chad/tansweat
1
0
Fork 0
Code Issues Pull requests 2 Projects Releases Packages Wiki Activity Actions

feat(prs): diff breakdown for each PR, shown in the header PR pill #50

Merged
chad merged 3 commits from feat/pr-breakdown into main 2026-11-06 21:13:34 -05:00
Conversation 0 Commits 3 Files changed 10 +657 -44
glen commented 2026-11-06 21:13:20 -05:00
Collaborator
Copy link

Ports sweatshop's PR breakdown. Every PR a thread opens now records how its diff splits across Generated / Tests / Docs / Migrations / Config / Code. The header PR pill shows the totals, and opens a list of every PR plus a combined category table.

What changed

  1. lib/diff-breakdown.ts (new, pure, tested). Ported from sweatshop: a git diff --numstat -z parser (renames keep the new path, binaries count 0), first-match-wins categories, and mergeBreakdowns. Categories are tuned to tansweat:
    • Generated: pnpm-lock.yaml, *.lock, **/routeTree.gen.ts, docs/vendor/**, *.snap
    • Tests: *.test.*, *.spec.*, __tests__/, fixtures/, *.fixture.*
    • Docs: docs/**, *.md, *.mdx
    • Migrations: no patterns. tansweat's SQL migrations live inline in server/db.ts, so matching that file would count real code as migrations. A comment says so, and empty categories don't show.
    • Config: *.config.*, package.json, pnpm-workspace.yaml, tsconfig*.json, .forgejo/**, Containerfile*, patches/**
    • Code: everything else
  2. create_pr records it. After the PR opens, a separate deps.once('diff-breakdown') step runs git diff --numstat -z origin/<base>...HEAD. <base> was already fetched in an earlier step, and a replay returns the recorded value. It's best-effort: if numstat fails, the result has no breakdown and the PR still succeeds. The result contract in docs/architecture/git.md is updated.
  3. lib/prs.ts. ThreadPR gains an optional breakdown, and title taken from the call's input.
  4. Header pill. After the existing state label it shows N files +A −D. Clicking opens a popover: one row per PR (live state dot, owner/repo#n, title, +/−, link), then the category table with a Total. A lone PR with no stats still links straight out. This adds components/ui/popover.tsx, the shadcn popover on the already-pinned radix-ui, so no new dependency.

Verification

  • pnpm typecheck and pnpm test pass (1331 passed, up from 1295). The new tests cover the parser, the categories, merging, create_pr with and without a numstat failure, and prsFromMessages reading title and breakdown.
  • Real numstat:
    • I ran the parser on git diff --numstat -z for the actual #47, #48 and #49 diffs. The totals match git --shortstat, and #48's split matches Forgejo's own breakdown (Docs 1 +3, Code 27 +97 −61).
    • A probe repo with a rename, a binary file, a lockfile and a vendored doc parsed correctly. The rename kept its new path and was filed as Tests.
  • Sandbox output: the docker/podman sandbox decodes stdout by concatenating the raw bytes, which keeps the NUL separators. I checked that with the same decode on real git output. The git() helper's .trim() only strips the trailing NUL.
  • Browser: I rendered the pill with #47–#49's real stats plus an older PR without stats (screenshots below):
    • The pill showed 4 PRs and 43 files +642 −118.
    • The table read Tests 2 / Docs 3 / Code 38, with the older PR falling back to its branch name.

Limits

  • Only PRs opened after this ships get stats. Older ones keep their row and live state, without numbers.
  • Not yet run through a live create_pr. The first real PR will be the end-to-end check.
  • Breakdown on the approval card is a planned follow-up, not in this PR.

Screenshots

PR popover, light

PR popover, dark

Ports sweatshop's PR breakdown. Every PR a thread opens now records how its diff splits across Generated / Tests / Docs / Migrations / Config / Code. The header PR pill shows the totals, and opens a list of every PR plus a combined category table. ## What changed 1. **`lib/diff-breakdown.ts` (new, pure, tested).** Ported from sweatshop: a `git diff --numstat -z` parser (renames keep the new path, binaries count 0), first-match-wins categories, and `mergeBreakdowns`. Categories are tuned to tansweat: - **Generated:** `pnpm-lock.yaml`, `*.lock`, `**/routeTree.gen.ts`, `docs/vendor/**`, `*.snap` - **Tests:** `*.test.*`, `*.spec.*`, `__tests__/`, `fixtures/`, `*.fixture.*` - **Docs:** `docs/**`, `*.md`, `*.mdx` - **Migrations:** no patterns. tansweat's SQL migrations live inline in `server/db.ts`, so matching that file would count real code as migrations. A comment says so, and empty categories don't show. - **Config:** `*.config.*`, `package.json`, `pnpm-workspace.yaml`, `tsconfig*.json`, `.forgejo/**`, `Containerfile*`, `patches/**` - **Code:** everything else 2. **`create_pr` records it.** After the PR opens, a separate `deps.once('diff-breakdown')` step runs `git diff --numstat -z origin/<base>...HEAD`. `<base>` was already fetched in an earlier step, and a replay returns the recorded value. It's best-effort: if numstat fails, the result has no `breakdown` and the PR still succeeds. The result contract in `docs/architecture/git.md` is updated. 3. **`lib/prs.ts`.** `ThreadPR` gains an optional `breakdown`, and `title` taken from the call's input. 4. **Header pill.** After the existing state label it shows `N files +A −D`. Clicking opens a popover: one row per PR (live state dot, `owner/repo#n`, title, +/−, link), then the category table with a Total. A lone PR with no stats still links straight out. This adds `components/ui/popover.tsx`, the shadcn popover on the already-pinned `radix-ui`, so no new dependency. ## Verification - `pnpm typecheck` and `pnpm test` pass (1331 passed, up from 1295). The new tests cover the parser, the categories, merging, `create_pr` with and without a numstat failure, and `prsFromMessages` reading title and breakdown. - **Real numstat:** - I ran the parser on `git diff --numstat -z` for the actual #47, #48 and #49 diffs. The totals match `git --shortstat`, and #48's split matches Forgejo's own breakdown (Docs 1 +3, Code 27 +97 −61). - A probe repo with a rename, a binary file, a lockfile and a vendored doc parsed correctly. The rename kept its new path and was filed as Tests. - **Sandbox output:** the docker/podman sandbox decodes stdout by concatenating the raw bytes, which keeps the NUL separators. I checked that with the same decode on real git output. The `git()` helper's `.trim()` only strips the trailing NUL. - **Browser:** I rendered the pill with #47–#49's real stats plus an older PR without stats (screenshots below): - The pill showed 4 PRs and 43 files +642 −118. - The table read Tests 2 / Docs 3 / Code 38, with the older PR falling back to its branch name. ## Limits - Only PRs opened after this ships get stats. Older ones keep their row and live state, without numbers. - Not yet run through a live `create_pr`. The first real PR will be the end-to-end check. - Breakdown on the approval card is a planned follow-up, not in this PR. ## Screenshots ![PR popover, light](./pr-popover-light.png) ![PR popover, dark](./pr-popover-dark.png)
Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
feat(ui): PR pill popover with per-PR rows and diff breakdown table
All checks were successful
build / build-and-test (pull_request) Successful in 2m40s
Details
build / build-sandbox-image (pull_request) Successful in 1s
Details
build / deploy (pull_request) Has been skipped
Details
cb42fef1a6 
Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
chad merged commit b07776db07 into main 2026-11-06 21:13:34 -05:00
Sign in to join this conversation.
Reviewers
No reviewers
Labels
Clear labels
No items
No labels
Milestone
Clear milestone
No items
No milestone
Projects
Clear projects
No items
No project
Assignees
Clear assignees
No assignees
1 participant
Notifications
Due date
The due date is invalid or out of range. Please use the format "yyyy-mm-dd".

No due date set.

Dependencies

No dependencies set.

Reference
chad/tansweat!50
No description provided.