Repository navigation
Conversation
The bundled-Otto check captured help with `|| true`, so any way the invocation broke — a non-zero exit, a panic, no output at all — reached the same branch and printed "Astro CLI 1.45.0 does not bundle Otto. Otto requires Astro CLI >= 1.42.0". 1.45.0 does bundle Otto. A run failed that way and the rerun was green, which sent someone looking at the CLI version for a problem that was never there. The check now keeps the exit status alongside the help text and reports the two cases apart: astro could not be asked, or astro answered and the answer has no otto in it. Both print what astro actually returned. It also retries once. `astro --help` makes no network calls, so an empty or non-zero result is the invocation breaking rather than the CLI. `astro otto --help` downloads Otto on first use, which is a real network call and a real reason to give it a second go. Same change in the GitLab adapter, which carries its own copy of the check.
Otto ReviewSplit the Otto verification into "astro couldn't be asked" vs "astro answered without otto", with a one-shot retry, in both the composite action and the GitLab script. No issues found The refactor cleanly separates the two failure modes the old |
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
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
Description
A run failed with
Astro CLI 1.45.0 at /home/runner/.local/bin/astro does not bundle Otto. Otto requires Astro CLI >= 1.42.0. Set the 'astro-cli-version' input to '1.42.0' or newer.1.45.0 does bundle Otto — I pulled the linux 1.45.0 binary in a clean container and ran the check as written, and both greps pass. The rerun was green. The message sent someone looking at the CLI version for a problem that was never there.The cause is
|| trueon the help capture.astro --helpexiting non-zero, panicking, or printing nothing all land in the same branch as an old CLI, and the branch prints a claim the step never established — it only knows it did not see the wordotto. The check now keeps the exit status next to the help text and reports the two cases apart: astro could not be asked, or astro answered and the answer has noottoin it. Both print what astro actually returned, so the next one is diagnosable from the log.It also retries once.
astro --helpmakes no network calls and is deterministic, so an empty or non-zero result is the invocation breaking rather than the CLI.astro otto --helpdownloads Otto on first use, which is a real network call and a real reason to give it a second go.Call-stack diff
gitlab/run-review.shcarries its own copy of this check and gets the same treatment, callingfailwhere the action writes::error::. Unchanged: the two greps, the step's inputs and outputs, and every other step in the action.What each case prints now, against a real linux astro 1.45.0 and against stubs:
astro --helpmakes no network calls and has no update check, so the help text itself is not flaky — something about the invocation was, and the old code threw away the evidence. This change does not identify that; it retries past it and records it next time.astro versionand comparing against 1.42.0. It would have caught this case, but it goes stale every time the floor moves, and it cannot see thereviewerpersona at all.Related Issue(s)
Related: AI-710
Breaking Change?
No. Inputs, outputs, and the pass/fail conditions are all unchanged — only the messages on failure, and the retry.
Checklist
README.mdand theaction.yamlinput/output descriptions if the action's interface changed — the interface did not changeLint YAMLpasses — yamllint against.yamllint, cleane2e/harness (or explained why that wasn't possible) — the e2e harness exercises the happy path, which is the one case that was already green. I pulled the verify step back out of the YAML and ran it in an amd64 ubuntu container against a real astro 1.45.0 and against stubs for each failure mode (the table above); same for the GitLab block. shellcheck is clean on both.