{"entries":[{"lastCommit":{"author":{"email":"claas@claas.dev","name":"Claas","when":"2026-08-06T09:09:35.000000+02:00"},"committer":{"email":"claas@claas.dev","name":"Claas","when":"2026-08-06T17:29:41.000000+02:00"},"hash":"316a0f76438414e6321d2debd9f60d4410436dbc","message":"Draw the commit graph\n\nThe history is an ordinary list of buttons. Each row owns a small SVG\nfor its lane segment, marked aria-hidden, over content that already\nreads correctly without it — a summary, an author, a , and a\nshort id. Row is a button because selecting a commit is an action and a\nbutton brings its own role, focus ring and Enter handling; that limits\nit to phrasing content, which is why ref badges are spans with a\nvisually hidden word for what each one is.\n\nTen lane colours come from one CSS rule rather than ten values: the\nlane index rotates an oklch hue, starting near the seed colour so the\ngraph belongs to the same palette as everything else. Nothing about\ncolour reaches the Rust side, and switching light to dark re-themes the\nwhole graph without a re-render.\n\nNothing here needs a virtual list. Rows arrive in batches as the end of\nthe list nears the viewport, so the DOM only holds what has been\nscrolled to, and rows past the first forty get content-visibility so\nthe browser skips laying out and painting the rest. `none` for the\nintrinsic width keeps an off-screen row from collapsing horizontally,\nwhich would make the scroll width jump on a wide graph. Which rows are\ndeferred is decided by nth-child, so no JavaScript is involved.\n\nWide graphs scroll rather than compress. A graph's width comes from the\nhistory — aspnetcore lays out 115 columns — so no window size makes it\nfit and capping would only hide branches.\n\nTwo things caught while driving the real components in a browser:\n\n- Five arrow presses moved the selection one row. Solid 2 batches\n signal writes, so reading selected() to compute the next value goes\n stale when presses land in one batch. Now the update goes through the\n functional setter, which sees the pending value.\n- jsx-a11y assumes React's htmlFor, so Solid's plain `for` read as an\n unassociated label. Configured rather than worked around.\n\nSafari drops list semantics from a list styled with list-style: none,\nwhich Tailwind's preflight applies — and Tauri is WKWebView on macOS,\nso that is the default here. An empty-string marker renders nothing and\nleaves the list a list, which beats patching the role back with ARIA.\n\nOpening is a path field for now. A native folder picker needs the Tauri\ndialog plugin, which is a dependency this change does not need; the\npicker arrives with the repository-picker work.\n\nInfinite scrolling is not verified yet — IntersectionObserver does not\nrun in a hidden document, and the browser pane available here cannot\nmake one visible. The Playwright harness covers it next.\n\nCo-Authored-By: Claude Opus 5 \n","tree":"2b0493859691be843589c8b062e5e045085099f5"},"mode":"0100644","name":"CommitRow.tsx","size":3151},{"lastCommit":{"author":{"email":"claas@claas.dev","name":"Claas","when":"2026-08-06T09:09:35.000000+02:00"},"committer":{"email":"claas@claas.dev","name":"Claas","when":"2026-08-06T17:29:41.000000+02:00"},"hash":"316a0f76438414e6321d2debd9f60d4410436dbc","message":"Draw the commit graph\n\nThe history is an ordinary list of buttons. Each row owns a small SVG\nfor its lane segment, marked aria-hidden, over content that already\nreads correctly without it — a summary, an author, a , and a\nshort id. Row is a button because selecting a commit is an action and a\nbutton brings its own role, focus ring and Enter handling; that limits\nit to phrasing content, which is why ref badges are spans with a\nvisually hidden word for what each one is.\n\nTen lane colours come from one CSS rule rather than ten values: the\nlane index rotates an oklch hue, starting near the seed colour so the\ngraph belongs to the same palette as everything else. Nothing about\ncolour reaches the Rust side, and switching light to dark re-themes the\nwhole graph without a re-render.\n\nNothing here needs a virtual list. Rows arrive in batches as the end of\nthe list nears the viewport, so the DOM only holds what has been\nscrolled to, and rows past the first forty get content-visibility so\nthe browser skips laying out and painting the rest. `none` for the\nintrinsic width keeps an off-screen row from collapsing horizontally,\nwhich would make the scroll width jump on a wide graph. Which rows are\ndeferred is decided by nth-child, so no JavaScript is involved.\n\nWide graphs scroll rather than compress. A graph's width comes from the\nhistory — aspnetcore lays out 115 columns — so no window size makes it\nfit and capping would only hide branches.\n\nTwo things caught while driving the real components in a browser:\n\n- Five arrow presses moved the selection one row. Solid 2 batches\n signal writes, so reading selected() to compute the next value goes\n stale when presses land in one batch. Now the update goes through the\n functional setter, which sees the pending value.\n- jsx-a11y assumes React's htmlFor, so Solid's plain `for` read as an\n unassociated label. Configured rather than worked around.\n\nSafari drops list semantics from a list styled with list-style: none,\nwhich Tailwind's preflight applies — and Tauri is WKWebView on macOS,\nso that is the default here. An empty-string marker renders nothing and\nleaves the list a list, which beats patching the role back with ARIA.\n\nOpening is a path field for now. A native folder picker needs the Tauri\ndialog plugin, which is a dependency this change does not need; the\npicker arrives with the repository-picker work.\n\nInfinite scrolling is not verified yet — IntersectionObserver does not\nrun in a hidden document, and the browser pane available here cannot\nmake one visible. The Playwright harness covers it next.\n\nCo-Authored-By: Claude Opus 5 \n","tree":"2b0493859691be843589c8b062e5e045085099f5"},"mode":"0100644","name":"GraphLanes.tsx","size":2595},{"lastCommit":{"author":{"email":"claas@claas.dev","name":"Claas","when":"2026-08-06T09:09:35.000000+02:00"},"committer":{"email":"claas@claas.dev","name":"Claas","when":"2026-08-06T17:29:41.000000+02:00"},"hash":"316a0f76438414e6321d2debd9f60d4410436dbc","message":"Draw the commit graph\n\nThe history is an ordinary list of buttons. Each row owns a small SVG\nfor its lane segment, marked aria-hidden, over content that already\nreads correctly without it — a summary, an author, a , and a\nshort id. Row is a button because selecting a commit is an action and a\nbutton brings its own role, focus ring and Enter handling; that limits\nit to phrasing content, which is why ref badges are spans with a\nvisually hidden word for what each one is.\n\nTen lane colours come from one CSS rule rather than ten values: the\nlane index rotates an oklch hue, starting near the seed colour so the\ngraph belongs to the same palette as everything else. Nothing about\ncolour reaches the Rust side, and switching light to dark re-themes the\nwhole graph without a re-render.\n\nNothing here needs a virtual list. Rows arrive in batches as the end of\nthe list nears the viewport, so the DOM only holds what has been\nscrolled to, and rows past the first forty get content-visibility so\nthe browser skips laying out and painting the rest. `none` for the\nintrinsic width keeps an off-screen row from collapsing horizontally,\nwhich would make the scroll width jump on a wide graph. Which rows are\ndeferred is decided by nth-child, so no JavaScript is involved.\n\nWide graphs scroll rather than compress. A graph's width comes from the\nhistory — aspnetcore lays out 115 columns — so no window size makes it\nfit and capping would only hide branches.\n\nTwo things caught while driving the real components in a browser:\n\n- Five arrow presses moved the selection one row. Solid 2 batches\n signal writes, so reading selected() to compute the next value goes\n stale when presses land in one batch. Now the update goes through the\n functional setter, which sees the pending value.\n- jsx-a11y assumes React's htmlFor, so Solid's plain `for` read as an\n unassociated label. Configured rather than worked around.\n\nSafari drops list semantics from a list styled with list-style: none,\nwhich Tailwind's preflight applies — and Tauri is WKWebView on macOS,\nso that is the default here. An empty-string marker renders nothing and\nleaves the list a list, which beats patching the role back with ARIA.\n\nOpening is a path field for now. A native folder picker needs the Tauri\ndialog plugin, which is a dependency this change does not need; the\npicker arrives with the repository-picker work.\n\nInfinite scrolling is not verified yet — IntersectionObserver does not\nrun in a hidden document, and the browser pane available here cannot\nmake one visible. The Playwright harness covers it next.\n\nCo-Authored-By: Claude Opus 5 \n","tree":"2b0493859691be843589c8b062e5e045085099f5"},"mode":"0100644","name":"History.tsx","size":4352},{"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":"RecentRepositories.tsx","size":3077},{"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":"0100644","name":"RefFilter.tsx","size":1636},{"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":"0100644","name":"RepositoryNav.tsx","size":4948},{"lastCommit":{"author":{"email":"claas@claas.dev","name":"Claas","when":"2026-08-06T09:09:35.000000+02:00"},"committer":{"email":"claas@claas.dev","name":"Claas","when":"2026-08-06T17:29:41.000000+02:00"},"hash":"316a0f76438414e6321d2debd9f60d4410436dbc","message":"Draw the commit graph\n\nThe history is an ordinary list of buttons. Each row owns a small SVG\nfor its lane segment, marked aria-hidden, over content that already\nreads correctly without it — a summary, an author, a , and a\nshort id. Row is a button because selecting a commit is an action and a\nbutton brings its own role, focus ring and Enter handling; that limits\nit to phrasing content, which is why ref badges are spans with a\nvisually hidden word for what each one is.\n\nTen lane colours come from one CSS rule rather than ten values: the\nlane index rotates an oklch hue, starting near the seed colour so the\ngraph belongs to the same palette as everything else. Nothing about\ncolour reaches the Rust side, and switching light to dark re-themes the\nwhole graph without a re-render.\n\nNothing here needs a virtual list. Rows arrive in batches as the end of\nthe list nears the viewport, so the DOM only holds what has been\nscrolled to, and rows past the first forty get content-visibility so\nthe browser skips laying out and painting the rest. `none` for the\nintrinsic width keeps an off-screen row from collapsing horizontally,\nwhich would make the scroll width jump on a wide graph. Which rows are\ndeferred is decided by nth-child, so no JavaScript is involved.\n\nWide graphs scroll rather than compress. A graph's width comes from the\nhistory — aspnetcore lays out 115 columns — so no window size makes it\nfit and capping would only hide branches.\n\nTwo things caught while driving the real components in a browser:\n\n- Five arrow presses moved the selection one row. Solid 2 batches\n signal writes, so reading selected() to compute the next value goes\n stale when presses land in one batch. Now the update goes through the\n functional setter, which sees the pending value.\n- jsx-a11y assumes React's htmlFor, so Solid's plain `for` read as an\n unassociated label. Configured rather than worked around.\n\nSafari drops list semantics from a list styled with list-style: none,\nwhich Tailwind's preflight applies — and Tauri is WKWebView on macOS,\nso that is the default here. An empty-string marker renders nothing and\nleaves the list a list, which beats patching the role back with ARIA.\n\nOpening is a path field for now. A native folder picker needs the Tauri\ndialog plugin, which is a dependency this change does not need; the\npicker arrives with the repository-picker work.\n\nInfinite scrolling is not verified yet — IntersectionObserver does not\nrun in a hidden document, and the browser pane available here cannot\nmake one visible. The Playwright harness covers it next.\n\nCo-Authored-By: Claude Opus 5 \n","tree":"2b0493859691be843589c8b062e5e045085099f5"},"mode":"0100644","name":"StatusBar.tsx","size":671},{"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