{"entries":[{"lastCommit":{"author":{"email":"claas@claas.dev","name":"Claas","when":"2026-08-06T09:24:26.000000+02:00"},"committer":{"email":"claas@claas.dev","name":"Claas","when":"2026-08-06T17:29:41.000000+02:00"},"hash":"f1c9d3538543b0d8e6cdac72ab0504ddb908f6ae","message":"Test the UI in real browsers without a native build\n\nThe frontend is an ordinary web page, so it can be driven in a real\nbrowser with the IPC boundary mocked — no Tauri build, no app window,\nand no system permissions. That was the gap: the only way to look at\nthis before was screen recording, which needs a permission that cannot\nbe granted from a terminal, and poking at a console by hand, which HMR\nwiped on every reload.\n\nVitest covers components and stores, Playwright covers flows across\nWebKit and Chromium. WebKit is the one that earns its place: it is what\nWKWebView on macOS actually is, so it is the only way to find out\nwhether something works in the shipped app rather than only in Chrome.\n\nIt found a real bug within minutes. WebKit leaves links out of the tab\norder unless the system's keyboard navigation setting is on, and that\nis off by default — so the skip link, whose entire job is to be the\nfirst thing a keyboard reaches, was unreachable in the shipped app.\nFixed one branch down; the assertion that catches it lives here.\n\nAlso verified here for the first time: infinite scrolling. Loading more\non scroll depends on IntersectionObserver, which does not run in a\nhidden document, so it could not be checked in the tooling available\nbefore.\n\nSome deliberate choices:\n\n- No screenshot baselines. They differ by operating system and font\n rendering, so they fail for reasons unrelated to the change. The\n appearance tests assert computed colours instead, which is the actual\n claim being made — that light-dark() resolves per scheme with no\n JavaScript.\n- Fixtures are typed as the real GraphRow, so a change to the Rust\n types stops the tests compiling rather than letting them drift.\n- One file knows about window.__TAURI_INTERNALS__, so an upgrade that\n moves it breaks in one place.\n\nTwo lint rules needed configuring for tests specifically: a mock has to\nkeep the signature of what it imitates, and commands reject with a\nplain { kind, message }, so a mock throwing an Error would misrepresent\nwhat the frontend has to handle.\n\nCo-Authored-By: Claude Opus 5 \n","tree":"52937c4fa97d11826f7935910c7b6cfb2f2480e4"},"mode":"0100644","name":"format.test.ts","size":2089},{"lastCommit":{"author":{"email":"claas@claas.dev","name":"Claas","when":"2026-08-06T09:24:26.000000+02:00"},"committer":{"email":"claas@claas.dev","name":"Claas","when":"2026-08-06T17:29:41.000000+02:00"},"hash":"f1c9d3538543b0d8e6cdac72ab0504ddb908f6ae","message":"Test the UI in real browsers without a native build\n\nThe frontend is an ordinary web page, so it can be driven in a real\nbrowser with the IPC boundary mocked — no Tauri build, no app window,\nand no system permissions. That was the gap: the only way to look at\nthis before was screen recording, which needs a permission that cannot\nbe granted from a terminal, and poking at a console by hand, which HMR\nwiped on every reload.\n\nVitest covers components and stores, Playwright covers flows across\nWebKit and Chromium. WebKit is the one that earns its place: it is what\nWKWebView on macOS actually is, so it is the only way to find out\nwhether something works in the shipped app rather than only in Chrome.\n\nIt found a real bug within minutes. WebKit leaves links out of the tab\norder unless the system's keyboard navigation setting is on, and that\nis off by default — so the skip link, whose entire job is to be the\nfirst thing a keyboard reaches, was unreachable in the shipped app.\nFixed one branch down; the assertion that catches it lives here.\n\nAlso verified here for the first time: infinite scrolling. Loading more\non scroll depends on IntersectionObserver, which does not run in a\nhidden document, so it could not be checked in the tooling available\nbefore.\n\nSome deliberate choices:\n\n- No screenshot baselines. They differ by operating system and font\n rendering, so they fail for reasons unrelated to the change. The\n appearance tests assert computed colours instead, which is the actual\n claim being made — that light-dark() resolves per scheme with no\n JavaScript.\n- Fixtures are typed as the real GraphRow, so a change to the Rust\n types stops the tests compiling rather than letting them drift.\n- One file knows about window.__TAURI_INTERNALS__, so an upgrade that\n moves it breaks in one place.\n\nTwo lint rules needed configuring for tests specifically: a mock has to\nkeep the signature of what it imitates, and commands reject with a\nplain { kind, message }, so a mock throwing an Error would misrepresent\nwhat the frontend has to handle.\n\nCo-Authored-By: Claude Opus 5 \n","tree":"52937c4fa97d11826f7935910c7b6cfb2f2480e4"},"mode":"0100644","name":"graph-lanes.test.tsx","size":4723},{"lastCommit":{"author":{"email":"claas@claas.dev","name":"Claas","when":"2026-08-06T09:24:26.000000+02:00"},"committer":{"email":"claas@claas.dev","name":"Claas","when":"2026-08-06T17:29:41.000000+02:00"},"hash":"f1c9d3538543b0d8e6cdac72ab0504ddb908f6ae","message":"Test the UI in real browsers without a native build\n\nThe frontend is an ordinary web page, so it can be driven in a real\nbrowser with the IPC boundary mocked — no Tauri build, no app window,\nand no system permissions. That was the gap: the only way to look at\nthis before was screen recording, which needs a permission that cannot\nbe granted from a terminal, and poking at a console by hand, which HMR\nwiped on every reload.\n\nVitest covers components and stores, Playwright covers flows across\nWebKit and Chromium. WebKit is the one that earns its place: it is what\nWKWebView on macOS actually is, so it is the only way to find out\nwhether something works in the shipped app rather than only in Chrome.\n\nIt found a real bug within minutes. WebKit leaves links out of the tab\norder unless the system's keyboard navigation setting is on, and that\nis off by default — so the skip link, whose entire job is to be the\nfirst thing a keyboard reaches, was unreachable in the shipped app.\nFixed one branch down; the assertion that catches it lives here.\n\nAlso verified here for the first time: infinite scrolling. Loading more\non scroll depends on IntersectionObserver, which does not run in a\nhidden document, so it could not be checked in the tooling available\nbefore.\n\nSome deliberate choices:\n\n- No screenshot baselines. They differ by operating system and font\n rendering, so they fail for reasons unrelated to the change. The\n appearance tests assert computed colours instead, which is the actual\n claim being made — that light-dark() resolves per scheme with no\n JavaScript.\n- Fixtures are typed as the real GraphRow, so a change to the Rust\n types stops the tests compiling rather than letting them drift.\n- One file knows about window.__TAURI_INTERNALS__, so an upgrade that\n moves it breaks in one place.\n\nTwo lint rules needed configuring for tests specifically: a mock has to\nkeep the signature of what it imitates, and commands reject with a\nplain { kind, message }, so a mock throwing an Error would misrepresent\nwhat the frontend has to handle.\n\nCo-Authored-By: Claude Opus 5 \n","tree":"52937c4fa97d11826f7935910c7b6cfb2f2480e4"},"mode":"0100644","name":"graph-view.test.ts","size":5634},{"lastCommit":{"author":{"email":"claas@claas.dev","name":"Claas","when":"2026-08-06T09:24:26.000000+02:00"},"committer":{"email":"claas@claas.dev","name":"Claas","when":"2026-08-06T17:29:41.000000+02:00"},"hash":"f1c9d3538543b0d8e6cdac72ab0504ddb908f6ae","message":"Test the UI in real browsers without a native build\n\nThe frontend is an ordinary web page, so it can be driven in a real\nbrowser with the IPC boundary mocked — no Tauri build, no app window,\nand no system permissions. That was the gap: the only way to look at\nthis before was screen recording, which needs a permission that cannot\nbe granted from a terminal, and poking at a console by hand, which HMR\nwiped on every reload.\n\nVitest covers components and stores, Playwright covers flows across\nWebKit and Chromium. WebKit is the one that earns its place: it is what\nWKWebView on macOS actually is, so it is the only way to find out\nwhether something works in the shipped app rather than only in Chrome.\n\nIt found a real bug within minutes. WebKit leaves links out of the tab\norder unless the system's keyboard navigation setting is on, and that\nis off by default — so the skip link, whose entire job is to be the\nfirst thing a keyboard reaches, was unreachable in the shipped app.\nFixed one branch down; the assertion that catches it lives here.\n\nAlso verified here for the first time: infinite scrolling. Loading more\non scroll depends on IntersectionObserver, which does not run in a\nhidden document, so it could not be checked in the tooling available\nbefore.\n\nSome deliberate choices:\n\n- No screenshot baselines. They differ by operating system and font\n rendering, so they fail for reasons unrelated to the change. The\n appearance tests assert computed colours instead, which is the actual\n claim being made — that light-dark() resolves per scheme with no\n JavaScript.\n- Fixtures are typed as the real GraphRow, so a change to the Rust\n types stops the tests compiling rather than letting them drift.\n- One file knows about window.__TAURI_INTERNALS__, so an upgrade that\n moves it breaks in one place.\n\nTwo lint rules needed configuring for tests specifically: a mock has to\nkeep the signature of what it imitates, and commands reject with a\nplain { kind, message }, so a mock throwing an Error would misrepresent\nwhat the frontend has to handle.\n\nCo-Authored-By: Claude Opus 5 \n","tree":"52937c4fa97d11826f7935910c7b6cfb2f2480e4"},"mode":"0100644","name":"history.test.tsx","size":6487},{"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":"recents.test.ts","size":2641},{"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":"refs.test.ts","size":8697},{"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":"repository-nav.test.tsx","size":8133}],"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"}}