Fix evidence-regression workflow: pin past v0, add 'pr' field to plan#579
Merged
Merged
Conversation
…r' to plan The first attempt to dispatch evidence-regression.yml against the cyclomatic pilot failed with "Missing required option --repo." after the action invoked the evidence CLI as bare `evidence capture-pr`. Root cause: the `v0` tag in RiddimSoftware/evidence was created before EVI-7 (#39) wired the capture-pr argument translation into action.yml. At @v0, the action's case statement has no `capture-pr)` branch, so it runs the CLI without `--repo`, `--pr`, `--plan`, `--output`, etc. Pinning to a SHA past EVI-7 (fc65484f400ec8bc0ded78a70fb54ed32f7dfa6d, the head of evidence/main as of this commit) restores the wiring. This is a temporary fix — the right long-term move is to fast-forward the `v0` tag in RiddimSoftware/evidence, after which we can move back to `@v0` for cross-repo consistency. The second issue was the plan itself: regression-parliament-calendar.json omitted the `pr` field, which the CLI requires even when --pr is passed as an override. Added `pr: 542` as the conventional placeholder (the cyclomatic-project trigger PR). The CLI's --pr flag and the workflow's pr_number input still override at dispatch time, so this value only affects metadata in manifest.json when no override is supplied. Local capture-pr against project-boundary SHAs (9ba487c → 79e3b0b) is currently running to validate the plan end-to-end before the next CI dispatch.
sunnypurewal
added a commit
that referenced
this pull request
May 25, 2026
## Why [PR #579](#579) pinned `RiddimSoftware/evidence@fc65484` to get past EVI-7's capture-pr argument wiring. That pin is **still missing EVI-11 (#41)**, which fixed a pipe-buffer deadlock in `CommandRunner.swift`. Diagnosed during local-first validation of the cyclomatic-complexity pilot — the pre-EVI-11 CLI hung at 12+ minutes post-compile on local runs. ## What the bug was The pre-EVI-11 `ProcessCommandRunner.run(_:_:workingDirectory:environment:)` created pipes for the child process's stdout/stderr, then called `process.waitUntilExit()`, then read the pipes. macOS pipe buffer is 16-64 KB; `xcodebuild` emits megabytes during a build. The buffer fills mid-build, xcodebuild's writer thread blocks waiting for a drain, no one drains until after the wait — **classic textbook deadlock**. `.app` files appear on disk (the build steps themselves succeed) but xcodebuild can never exit. EVI-11 fixed it with `readabilityHandler`-based concurrent draining. Verified locally: pre-fix binary hung at 12+ minutes on a build that the post-fix binary completes in ~1 minute. CI happens to mask this because the action's `${cli} ${args[@]}` invocation has a downstream shell pipe that drains the stdout transitively. Local invocations (no downstream pipe) hit the deadlock directly. ## What changed Bump `RiddimSoftware/evidence@fc65484f400ec8bc0ded78a70fb54ed32f7dfa6d` → `RiddimSoftware/evidence@e76e7991b7eda56857080e46fd84ec6ebb642161` (current `evidence/main`, post-EVI-11). The inline workflow comment is updated to explain the second pin requirement so future maintainers know why this isn't `@v0`. ## Trade-offs not taken - **Did not move the `v0` tag in `RiddimSoftware/evidence`.** That's the right long-term fix (so we can return to `@v0`), but it would silently affect every consumer. Pinning here is reversible and contained. - **Did not change the local CLI invocation.** Local runs were also affected, but the fix lives in the upstream evidence repo, not in any consuming workflow. ## Test plan - `actionlint .github/workflows/evidence-regression.yml` — clean. - Local CLI run with a binary rebuilt from `e76e7991b7eda56857080e46fd84ec6ebb642161` of `evidence` — used the same pipeline to validate the fix worked (the pipeline now produces evidence end-to-end where the prior CLI hung). - After this merges, the canonical dispatch is unchanged: ``` gh workflow run evidence-regression.yml -R RiddimSoftware/epac --ref main \ -f plan_path=.evidence/regression-parliament-calendar.json \ -f pr_number=542 \ -f before_sha=9ba487ca2fdb2ecfa4fe722d0c353944284bc9ff \ -f after_sha=79e3b0bc2904dcc69e53975142deb4283f5f8c6d \ -f linear_issue_id=EPAC-1997 \ -f output_subdir=pilot-cyclomatic-complexity ``` Reviewer-Boundary: review-only --------- Co-authored-by: Sunny Purewal <sunny@riddimsoftware.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
The first dispatch of
evidence-regression.ymlfailed in ~1 second withMissing required option --repo.and a separate local CLI attempt failed withInvalid PR change evidence plan: missing required field 'pr'. Two distinct bugs:RiddimSoftware/evidence@v0— thev0tag was created inRiddimSoftware/evidencebefore EVI-7 (#39) wiredcapture-prargument translation intoaction.yml. At@v0, thecase "${subcommand}"block has nocapture-pr)branch, so the action invokes the CLI as bareevidence capture-prwith no flags. The 30-minute build never happens — the action fails at CLI argument parsing.regression-parliament-calendar.jsonwas missing theprfield — the CLI's plan loader validatespras required even when the--prflag is passed as override. Local dispatch hit this gate before any build.What changed
.github/workflows/evidence-regression.yml— pin toRiddimSoftware/evidence@fc65484f400ec8bc0ded78a70fb54ed32f7dfa6d, the current HEAD ofevidence/main, which is past EVI-7. Inline comment notes the temporary nature: the right long-term fix is to fast-forward thev0tag inRiddimSoftware/evidenceso all consumers get the fix, after which we can return to@v0..evidence/regression-parliament-calendar.json— add"pr": 542as the conventional placeholder (the cyclomatic-project's trigger PR, EPAC-1985). The CLI's--prflag and the workflow'spr_numberinput still override at dispatch time, so this only sets metadata inmanifest.jsonwhen no override is supplied.Trade-offs not taken
v0tag inRiddimSoftware/evidence. That would silently affect every consumer. Pinning my workflow to a specific SHA is reversible and contained.swift teststep that parses every.evidence/*.jsonthrough the plan loader.Test plan
actionlint .github/workflows/evidence-regression.yml— clean.evidence capture-pr ... --plan .evidence/regression-parliament-calendar.json ...against the project boundary SHAs (9ba487ca→79e3b0bc) — the CLI accepted the plan and started both builds (the local Mac hit an unrelatedxcodebuildpost-build network stall that does not reproduce on macos-14 runners; documented in the session, not blocking this PR).-f pr_number=542is now required (the action's pre-check at any post-EVI-7 SHA requires it; prior attempt omitted it).Reviewer-Boundary: review-only