🙈 chore(build): untrack dist/, keep publishing it

[vtempest]

Hides dist/ from git while keeping it in every npm tarball.

The problem

dist was already listed in .gitignore, but 130 build artifacts were tracked anyway — .gitignore has no effect on files git already follows. Worse, the committed bundle had gone stale: it predated a package reshuffle, so its .d.ts tree (dist/grab-api/*.d.ts) no longer matched the current source layout (dist/grab-api/src/*.d.ts). That stale bundle was masking the fact that npm run build has been failing on master.

What changed

Untracked (files stay on disk; only git stops following them):

Path	Why
dist/	87 root build artifacts
packages/archiver-web/dist/	35 stale copies of the root build — contains grab-api, grab-url-cli, log, icons, nothing of archiver-web's
packages/quantum-sphere-loading-animation/dist/	8 tsup outputs

Kept tracked: packages/native-app-wrapper/dist/index.html. It is not build output — it is the hand-written UI Tauri serves as frontendDist (src-tauri/tauri.conf.json), and its own comment says "no build step and no npm dependency". .gitignore re-includes that one directory.

Build fix (separate commit, required to make the above safe): packages/grab-url-cli/src/index.ts carried a duplicate export { … } from './transfer/ytdlp-transfer.js' block whose names were either already exported a few lines above or never existed at all. page/archive-page.ts imported that same never-written pair, so the --page video step could not have run. Implemented probeYtDlp(url) and runYtDlpDownload(url, opts) in ytdlp-transfer.ts under the names archive-page.ts and test/page-archive.test.ts already expect.

Why publishing still works

A files entry cannot be excluded by .gitignore or .npmignore — so npm pack still carries dist, and prepublishOnly rebuilds it beforehand. Also adds the missing prepublishOnly to quantum-sphere-loading-icon so a manual npm publish cannot ship without one.

Verification

• rm -rf dist && npm run build → rebuilds clean; all 18 paths named by package.json's main/types/module/bin/exports resolve.
• npm pack --dry-run → 91 files, identical to before (125 dist entries after the rebuild, up from the stale 87).
• git status clean after a build — no dist noise.
• npx vitest run → 343 passed / 11 failed, byte-identical to the master baseline (verified via a git worktree of HEAD). Those 11 failures are pre-existing and untouched here.

Left for a follow-up

master is red on npm test independently of this PR: test/page-archive.test.ts is written against a different yt-dlp module design than the one on disk — it expects parseYtDlpSize, parseYtDlpEta, a percent-bearing parseYtDlpProgress that parses yt-dlp's human console line, and buildYtDlpArgs({ filename }), whereas the module implements a --progress-template sentinel design with output. Reconciling those two is a design decision, not a build fix, so it is deliberately out of scope.

Worth noting: CI (tests.yml) runs only npm run test:coverage, never npm run build — which is how a broken build reached master unnoticed.

🤖 Generated with Claude Code

https://claude.ai/code/session_01R3wkKQsRoi6k3DTvBRQhHd

---

Generated by Claude Code

[vercel]

The latest updates on your projects. Learn more about Vercel for GitHub.

Project	Deployment	Actions	Updated
grab-url	Error		Sep 11, 2026 2:22pm UTC

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.

📢 Thoughts on this report? Let us know!

[vtempest]

CI: Tests is red here, and it is red on master too

Not standing down quietly — here is the evidence and a proposed patch.

The check is failing on the base branch independently of this PR. Tests run #175 on master (18201a0, the merge of #50) is a failure, as is #173 on the merge of #49. I also ran the suite locally against a git worktree of the base commit and against this branch:

	Tests
master baseline	343 passed / 11 failed
this branch	343 passed / 11 failed

Same 11, same names. This PR neither causes nor worsens them.

What the 11 failures actually are. test/page-archive.test.ts is written against a different generation of packages/grab-url-cli/src/transfer/ytdlp-transfer.ts than the one on disk — the --page feature (#49) landed with tests for one design while the module implements another:

The test expects	The module implements
parseYtDlpSize('1.00KiB') → 1024	(does not exist)
parseYtDlpEta('01:02:03') → 3723	(does not exist)
parseYtDlpProgress() parses yt-dlp's human console line ([download] 23.4% of ~12.00MiB at 1.00MiB/s ETA 00:42) and returns a percent	parses the @GRAB@ --progress-template sentinel and returns status / estimated / fragmentIndex
buildYtDlpArgs(url, { filename }) → 'My Video.%(ext)s'	takes output, used verbatim
describeYtDlpExit(1) === 'download failed'	'download failed — the media may be private, region-locked or removed'

Why I am not fixing it here. Nine of the eleven are additive and safe (parseYtDlpSize, parseYtDlpEta, accepting filename alongside output, shortening the exit string). The remaining two are not: parseYtDlpProgress cannot satisfy both designs at once, and the sentinel version is the one runYtDlpTransfer feeds its live progress bar from. Fixing nine leaves CI red anyway, so it buys nothing and widens a .gitignore PR into a rewrite of the downloader's progress path. Which design is canonical is a call for the repo owner, not something to decide inside this PR.

Proposed patch, for whoever picks it up. Keep the sentinel design — it is what production uses, and it is strictly more informative. Then:

1. Add parseYtDlpSize / parseYtDlpEta as pure helpers (the tests fully specify them).
2. Accept filename in YtDlpOptions as filename + '.%(ext)s', keeping output for the verbatim case.
3. Update the parseYtDlpProgress tests to the sentinel format rather than the console format — the console-parsing API they describe is not reachable from any caller.
4. Decide on the describeYtDlpExit(1) wording and align test and module.

This is not a flake, so I have not spent a re-run on it: a re-run would fail identically.

One thing this PR does fix: npm run build was also broken on master (duplicate and non-existent re-exports in grab-url-cli/src/index.ts, plus archive-page.ts importing a runYtDlpDownload that was never written). That is repaired in 596d7b1 and the build is green. It went unnoticed because tests.yml runs only npm run test:coverage and never npm run build — worth adding a build step to CI in a follow-up.

Vercel is a separate known issue: the project's Root Directory is still docs, a folder #47 deleted, so every deployment fails at clone. That is a dashboard setting, already documented in .claude/architecture/documentation.md by #51 — nothing in this diff affects it.

---

Generated by Claude Code