{"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":"recent-repositories.test.tsx","size":4361},{"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":"2026-08-06T15:39:56.000000+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