feat: dist-diff — show reviewers what a PR changes in the built artifacts - #332
Draft
Peetee06 wants to merge 5 commits into
Draft
feat: dist-diff — show reviewers what a PR changes in the built artifacts#332Peetee06 wants to merge 5 commits into
Peetee06 wants to merge 5 commits into
Conversation
A PR's source diff doesn't show its effect on the built artifacts: an 8-line config change can add a wizard cliEntry, a marketplace plugin, and a mirror entry, while a naive diff of two builds reports 221 changed files of which 216 are noise (zip entry mtimes, manifest buildTimestamp). dist-diff compares two normalized dist/ trees and reports the true artifact delta per consumer surface, as a sticky PR comment (posted by a minimal privileged workflow_run job so fork PRs get identical treatment) plus a full report with content hunks in the step summary. A twice-build self-check guards the normalization on pipeline-touching PRs and main. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01GDjuSAs927XvcyziFJk865
- Root-level files no longer fold into a nameless directory bucket (the empty-dir key was a string-prefix of every same-kind key) - Budget truncation now trims inside diff blocks, so surface headers and the "✓ unchanged" assertions always survive; removes the old tail-cut that could clobber a line and leave an open fence - Content-hunk size gate moved to decompressed inner-entry size — a tiny archive holding a huge entry no longer reaches the differ unchecked - Full report diffs added/removed files against the empty side - Corrupt-zip failures now name the offending file (still fail loudly: a zip our own build wrote but can't read back is a broken build) - Dedupe delta-key and inner-suffix construction; fix stale --self-check reference in the module header Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01GDjuSAs927XvcyziFJk865
An identical-delta fan-out is one content change that happens to land in many archives — so the full report now leads with what changed (the shared inner file) and the hunk itself, and collapses the member list into <details>. Replaces the per-surface tree + hunks-in-details layout where 68 zip lines repeated the same inner path. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01GDjuSAs927XvcyziFJk865
skills-mcp-resources.zip is derived (manifest + every skill zip), so its diff repeats what the report already shows. Cross-check each inner change against the reported constituent changes: consistent collapses to one verified summary line; an unexplained inner change — a broken bundling step — goes loud with a warning in both reports. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01GDjuSAs927XvcyziFJk865
… block Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01GDjuSAs927XvcyziFJk865
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
A PR's source diff doesn't show its impact on this repo's actual product — the generated
dist/tree the wizard, MCP server, and plugin marketplace consume. Reviewers reconstruct that mapping in their heads.Measured on #330 (8 added YAML lines): the real effect was a new wizard
cliEntry, a new marketplace plugin, and a mirror entry — while a naivediff -rof the two builds reports 221 changed files, 216 of them noise (zip entry mtimes, manifestbuildTimestamp).What
npm run diff+ two workflows that post a sticky PR comment with the normalized artifact delta, grouped by consumer surface. This is #330's comment, generated from its real build outputs:The workflow step summary carries the full report with content-level hunks (one per unique delta, so a shared-doc change fanning into 68 variant zips shows its hunk once).
Design invariants
dist/is compared; unknown future artifact types degrade to byte comparison — noise at worst, never silently skipped.scripts/,package.json, the lockfile, or the workflow — content PRs can never go red here. On failure the report still posts (warning-stamped) before the job fails..docs-cache, so upstream doc drift can't appear in the diff.contents: readonly; a minimalworkflow_runjob posts the sticky comment, verifying the artifact's PR number against the run's head SHA.fflate(same zip library the MCP server uses on these archives) anddiff(jsdiff), both dev-tier.Verification
A follow-up issue will propose a GitHub-native alternative for the full report (push normalized base/head trees to a throwaway ref and let GitHub's commit view render the diff) — it would delete the hunk-rendering code here but needs a maintainer call on
contents: writeand ref hygiene.🤖 Generated with Claude Code
https://claude.ai/code/session_01GDjuSAs927XvcyziFJk865