{"entries":[{"lastCommit":{"author":{"email":"claas@claas.dev","name":"Claas","when":"2026-08-06T15:56:11.000000+02:00"},"committer":{"email":"claas@claas.dev","name":"Claas","when":"2026-08-06T17:29:43.000000+02:00"},"hash":"b3f65e3f940555f46ddcd5d0ea909490c4f1bdfb","message":"Fill the sidebar with the repository's refs\n\nThe placeholder sections from the first change now show what is\nactually there. `list_references` already existed and was tested; this\nis the store, the grouping and the markup that use it.\n\nRefs are grouped by what their namespace says they are, and sorted with\na numeric collator so v9 comes before v10 rather than after it. Refs in\nthe `other` group — notes, refs/stash, replacements — are left out:\nthey are machinery rather than things a person navigates by, and\nshowing them would bury the branches.\n\nThe Stashes section is gone rather than left empty. Listing stashes\nmeans reading the reflog of refs/stash, not enumerating refs, so it\ncould only ever have shown whether a stash ref exists. A heading that\ncan never be right is worse than no heading.\n\nEntries are text, not buttons. Nothing happens when you click a branch\nyet, and a control that looks interactive but is not is worse than\nplain text. They become controls when checking out exists.\n\nThe checked-out branch is marked in words as well as colour, since\ncolour alone says nothing out loud. Before a repository is open the\nsections read \"—\" rather than \"None\", because \"None\" would be a claim\nabout a repository nothing has read. Each section carries a count,\nwhich earns its place on a repository with hundreds of remote branches.\n\nLoading refs is not awaited before the graph starts: the two are\nindependent, and the sidebar should not hold up the first commits. A\nfailure in one leaves the other alone.\n\n21 tests, including a repository with 400 refs, a remote branch whose\nname ends with the checked-out branch's name, and that grouping does\nnot mutate what it was handed.\n\nCo-Authored-By: Claude Opus 5 \n","tree":"e5811c0ddfb29b33f573e63e317864fe9b67212f"},"mode":"0100644","name":"App.tsx","size":4770},{"lastCommit":{"author":{"email":"claas@claas.dev","name":"Claas","when":"2026-08-06T15:39:56.000000+02:00"},"committer":{"email":"claas@claas.dev","name":"Claas","when":"2026-08-06T17:29:42.000000+02:00"},"hash":"f24bc1084fdab7735009fdaa5ff5858fad2f12b9","message":"Pick repositories with a folder dialog and a recents list\n\nReplaces the placeholder path field with the native folder picker and a\nwelcome pane that offers what was opened before.\n\nThe picker is wrapped in one module so nothing else imports the dialog\nplugin, which is also what lets the tests stand in for it in one place.\nDismissing the dialog resolves to undefined and is treated as an\nordinary outcome — cancelling is not a failure to report.\n\nA recent entry opens by its stored path rather than its name, and stays\nopenable when the folder is missing: a drive may simply not be mounted,\nand refusing to try would be presumptuous. It is marked so the reason a\nclick might fail is visible before it does. Forgetting removes the entry\nstraight away and puts it back if the call fails — waiting on a round\ntrip to see your own click take effect feels broken.\n\nA row is two controls side by side rather than one containing another,\nwhich HTML does not allow. Every forget button reads \"Forget\" on screen,\nso each carries the repository name as hidden text; otherwise they would\nbe indistinguishable by voice.\n\nTwo things the tests pushed on:\n\n- The recents list is named after its heading, so it can be addressed as\n the recent list rather than as \"a list\" — the ref sidebar is a list\n too, and an unscoped query counts four extra rows. Naming a

would invent a region landmark.\n- The list-semantics workaround moved from the history list to a base\n rule for every ol and ul. Two lists needed it, and a third would have\n been a bug waiting to happen.\n\nVerified in WebKit and Chromium, light and dark: 15 unit tests, 10\nend-to-end, and the whole suite still green at 56 across both engines.\n\nCo-Authored-By: Claude Opus 5 \n","tree":"74125a833df45dcdc3d32c4b13b894235958add4"},"mode":"0100644","name":"app.css","size":2959},{"lastCommit":{"author":{"email":"claas@claas.dev","name":"Claas","when":"2026-08-06T16:24:11.000000+02:00"},"committer":{"email":"claas@claas.dev","name":"Claas","when":"2026-08-06T17:29:43.000000+02:00"},"hash":"a177e49aaea39aeca74c8fac8a60cdcaba082f54","message":"Filter the sidebar refs\n\nFour hundred branches is a scroll; three is a list. A search field at\nthe top of the sidebar narrows branches, remotes and tags at once,\nwhich is the point — a filter that only applied to one section would\nmean knowing which section a name is in before you can look for it.\n\nMatching is a plain case-insensitive substring on the shorthand, not\nanything fuzzy. It is predictable, and a person can see why a result\nappeared without learning the rules. The shorthand is what the sidebar\nshows, so it is what gets matched; searching text that is not on screen\nwould be a surprise.\n\nThe matched run of each name is wrapped in , which is exactly\nwhat that element is for: text made relevant by something the reader is\ndoing right now. Splitting is done by scanning rather than with a\nregular expression — a branch name can contain `.` or `+`, and a lone\n`*` would otherwise throw rather than simply matching nothing. There is\na test for each.\n\nCounts read \"1 of 2\" while filtering, and an — a polite live\nregion — says how many matched altogether, so the result of typing is\nspoken without focus leaving the field.\n\nA section with nothing left in it disappears rather than saying \"None\",\nwhich would read as \"this repository has no tags\" when the truth is \"no\ntags match\". When nothing matches anywhere, one message says so instead\nof three sections each saying it separately. The e2e tests caught that\ninconsistency: the whole-sidebar case was handled and the per-section\ncase was not.\n\nEscape clears the field. Whether a search input does that itself is\nleft to the browser, and WebKit and Chromium disagree, so it is handled\nexplicitly and tested in both.\n\nThe filter resets when another repository opens. One written for one\nrepository rarely means anything in the next, and leaving it on would\nmake a full sidebar look empty.\n\nThe nav's existing tests now drive the real store rather than a\nhand-written stand-in, which is why they needed changing: the stub had\ngone stale against the store it was imitating. Using the real one means\nthey cannot drift again.\n\nCo-Authored-By: Claude Opus 5 \n","tree":"a6e2e68e816635014c94e34c541cf67d04359049"},"mode":"0040000","name":"components","size":0},{"lastCommit":{"author":{"email":"claas@claas.dev","name":"Claas","when":"2026-08-06T01:05:49.000000+02:00"},"committer":{"email":"claas@claas.dev","name":"Claas","when":"2026-08-06T17:29:37.000000+02:00"},"hash":"39172c3e042f42852a348f49638697401ac6001f","message":"Set up the frontend toolchain and app shell\n\nReplaces the create-tauri-app demo with the toolchain the project will\nactually use, and a shell to hang the real UI off.\n\n- Solid 2.0 beta. The JSX runtime moved to @solidjs/web, so jsxImportSource\n and the render import point there now. @solidjs/signals is a direct\n dependency so resolve.dedupe can actually resolve it under pnpm —\n more than one copy of the Solid packages causes reactivity bugs that\n are painful to trace.\n- TypeScript 7, the native compiler. Type-checking only; Vite transpiles.\n- Tailwind v4 with CSS-first config, plus @claas.dev/material-tailwind to\n generate the Material 3 colour system from a single seed. The plugin\n emits every colour as light-dark(), so light and dark themes work with\n no dark: variants at all — only a color-scheme declaration.\n- oxlint and oxfmt instead of ESLint and Prettier, with Tailwind class\n sorting pointed at src/app.css. Verified it resolves plugin-generated\n utilities, not just core Tailwind ones. jsx-a11y is enabled and works\n on Solid's JSX; react is deliberately off. Vendored skill files and\n Tauri's generated JSON are excluded from formatting.\n\nThe shell is landmarks and headings only: banner, a labelled navigation\nsidebar, main, and a contentinfo status bar with an live region.\nA skip link targets main, which takes tabindex=\"-1\" so focus actually\nmoves rather than only scrolling.\n\nThe scaffold's greet command in src-tauri is left alone; the repository\nactor change replaces it.\n\nCo-Authored-By: Claude Opus 5 \n","tree":"2a0be45da83e293c31775c2ec079aaac96764aed"},"mode":"0100644","name":"index.tsx","size":262},{"lastCommit":{"author":{"email":"claas@claas.dev","name":"Claas","when":"2026-08-06T16:24:11.000000+02:00"},"committer":{"email":"claas@claas.dev","name":"Claas","when":"2026-08-06T17:29:43.000000+02:00"},"hash":"a177e49aaea39aeca74c8fac8a60cdcaba082f54","message":"Filter the sidebar refs\n\nFour hundred branches is a scroll; three is a list. A search field at\nthe top of the sidebar narrows branches, remotes and tags at once,\nwhich is the point — a filter that only applied to one section would\nmean knowing which section a name is in before you can look for it.\n\nMatching is a plain case-insensitive substring on the shorthand, not\nanything fuzzy. It is predictable, and a person can see why a result\nappeared without learning the rules. The shorthand is what the sidebar\nshows, so it is what gets matched; searching text that is not on screen\nwould be a surprise.\n\nThe matched run of each name is wrapped in , which is exactly\nwhat that element is for: text made relevant by something the reader is\ndoing right now. Splitting is done by scanning rather than with a\nregular expression — a branch name can contain `.` or `+`, and a lone\n`*` would otherwise throw rather than simply matching nothing. There is\na test for each.\n\nCounts read \"1 of 2\" while filtering, and an — a polite live\nregion — says how many matched altogether, so the result of typing is\nspoken without focus leaving the field.\n\nA section with nothing left in it disappears rather than saying \"None\",\nwhich would read as \"this repository has no tags\" when the truth is \"no\ntags match\". When nothing matches anywhere, one message says so instead\nof three sections each saying it separately. The e2e tests caught that\ninconsistency: the whole-sidebar case was handled and the per-section\ncase was not.\n\nEscape clears the field. Whether a search input does that itself is\nleft to the browser, and WebKit and Chromium disagree, so it is handled\nexplicitly and tested in both.\n\nThe filter resets when another repository opens. One written for one\nrepository rarely means anything in the next, and leaving it on would\nmake a full sidebar look empty.\n\nThe nav's existing tests now drive the real store rather than a\nhand-written stand-in, which is why they needed changing: the stub had\ngone stale against the store it was imitating. Using the real one means\nthey cannot drift again.\n\nCo-Authored-By: Claude Opus 5 \n","tree":"a6e2e68e816635014c94e34c541cf67d04359049"},"mode":"0040000","name":"lib","size":0},{"lastCommit":{"author":{"email":"claas@claas.dev","name":"Claas","when":"2026-08-05T23:11:46.000000+02:00"},"committer":{"email":"claas@claas.dev","name":"Claas","when":"2026-08-05T23:11:46.000000+02:00"},"hash":"30d6fc6273d153ec656cd3d6673037d1e74ee756","message":"Scaffold tauri project\n","tree":"57f5c1afe32dc29a10085f8742a5a8205e0b9d37"},"mode":"0100644","name":"vite-env.d.ts","size":38}],"lastCommit":{"author":{"email":"claas@claas.dev","name":"Claas","when":"2026-08-06T16:24:11.000000+02:00"},"committer":{"email":"claas@claas.dev","name":"Claas","when":"2026-08-06T17:29:43.000000+02:00"},"hash":"a177e49aaea39aeca74c8fac8a60cdcaba082f54","message":"Filter the sidebar refs\n\nFour hundred branches is a scroll; three is a list. A search field at\nthe top of the sidebar narrows branches, remotes and tags at once,\nwhich is the point — a filter that only applied to one section would\nmean knowing which section a name is in before you can look for it.\n\nMatching is a plain case-insensitive substring on the shorthand, not\nanything fuzzy. It is predictable, and a person can see why a result\nappeared without learning the rules. The shorthand is what the sidebar\nshows, so it is what gets matched; searching text that is not on screen\nwould be a surprise.\n\nThe matched run of each name is wrapped in , which is exactly\nwhat that element is for: text made relevant by something the reader is\ndoing right now. Splitting is done by scanning rather than with a\nregular expression — a branch name can contain `.` or `+`, and a lone\n`*` would otherwise throw rather than simply matching nothing. There is\na test for each.\n\nCounts read \"1 of 2\" while filtering, and an — a polite live\nregion — says how many matched altogether, so the result of typing is\nspoken without focus leaving the field.\n\nA section with nothing left in it disappears rather than saying \"None\",\nwhich would read as \"this repository has no tags\" when the truth is \"no\ntags match\". When nothing matches anywhere, one message says so instead\nof three sections each saying it separately. The e2e tests caught that\ninconsistency: the whole-sidebar case was handled and the per-section\ncase was not.\n\nEscape clears the field. Whether a search input does that itself is\nleft to the browser, and WebKit and Chromium disagree, so it is handled\nexplicitly and tested in both.\n\nThe filter resets when another repository opens. One written for one\nrepository rarely means anything in the next, and leaving it on would\nmake a full sidebar look empty.\n\nThe nav's existing tests now drive the real store rather than a\nhand-written stand-in, which is why they needed changing: the stub had\ngone stale against the store it was imitating. Using the real one means\nthey cannot drift again.\n\nCo-Authored-By: Claude Opus 5 \n","tree":"a6e2e68e816635014c94e34c541cf67d04359049"}}