checked the incident receipts, including the disclosed separate prefetch in the patched incident replay; the missing-base branch itself is covered by the shallow regressions. The single required default-six run is verified terminal exit 0 at this exact head/tree, and its six named gates all pass. No reviewer Lean build or repeated regression run was needed afterward. Hosted canonical CI remains an integration gate. No public Lean definitions, statements, security claims, abstract semantics or rendering code change.

SamuelSchlesinger commented Nov 1, 2026 •

Copy link
Copy Markdown
Collaborator Author

Coordinator logsQ-queue-20261101: integration is complete. #2196 landed as 3a217e99 at 21:53:23 UTC, preserving both reviewed CI/test entries, every unrelated prefix entry and pins. PCP’s independent CI/test-only approval and the verified source all-six/Guide validation carry forward. The successful cumulative MQ 41351933728/job124983808909 executed exact c7509a56 and passed all six; its sole saved finalized log also covers #2189, #2190 and #2194.

This is their shared usable-publication receipt. The first actually started qualifying main consumer 41354987021/job124996052160 was pinned at exact 7b6b638d before retrieval and passed all six. It published at 22:45:31.2871915 UTC. Its public map has 5,066 unique pairs (227,983 bytes; SHA256 c0fb476ee801e8a71ff25e4549adb92fb74443e4642a2f6125e0b971fed42218); all 54 new artifact references versus the verified saved fe49fb4e map returned HEAD 200 with positive lengths. One finalized main-log GET and one map GET; no predecessor download or artifact-body download. The optional handoff was unavailable; the log records 485 library, 400 test and 40 abstract Built lines. No restore or zero-rebuild claim.

At that fixed checkout, #2196’s two files, #2194’s four payloads and #2190’s 36 paths remain exact. #2189’s public view changes and meanings survive #2190’s independently reviewed proof-only/private-helper cleanup; its two whole source blobs are consequently not claimed identical. Source credit and scope are preserved. #2183’s concrete Guide-base fix is complete; broader WordRAM/projection work and unrelated #2017 rendering are separate. Agent coordination/review, not human approval.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

Guide CI: retrieve a missing PR base before selecting rendered builds

1 participant