Tests / @ogygia/devtools
@ogygia/devtools
Version 0.14.0 latest
620ddcca 925 test runs (925 tests, each run in Node, the browsers or both)
0.41× lines of test for each line of source (26,220 to 63,558) Only real test code counts: test files and their helpers, not fixtures, sample apps or benches.
Blank lines are left out on both sides.
Exactly what is counted
non-blank lines of test code across the repo (*.test.ts, *.test-d.ts, *.spec.ts, and the .ts helpers in test/ and e2e/ folders; no fixtures, fixture apps, stubs or benches), over non-blank lines of packages/*/src (no .d.ts or generated files)59.6% of source lines run by the tests
- 894Unit
- 0Browser
- 31End to end
- 0Type tests
- 0Lint rules
- 0Failing
Coverage
- 59.6%Lines
- 52.8%Branches
- 46.0%Functions
- 56.9%Statements
Every test
925 tests
| Test | Kind | Node | Chromium | Firefox | WebKit |
|---|---|---|---|---|---|
packages/devtools/e2e/dev-installed.spec.ts | |||||
| a snippet-prop island hydrates, and no ogygia module is loaded twice | End to end | not run | pass | pass | pass |
| opening the devtools dock on a cold server finds nothing late: no re-optimize, no reload | End to end | not run | pass | pass | pass |
| the devtools name it when it happens anyway: an app pre-bundling ogygia's boot runs two copies | End to end | not run | pass | pass | pass |
packages/devtools/e2e/devtools.spec.ts | |||||
| a cold dev server: opening the dock and its tabs finds no dependency late (no re-optimize, no reload) | End to end | not run | pass | pass | pass |
| a recording stopped a second time leaves the recording after it alone (#240)#240 | End to end | not run | pass | pass | pass |
| hub: populated — /transportable shows a shared instance reunited across islands | End to end | not run | pass | pass | pass |
| page: the lab page's planted problems are found, pinned on their islands, and shown | End to end | not run | pass | pass | pass |
| profiler: a run renders natively and its island rows open the island detail | End to end | not run | pass | pass | pass |
| record: a session finds the slow click and names its handler; hydration lists each island | End to end | not run | pass | pass | pass |
| schema, server realm, panel tabs, event-driven wake/replay, identity spine, nav, trace, timeline | End to end | not run | pass | pass | pass |
| the dock’s code fails to download once: the button does not stick on “loading…”, a click loads it again (#201)#201 | End to end | not run | pass | pass | pass |
| twin islands (one fingerprint, two elements) and a Kit-hydrated page | End to end | not run | pass | pass | pass |
packages/devtools/e2e/dock-switch.spec.ts | |||||
| an app with the handle, devtools on: the server’s tag is the beacon, no chunk is asked for | End to end | not run | pass | pass | pass |
| an app without ogygia’s handle, devtools off: no beacon, and its code is never asked for | End to end | not run | pass | pass | pass |
| an app without ogygia’s handle, devtools on: the dock starts the page beacon | End to end | not run | pass | pass | pass |
| off while the dock's code downloads: the dock never mounts | End to end | not run | pass | pass | pass |
| only the launcher up, then off: Alt+O loads no dock | End to end | not run | pass | pass | pass |
| the dock open, then off, then on and open: one dock runs | End to end | not run | pass | pass | pass |
| the dock open, then off: its listeners and observers go, and Alt+O does nothing | End to end | not run | pass | pass | pass |
packages/devtools/e2e/profiler-beacon-standalone.spec.ts | |||||
| a Kit page reports its vitals once and its visit, with no regions, from the one page beacon | End to end | not run | pass | pass | pass |
| a page hidden right after it loads sends its visit once, and no copy after it | End to end | not run | pass | pass | pass |
| a page restored from the back/forward cache reports its visit again | End to end | not run | pass | pass | pass |
| coming back to the tab opens the visit again: the next hide sends it | End to end | not run | pass | pass | pass |
packages/devtools/e2e/profiler-dev-css.spec.ts | |||||
| an app page whose graph reaches the profiler shell carries none of its global sheet | End to end | not run | pass | pass | pass |
packages/devtools/e2e/profiler-serverless.spec.ts | |||||
| every report action works after the recording instance is gone | End to end | not run | pass | pass | pass |
| on AWS Lambda a recording of the heaviest page answers inside the 30 s cut, with its passes in | End to end | not run | pass | pass | pass |
| pick your fixes: the checklist recounts the forecast in the browser | End to end | not run | pass | pass | pass |
| profile again & compare renders from this browser when no instance holds either report | End to end | not run | pass | pass | pass |
| the browser's CPU names an island by its component, not the minified name in its chunk | End to end | not run | pass | skip | skip |
packages/devtools/e2e/profiler-window-instances.spec.ts | |||||
| a window started on one instance gathers the requests every instance served, and stops on any | End to end | not run | pass | pass | pass |
| stopped on the instance that started it: its CPU profile and every instance’s requests | End to end | not run | pass | pass | pass |
packages/devtools/test/build-module-map.test.ts | |||||
| build_module_map > each module at its lines of the chunk, ids relative to the root, queries dropped | Unit | pass | not run | not run | not run |
| build_module_map: a module starting with a newline > the modules after it keep their lines | Unit | pass | not run | not run | not run |
| build_module_map: a module whose imports the bundler dropped > is found by its first solid line, and the modules after it keep their lines | Unit | pass | not run | not run | not run |
| build_module_map: the app's own folding comments inside a module > a //#region that names no module is not a module, and does not end the real one early | Unit | pass | not run | not run | not run |
| build_module_map: the bundler's own region markers > each //#region … //#endregion is a module at exactly those lines | Unit | pass | not run | not run | not run |
| module_category > the profiler, ogygia, a package, or (undefined) the app | Unit | pass | not run | not run | not run |
packages/devtools/test/build-profiler-maps.test.ts | |||||
| a chunk with no map > the module map's script module at the line, its lines found by their text | Unit | pass | not run | not run | not run |
| a chunk with no map: what a build does to lines > types stripped, two lines folded into one, one line wrapped over five | Unit | pass | not run | not run | not run |
| embedded_key > a browser chunk: the same key from a local copy of the client output or from its URL | Unit | pass | not run | not run | not run |
| embedded_key > a script's path under the server output, from Kit's output or an adapter map's source | Unit | pass | not run | not run | not run |
| embedded_source > only the app's own src/ files: a package's src/ and a dependency are not filed under the app | Unit | pass | not run | not run | not run |
| fill_profiler_maps > embeds the browser's maps for the app's own chunks from the stash, and removes the stash | Unit | pass | not run | not run | not run |
| fill_profiler_maps > writes the module map and every chunk's map into the placeholder, and nothing else changes | Unit | pass | not run | not run | not run |
| module_imports > a server file's $lib and relative script imports, resolved, with the files each may be; packages and components left out | Unit | pass | not run | not run | not run |
| route_imports > an import spread over lines | Unit | pass | not run | not run | not run |
| route_imports > the same-named component the route imports | Unit | pass | not run | not run | not run |
| the resolver follows a map to another built file > adapter chunk → Kit's chunk → the source | Unit | pass | not run | not run | not run |
packages/devtools/test/build-region-bytes.test.ts | |||||
| island_subgraph_bytes > a barrel row says why the barrel pass left it, joined by the barrel file and its importers | Unit | pass | not run | not run | not run |
| island_subgraph_bytes > a barrel the island still imports whole: little code, many app modules behind it | Unit | pass | not run | not run | not run |
| island_subgraph_bytes > a site-kit island (ogygia's content components): no app bytes, and still its lines | Unit | pass | not run | not run | not run |
| island_subgraph_bytes > a small component composing many children is no barrel: it uses what it imports | Unit | pass | not run | not run | not run |
| island_subgraph_bytes > ignores non-island modules and cold (untransformed) modules | Unit | pass | not run | not run | not run |
| island_subgraph_bytes > its components' lines that draw differently in the browser, read from their sources on disk | Unit | pass | not run | not run | not run |
| island_subgraph_bytes > keeps the larger subgraph when two graph nodes share one island id (?v= variants) | Unit | pass | not run | not run | not run |
| island_subgraph_bytes > prunes the framework (svelte / ogygia runtime) — shared once per page, not per island | Unit | pass | not run | not run | not run |
| island_subgraph_bytes > prunes the shared transportables registry (would otherwise drag in the whole app) | Unit | pass | not run | not run | not run |
| island_subgraph_bytes > sums the component + its child app modules, not the entry glue | Unit | pass | not run | not run | not run |
packages/devtools/test/build-server-sourcemaps.test.ts | |||||
| server_sourcemaps_plugin > the app's own choice wins; no profiler, no dev server: nothing | Unit | pass | not run | not run | not run |
| server_sourcemaps_plugin > the server build gets hidden maps; the client build Kit starts after it keeps them hidden and moves them off its output | Unit | pass | not run | not run | not run |
| server_sourcemaps_plugin > two apps built in one process: one app's hidden maps are not taken from the other | Unit | pass | not run | not run | not run |
packages/devtools/test/devtools-bytes.test.ts | |||||
| an island's id from a dev entry and from a built one (the names key the bare id) | Unit | pass | not run | not run | not run |
| each file once, what only each island needs, the shared files with their users | Unit | pass | not run | not run | not run |
| two copies of an island are one row (their files once) | Unit | pass | not run | not run | not run |
packages/devtools/test/devtools-cache-headers.test.ts | |||||
| cache_findings > a header many files share is named once | Unit | pass | not run | not run | not run |
| cache_findings > a missing header is named as such; many files are cut to four names | Unit | pass | not run | not run | not run |
| cache_findings > every file cached for good, or none probed: quiet | Unit | pass | not run | not run | not run |
| cache_findings > files not cached long: named, with the header seen, the bytes, and the fix | Unit | pass | not run | not run | not run |
| cache_findings > many files: the heaviest four are named, the rest counted | Unit | pass | not run | not run | not run |
| cached_long: does this cache-control keep the file long enough > what makes a returning visitor fetch again | Unit | pass | not run | not run | not run |
| cached_long: does this cache-control keep the file long enough > what SvelteKit sends, and other long policies | Unit | pass | not run | not run | not run |
| content_named_files > a build: each island file and the runtime, once, by name | Unit | pass | not run | not run | not run |
| content_named_files > off a build (no src, or the src is the entry): nothing | Unit | pass | not run | not run | not run |
| probe_cache > a file's bytes are the page's load of it, not an earlier probe's empty HEAD | Unit | pass | not run | not run | not run |
| probe_cache > a probe that failed says nothing about the file: it is left out, never "no cache-control" | Unit | pass | not run | not run | not run |
| probe_cache > a refused probe (a 405 for HEAD) says nothing about the file either | Unit | pass | not run | not run | not run |
| probe_cache > asks the server with a HEAD past the browser cache, and reads its cache-control | Unit | pass | not run | not run | not run |
packages/devtools/test/devtools-client-realms.test.ts | |||||
| devtools' browser realms > with the dock the build registers the dock realm | Unit | pass | not run | not run | not run |
| devtools' browser realms > without the dock the build registers the beacon realm, which subscribes to the beacon hooks alone | Unit | pass | not run | not run | not run |
packages/devtools/test/devtools-hook-record.test.ts | |||||
| settled > a hole that woke and then failed for good, or was answered, is not in flight | Unit | pass | not run | not run | not run |
| settled > a live region that mounted (no regionHydrated after it) and one that left mid-wake are not in flight | Unit | pass | not run | not run | not run |
| settled > a region that mounted and is still waking holds it | Unit | pass | not run | not run | not run |
| settled > a region that woke and is still downloading (no regionMount yet) holds it past the quiet time | Unit | pass | not run | not run | not run |
| the hook record > holds a region only weakly and keeps no server markup, so a region the page left can go | Unit | pass | not run | not run | not run |
| the hook record > keeps each call with when it came and the region its element names | Unit | pass | not run | not run | not run |
| the hook record > the writer and the readers, in any copy, share one record, and a listener hears each call | Unit | pass | not run | not run | not run |
packages/devtools/test/devtools-last-nav.test.ts | |||||
| last_nav > a navigation that finished: its start, its address | Unit | pass | not run | not run | not run |
| last_nav > an overtaken navigation is not the page: the one that finished is | Unit | pass | not run | not run | not run |
| last_nav > one that started and never finished changes nothing | Unit | pass | not run | not run | not run |
packages/devtools/test/devtools-ledger-tab.test.ts | |||||
| the Bytes tab, with an island graph > lists the modules this page's last profile loaded twice | Unit | pass | not run | not run | not run |
| the Bytes tab, with an island graph > shows the page's exact ledger in place of the per-chunk estimate | Unit | pass | not run | not run | not run |
| the Bytes tab, without an island graph > lists one row per entry chunk, with how many regions share it | Unit | pass | not run | not run | not run |
| the Bytes tab, without an island graph > puts the dearest island first: by its whole subgraph, then by its own chunk | Unit | pass | not run | not run | not run |
| the Bytes tab, without an island graph > totals the loaded chunks and the runtime for the wire, and every subgraph for +deps | Unit | pass | not run | not run | not run |
packages/devtools/test/devtools-leftovers-ledger.test.ts | |||||
| a second copy of the readers (the testing engine a test injects) > reads the build’s dev flag, names and sidecar lookup the realm lent through the hook record | Unit | pass | not run | not run | not run |
| a second copy of the readers (the testing engine a test injects) > sees the islands the realm’s copy of the leftovers watch saw | Unit | pass | not run | not run | not run |
| a second copy of the readers (the testing engine a test injects) > with no realm on the page, the build is not the dev server’s and nothing is live | Unit | pass | not run | not run | not run |
packages/devtools/test/devtools-leftovers.test.ts | |||||
| leftovers > a copy alive now that was there to make it: not left behind (one kept across the navigation) | Unit | pass | not run | not run | not run |
| leftovers > a scroll-holding listener is the nearest island’s among its callers (Svelte’s code above it), else its file’s | Unit | pass | not run | not run | not run |
| leftovers > an island gone, its interval and listener still set: named, by its .svelte file in dev | Unit | pass | not run | not run | not run |
| leftovers > back on its page, a new copy alive: the old copy’s interval is still named | Unit | pass | not run | not run | not run |
| leftovers > by location in a build: the caller’s caller is the island’s file | Unit | pass | not run | not run | not run |
| leftovers > made before any copy started (module code, Svelte’s own delegation), or by no island: quiet | Unit | pass | not run | not run | not run |
| leftovers > reads each frame’s file: Chrome, Firefox/Safari, no position, no query | Unit | pass | not run | not run | not run |
packages/devtools/test/devtools-messages.test.ts | |||||
| the devtools messages > each code's page reads as plain sentences: no { } or < outside backticks | Unit | pass | not run | not run | not run |
| the devtools messages > the dock's failed download: its wording, then its page | Unit | pass | not run | not run | not run |
packages/devtools/test/devtools-page-asks.test.ts | |||||
| a file's Vary header is asked again after a failed ask | Unit | pass | not run | not run | not run |
| the dev server's pre-bundled packages are asked again after a failed ask | Unit | pass | not run | not run | not run |
packages/devtools/test/devtools-page-nav.test.ts | |||||
| since_nav > keeps only this page's events, shifted to the navigation | Unit | pass | not run | not run | not run |
| since_nav > no finding leans on the first page's paints or vitals | Unit | pass | not run | not run | not run |
packages/devtools/test/devtools-page-tab-read.test.ts | |||||
| the Page tab when reading the visit > a read that throws: said as a failure, with its message, not as "not measured" | Unit | pass | not run | not run | not run |
| the Page tab when reading the visit > no record: "not measured" | Unit | pass | not run | not run | not run |
packages/devtools/test/devtools-profile-busy.test.ts | |||||
| a Retry-After past any timer waits, and ends with its owner, never failing (#1096)#1096 | Unit | pass | not run | not run | not run |
| asks again when the 409's Retry-After says | Unit | pass | not run | not run | not run |
| says it is asking only while its owner lives: an owner that ends during the wait hears no more | Unit | pass | not run | not run | not run |
| stops asking once its owner ends | Unit | pass | not run | not run | not run |
packages/devtools/test/devtools-profile-store.test.ts | |||||
| island_row > joins by fingerprint, then by entry | Unit | pass | not run | not run | not run |
| run_profile when the profiler refuses > a heap too full to profile: the profiler's sentence, not its JSON | Unit | pass | not run | not run | not run |
| run_profile when the profiler refuses > another failure: its status and its words | Unit | pass | not run | not run | not run |
| slim_profile > keeps the score's lost points, what was left out, the score before, and the page's files | Unit | pass | not run | not run | not run |
| slim_profile > keeps what the dock shows, in order | Unit | pass | not run | not run | not run |
| slim_profile > survives a report missing everything | Unit | pass | not run | not run | not run |
| the profile store is bounded (#413) > keeps the most recently profiled pages, not one per page a tab ever profiled#413 | Unit | pass | not run | not run | not run |
| the slim profile keeps what the browser cannot read > a prerendered page: the file served, or a route that prerenders | Unit | pass | not run | not run | not run |
| the slim profile keeps what the browser cannot read > an island's line that draws differently in the browser: a rendered value before a guard | Unit | pass | not run | not run | not run |
| the slim profile keeps what the browser cannot read > the modules this page loads twice, with where each copy came from | Unit | pass | not run | not run | not run |
| the slim profile keeps what the browser cannot read > the page's Cache-Control, when the report has it | Unit | pass | not run | not run | not run |
packages/devtools/test/devtools-realm.test.ts | |||||
| two copies of the realm > a hook call both hear is in the record once | Unit | pass | not run | not run | not run |
| two copies of the realm > each role handles a call once, whichever copy asks first, and the roles do not share | Unit | pass | not run | not run | not run |
| two copies of the realm > the leftovers watch patches the page’s timers once | Unit | pass | not run | not run | not run |
packages/devtools/test/devtools-session-console.test.ts | |||||
| a recording and the console > a wrapper the page put on console.error during the recording stays after the stop | Unit | pass | not run | not run | not run |
| a recording and the console > with nothing on top, the stop puts console.error back as it was | Unit | pass | not run | not run | not run |
packages/devtools/test/devtools-session-insights.test.ts | |||||
| analyze_session > a click outside islands is not dead when the page changed (but not the islands other clicks hit) | Unit | pass | not run | not run | not run |
| analyze_session > a failed island is one finding (hydrate-failed), not also ogygia's own console log of it | Unit | pass | not run | not run | not run |
| analyze_session > a quiet session says nothing | Unit | pass | not run | not run | not run |
| analyze_session > a request's own server split (Server-Timing): the upstream it waited on | Unit | pass | not run | not run | not run |
| analyze_session > an interaction: its length is the longest event, its split the event that ran the handlers | Unit | pass | not run | not run | not run |
| analyze_session > every failed request, and the slowest interaction on each element however quick | Unit | pass | not run | not run | not run |
| analyze_session > repeated clicks are rage, a click with no effect is dead, a late answer is late | Unit | pass | not run | not run | not run |
| analyze_session > requests: failed, slow, repeated | Unit | pass | not run | not run | not run |
| analyze_session > shifts only count when no input caused them; churn names the island | Unit | pass | not run | not run | not run |
| analyze_session > the timeline puts everything on one clock, requests with their server part | Unit | pass | not run | not run | not run |
packages/devtools/test/devtools-since-load.test.ts | |||||
| a moved vital names its part that moved most (the shared rule); a module load needs 50 ms | Unit | pass | not run | not run | not run |
| fixed, new, and what moved | Unit | pass | not run | not run | not run |
| the dev server's own doing is never new or fixed: a dev-CSS shift, and every module warmer at once | Unit | pass | not run | not run | not run |
| the last load was the dev server’s first compile: nothing "fixed" by the warm-up, loads compared without it | Unit | pass | not run | not run | not run |
| the same finding on another island is a new one; another page compares to nothing | Unit | pass | not run | not run | not run |
packages/devtools/test/devtools-source-lines.test.ts | |||||
| lines_of_map > each served line → the source line of its first segment | Unit | pass | not run | not run | not run |
| lines_of_map > empty mappings map nothing | Unit | pass | not run | not run | not run |
| lines_of_map > only the wanted source counts (an inlined helper is not the file’s line) | Unit | pass | not run | not run | not run |
packages/devtools/test/devtools-styles-read.test.ts | |||||
| read_styles > counts a nested rule’s `&` as used: it needs its parent to mean anything | Unit | pass | not run | not run | not run |
| read_styles > counts a selector this browser cannot query (a vendor pseudo) as used | Unit | pass | not run | not run | not run |
| read_styles > leaves out a linked sheet that styles only the devtools panel, keeps one the page uses too | Unit | pass | not run | not run | not run |
| read_styles > leaves out devtools’ own sheets a dev server injected, from the package or its workspace source | Unit | pass | not run | not run | not run |
| read_styles > names a dev server’s injected <style> by its file, and a plain one inline | Unit | pass | not run | not run | not run |
| read_styles > reads the rules inside @media, @supports and nested blocks too | Unit | pass | not run | not run | not run |
| read_styles > takes a pseudo-element off up to the next space or combinator, past a space in its parens | Unit | pass | not run | not run | not run |
| read_styles > takes off a whole state only: `:target` leaves `:target-current` alone | Unit | pass | not run | not run | not run |
| scan_unscoped > reads an inline sheet’s marker from its node, a linked one’s from one same-origin fetch the HTTP cache answers | Unit | pass | not run | not run | not run |
packages/devtools/test/devtools-styles.test.ts | |||||
| selectors, ready to try against the page > splits on top-level commas only | Unit | pass | not run | not run | not run |
| selectors, ready to try against the page > takes off pseudo-elements and momentary states, keeps real pseudo-classes | Unit | pass | not run | not run | not run |
| styles findings > an unscoped component is a warning, first | Unit | pass | not run | not run | not run |
| styles findings > names the sheets that hold the unmatched rules | Unit | pass | not run | not run | not run |
| styles findings > says nothing about a small or mostly used sheet | Unit | pass | not run | not run | not run |
| the profiler finds the marker in the files it weighs > in a linked stylesheet | Unit | pass | not run | not run | not run |
| the profiler finds the marker in the files it weighs > in an inline <style>, but not in page text that quotes it | Unit | pass | not run | not run | not run |
| the profiler finds the marker in the files it weighs > nothing marked, nothing said | Unit | pass | not run | not run | not run |
| the unscoped-fallback marker > a reason that would close the comment cannot | Unit | pass | not run | not run | not run |
| the unscoped-fallback marker > carries the file name only and the reason, and reads back | Unit | pass | not run | not run | not run |
| the unscoped-fallback marker > each file once, however many sheets repeat it | Unit | pass | not run | not run | not run |
| the unscoped-fallback marker > the finding names them and the way to the full reason | Unit | pass | not run | not run | not run |
packages/devtools/test/devtools-timeline-tab.test.ts | |||||
| the Timeline tab > draws only this page: the regions of the page before the last navigation are gone | Unit | pass | not run | not run | not run |
| the Timeline tab > draws the client's events only: a server event makes no lane | Unit | pass | not run | not run | not run |
| the Timeline tab > keeps a region's first event of each kind: a live region's later hydrations do not stretch the axis | Unit | pass | not run | not run | not run |
| the Timeline tab > leaves the bar of a region still waking open | Unit | pass | not run | not run | not run |
packages/devtools/test/html-held-open.test.ts | |||||
| defer_keys > names each streamed id by its page.data key (one level down too), from the document or the seed alone | Unit | pass | not run | not run | not run |
| defer_keys > reads the seed script itself: not the page facts beside it, not its type written in text (#1370)#1370 | Unit | pass | not run | not run | not run |
| explain_held_open > decoys: a short stream, a paint that came with the end, islands that hydrated before the end | Unit | pass | not run | not run | not run |
| explain_held_open > nothing known of the stream: a large HTML is its download; else say what usually holds it | Unit | pass | not run | not run | not run |
| explain_held_open > the browser: painted early, the HTML kept coming, every island hydrated after its end — named with the key that held it | Unit | pass | not run | not run | not run |
| explain_held_open > the server alone (no visit): the tail left long after the rest, held by a key | Unit | pass | not run | not run | not run |
| stream_tail > splits at the longest pause and names what settled in the tail | Unit | pass | not run | not run | not run |
packages/devtools/test/page-beacon-start.test.ts | |||||
| a beacon chunk that loads starts the beacon and says nothing | Unit | pass | not run | not run | not run |
| a beacon that throws as it starts is warned with the same code, and the start never rejects | Unit | pass | not run | not run | not run |
| a failed beacon chunk download prints its warning once, however often the start is asked | Unit | pass | not run | not run | not run |
packages/devtools/test/page-score.test.ts | |||||
| curve > just under the p10 does not round into the green | Unit | pass | not run | not run | not run |
| curve > passes through its control points and never cliffs | Unit | pass | not run | not run | not run |
| page_score > a browser that cannot see long tasks or shifts: unknown, named — never a clean zero | Unit | pass | not run | not run | not run |
| page_score > a fast machine's clean visit does not wash out heavy JS | Unit | pass | not run | not run | not run |
| page_score > a lean islands page scores an A | Unit | pass | not run | not run | not run |
| page_score > an unweighed page (dev server) leaves JS out, says why, and never reads 0 B | Unit | pass | not run | not run | not run |
| page_score > each category reports the points it cost, and the worst is the biggest loss | Unit | pass | not run | not run | not run |
| page_score > every millisecond counts: a slower server always scores lower | Unit | pass | not run | not run | not run |
| page_score > every sub-score is 0–100 and the grade bands hold | Unit | pass | not run | not run | not run |
| page_score > hydration faults are steep and ordered: recovered > never woke > changed | Unit | pass | not run | not run | not run |
| page_score > no browser visit: vitals and integrity are left out, not scored 100 | Unit | pass | not run | not run | not run |
| page_score > render-blocking counts requests too, not only bytes | Unit | pass | not run | not run | not run |
| page_score > THE FAIRNESS CASE: 6 MB of JS with no islands scores far below the lean islands page | Unit | pass | not run | not run | not run |
| what a visit fetched for nothing > the page weight says it, and the score stays the weighed bytes | Unit | pass | not run | not run | not run |
packages/devtools/test/profiler-almost-answer.test.ts | |||||
| almost the same answer > its own kind, what changes quoted, a caching lever like the same answer | Unit | pass | not run | not run | not run |
| almost the same answer > the GETs made in every render with more than one answer | Unit | pass | not run | not run | not run |
packages/devtools/test/profiler-app-path.test.ts | |||||
| app_relative > a file under the app folder reads from there; the build output and packages do not | Unit | pass | not run | not run | not run |
| app_relative > the ledger labels a nested src/ folder from the app root, not from the innermost src/ | Unit | pass | not run | not run | not run |
packages/devtools/test/profiler-async-origin.test.ts | |||||
| async origins > library work after an await is charged to the app function that called into the library | Unit | pass | not run | not run | not run |
| Node work the profiler asked for is overhead > a node frame under a profiler frame (no app frame between) is profiler time; under app code it stays node | Unit | pass | not run | not run | not run |
packages/devtools/test/profiler-background-refused.test.ts | |||||
| a sampler the inspector refused says it stopped | Unit | pass | not run | not run | not run |
| a trap the inspector refused says it caught nothing and stopped | Unit | pass | not run | not run | not run |
| a window started after the trap stopped does not show the trap armed | Unit | pass | not run | not run | not run |
packages/devtools/test/profiler-beacon-big-body.test.ts | |||||
| a big body as the page goes away sends the slim copy; nothing that cannot fit is tried | Unit | pass | not run | not run | not run |
| a big body while the page is open goes by an ordinary request, which has no size limit | Unit | pass | not run | not run | not run |
| a page leaving by a link still reads "visible": the hide sends the slim copy as a beacon, not an ordinary request the navigation cancels | Unit | pass | not run | not run | not run |
| a refused request never becomes a page error | Unit | pass | not run | not run | not run |
| a small body goes by sendBeacon, as before | Unit | pass | not run | not run | not run |
| the 64 KB is shared by everything queued at once: a hide that sends several fits them or slims them | Unit | pass | not run | not run | not run |
packages/devtools/test/profiler-beacon-one-owner.test.ts | |||||
| ogygia/profiler/client carries no beacon of its own: its marks reach the page's, and an early one waits for it | Unit | pass | not run | not run | not run |
| the second copy of the regions' half hands its hydrations to the first | Unit | pass | not run | not run | not run |
packages/devtools/test/profiler-beacon-resource-type.test.ts | |||||
| the page beacon: what a file counts as > a link is not a stylesheet by itself | Unit | pass | not run | not run | not run |
| the page beacon: what a file counts as > the extension first, then what started it | Unit | pass | not run | not run | not run |
packages/devtools/test/profiler-beacon-room.test.ts | |||||
| before any navigation a full list keeps the first ones (the profiler’s page) | Unit | pass | not run | not run | not run |
| the page in view gets room; the first page keeps all of its own | Unit | pass | not run | not run | not run |
packages/devtools/test/profiler-body-cap.test.ts | |||||
| profiler request bodies are capped (#1150) > a posted report: a body past its cap is 413, and the rest is never read#1150 | Unit | pass | not run | not run | not run |
| profiler request bodies are capped (#1150) > a posted report: a body that breaks off mid-upload is a 400, not a thrown error (#1427)#1150 | Unit | pass | not run | not run | not run |
| profiler request bodies are capped (#1150) > an uploaded .ogp: a body past its cap is 413, and the rest is never read#1150 | Unit | pass | not run | not run | not run |
| profiler request bodies are capped (#1150) > an uploaded .ogp: a body that breaks off mid-upload is a 400, not a thrown error (#1427)#1150 | Unit | pass | not run | not run | not run |
| profiler request bodies are capped (#1150) > the beacon: a body past its cap is 413, and the rest is never read#1150 | Unit | pass | not run | not run | not run |
| profiler request bodies are capped (#1150) > the beacon: a body that breaks off mid-upload is a 400, not a thrown error (#1427)#1150 | Unit | pass | not run | not run | not run |
| profiler request bodies are capped (#1150) > the login says it did not unlock (ok: false) for a body past its cap, and for one that breaks off#1150 | Unit | pass | not run | not run | not run |
| profiler request bodies are capped (#1150) > the login, open before sign-in: a body past its cap is 413, and the rest is never read#1150 | Unit | pass | not run | not run | not run |
| profiler request bodies are capped (#1150) > the login, open before sign-in: a body that breaks off mid-upload is a 400, not a thrown error (#1427)#1150 | Unit | pass | not run | not run | not run |
packages/devtools/test/profiler-browser-findings.test.ts | |||||
| a clean visit has no browser findings; a later record of the visit updates the regions | Unit | pass | not run | not run | not run |
| a failed island's wake → failure span is kept (what held the islands below the fold); junk is not | Unit | pass | not run | not run | not run |
| a hole answered late: waited from the first paint, split by its server time | Unit | pass | not run | not run | not run |
| a hole that kept its fallback reaches the report, named from the hole rows | Unit | pass | not run | not run | not run |
| a lazy island's code counts at start only when it loaded before the load event (a scroll right after is not the start) | Unit | pass | not run | not run | not run |
| a page with more files than the visit lists: counts and bytes cover all of them | Unit | pass | not run | not run | not run |
| a preload downloaded again reaches the report, even on a page with no island | Unit | pass | not run | not run | not run |
| a preload nothing used reaches the report, on a visit with nothing else to say | Unit | pass | not run | not run | not run |
| a recovered island's unpaired markers survive the parse; a paired or malformed count does not | Unit | pass | not run | not run | not run |
| a server transform’s restore gone wrong rides the visit: a late host named in the report | Unit | pass | not run | not run | not run |
| a visit with the devtools dock open as it loaded: the report says the paints may be the dock’s | Unit | pass | not run | not run | not run |
| an image sent far bigger than shown reaches the report; a record that breaks the rule is dropped | Unit | pass | not run | not run | not run |
| an island that drew nothing on the visitor's screen: eager-hidden, not "below the first screen" | Unit | pass | not run | not run | not run |
| an island whose own file was gone reaches the report, named from the island rows | Unit | pass | not run | not run | not run |
| bad region records are dropped | Unit | pass | not run | not run | not run |
| content-named files fetched again reach the report, named from the island rows, even with no island awake | Unit | pass | not run | not run | not run |
| each planted problem, on its island, in the report | Unit | pass | not run | not run | not run |
| holes in one batch request: a batched hole’s wait is the server’s, never the gate’s; a refused batch is named | Unit | pass | not run | not run | not run |
| images below the first screen that loaded at start reach the report | Unit | pass | not run | not run | not run |
| in-app navigations reach the report: parsed, merged once, out-of-order clocks dropped | Unit | pass | not run | not run | not run |
| scrolling that stalled reaches the report, on a visit with nothing else to say | Unit | pass | not run | not run | not run |
| text a late font kept invisible reaches the report; a bad face record is dropped | Unit | pass | not run | not run | not run |
| the inline beacon for pages without the runtime is one valid script tag | Unit | pass | not run | not run | not run |
| the inline beacon names a preload the browser downloaded again (a Kit page has no runtime beacon) | Unit | pass | not run | not run | not run |
| the inline beacon names a reloaded file's region, past a region whose src is not a URL | Unit | pass | not run | not run | not run |
| the slowest interaction reaches the report: parsed, the slower of two records kept, a built island file named | Unit | pass | not run | not run | not run |
packages/devtools/test/profiler-client-late-beacon.test.ts | |||||
| a mark made 11 s before the beacon loads reaches it | Unit | pass | not run | not run | not run |
packages/devtools/test/profiler-client-windows.test.ts | |||||
| a window profile keeps only the samples inside it | Unit | pass | not run | not run | not run |
| each island window and the long tasks outside them, by name | Unit | pass | not run | not run | not run |
packages/devtools/test/profiler-codec-bits.test.ts | |||||
| the .ogp layout > is 'OGP1', a 16-byte salt, a 12-byte iv, the ciphertext and a 16-byte tag | Unit | pass | not run | not run | not run |
| the profiler session among many cookies > finds a session Kit's parser missed | Unit | pass | not run | not run | not run |
| the profiler session among many cookies > tries Kit's parsed value first, then every raw pair once | Unit | pass | not run | not run | not run |
| the saving arithmetic > imports nothing, so the report page runs the same code in the browser | Unit | pass | not run | not run | not run |
| the sink buffer > when full, drops the oldest request rows, never a window or a catch | Unit | pass | not run | not run | not run |
packages/devtools/test/profiler-codec-off-loop.test.ts | |||||
| the .ogp codec > derives the key and compresses off the event loop | Unit | pass | not run | not run | not run |
packages/devtools/test/profiler-compress.test.ts | |||||
| gzip_large > gzips a big HTML or JSON answer the browser takes gzip for | Unit | pass | not run | not run | not run |
| gzip_large > hands back an answer with no body as it is (a 204, a 304) | Unit | pass | not run | not run | not run |
| gzip_large > leaves the rest alone | Unit | pass | not run | not run | not run |
packages/devtools/test/profiler-core-features.test.ts | |||||
| allocation on the timeline > finds the bursts, attributes them to what ran, and flags a GC inside | Unit | pass | not run | not run | not run |
| data lineage from the code > gives every key a verdict | Unit | pass | not run | not run | not run |
| data lineage from the code > ignores attributes, import paths, comments and prose | Unit | pass | not run | not run | not run |
| data lineage from the code > never calls a key unread when a reader could not be scanned or takes the whole object | Unit | pass | not run | not run | not run |
| data lineage from the code > reads the keys a component touches, without regex | Unit | pass | not run | not run | not run |
| data lineage from the code > sees a renamed prop, a spread, and a pass-through as whole | Unit | pass | not run | not run | not run |
| the instance was not alone > a render reading a prerender function over HTTP: the call, logged at its own address, is the render's own | Unit | pass | not run | not run | not run |
| the instance was not alone > counts every overlapping request, not only the 40 it lists | Unit | pass | not run | not run | not run |
| the instance was not alone > is absent when nothing overlapped | Unit | pass | not run | not run | not run |
| the instance was not alone > names the overlapping requests with a bounded CPU estimate | Unit | pass | not run | not run | not run |
| the render, step by step > folds the smallest steps under the cap and takes the stack from the index | Unit | pass | not run | not run | not run |
| the render, step by step > merges one owner’s consecutive segments and carries running totals | Unit | pass | not run | not run | not run |
| the stack index > answers a range and an instant | Unit | pass | not run | not run | not run |
| the stack index > folds consecutive samples of one leaf and keeps the parent links | Unit | pass | not run | not run | not run |
| the stack index > is absent without a window and stride-samples past the cap | Unit | pass | not run | not run | not run |
| values, not just functions > ranges each numeric attribute and fits the time against it | Unit | pass | not run | not run | not run |
| values, not just functions > skips a constant value, an open span, and no spans | Unit | pass | not run | not run | not run |
packages/devtools/test/profiler-crypto.test.ts | |||||
| .ogp codec (Brotli + AES-GCM) > a file that unpacks past the limit is refused while it unpacks, with devtools_profiler_upload_too_large | Unit | pass | not run | not run | not run |
| .ogp codec (Brotli + AES-GCM) > dev fallback: encodes/decodes with no secret | Unit | pass | not run | not run | not run |
| .ogp codec (Brotli + AES-GCM) > each export has its own salt + iv (ciphertext differs for the same input) | Unit | pass | not run | not run | not run |
| .ogp codec (Brotli + AES-GCM) > fails on the wrong key (GCM tag rejects it) | Unit | pass | not run | not run | not run |
| .ogp codec (Brotli + AES-GCM) > is_ogp rejects non-.ogp bytes (plain JSON, random) | Unit | pass | not run | not run | not run |
| .ogp codec (Brotli + AES-GCM) > opens in ANY profiler that has the same key (cross-instance import) | Unit | pass | not run | not run | not run |
| .ogp codec (Brotli + AES-GCM) > round-trips with the right key | Unit | pass | not run | not run | not run |
packages/devtools/test/profiler-deep.test.ts | |||||
| as if the profiler were not there > a pause is charged as the app would have paid it: the profiler’s own share of the allocations before it is taken out | Unit | pass | not run | not run | not run |
| byte strip > a byte arrives in proportion to its place in the download | Unit | pass | not run | not run | not run |
| byte strip > a document without ogygia at all, or a broken one, still strips | Unit | pass | not run | not run | not run |
| byte strip > counts the elements as the browser makes them: none from a quoted `<`, a comment ends at `<!-->`, a nested template is one (#1370)#1370 | Unit | pass | not run | not run | not run |
| byte strip > counts the elements the markup makes: the document and each island; text, comments and templates are not elements | Unit | pass | not run | not run | not run |
| byte strip > cuts the document into ordered, gap-free segments by what each byte is | Unit | pass | not run | not run | not run |
| byte strip > names each <img> with no width or height by its src; sized, hidden, scripted and data-width ones are not | Unit | pass | not run | not run | not run |
| byte strip > reads the page seed’s text the way the byte strip finds the seed script | Unit | pass | not run | not run | not run |
| byte strip > reads the tags as a browser does: a quoted `>`, a comment, a value naming a marker, a data-src | Unit | pass | not run | not run | not run |
| data river > build_river: calls feed loads, loads feed keys, keys feed islands; waste and notes | Unit | pass | not run | not run | not run |
| data river > load_return_keys reads the object literals a load returns; spreads and non-literals are unknown | Unit | pass | not run | not run | not run |
| data river > merge_visits folds two records of one visit: arrays union, the later navigation and paints win | Unit | pass | not run | not run | not run |
| data river > seed_key_bytes measures each top-level key, unwrapping the seed shapes | Unit | pass | not run | not run | not run |
| dom diff + shift boxes > shift boxes normalize to the viewport, biggest first, skipping shifts with no rect | Unit | pass | not run | not run | not run |
| dom diff + shift boxes > tokenizes a tag whole whatever its values hold, in any quoting | Unit | pass | not run | not run | not run |
| dom diff + shift boxes > tokenizes tags and text, diffs by token, and reports what changed | Unit | pass | not run | not run | not run |
| more from the same snapshot > deopts: the reasons V8 wrote into the profile, per function, with self time | Unit | pass | not run | not run | not run |
| more from the same snapshot > simulate_awaits: parallel starts a call with the one it waited for, cache makes it instant, remove takes it out | Unit | pass | not run | not run | not run |
| more from the same snapshot > sync_io names the blocking node calls with their app callers; memo candidates need many calls at a steady cost | Unit | pass | not run | not run | not run |
| sink > buffers rows, drops the oldest request rows first when full, drains NDJSON that parses back | Unit | pass | not run | not run | not run |
| sink > cloud points, the layer cake per route, and the hot functions of a brushed range | Unit | pass | not run | not run | not run |
| visit + one clock > one_clock: a lane per hole, its answer cut as the devtools waterfall cuts it | Unit | pass | not run | not run | not run |
| visit + one clock > one_clock: the server render ends at the first byte, lanes per resource type, islands, marks, and a critical path to first paint and LCP | Unit | pass | not run | not run | not run |
| visit + one clock > parse_visit bounds and drops junk; rejects what is not a visit | Unit | pass | not run | not run | not run |
| who caused the GC > attribute_gc charges each pause to what was allocated since the previous one, and each maker its share of pause time | Unit | pass | not run | not run | not run |
| who caused the GC > gc_kind reads Node’s kind bits | Unit | pass | not run | not run | not run |
| who caused the GC > gc_running takes the whole capture CPU first, else the window CPU segments shifted onto the capture clock | Unit | pass | not run | not run | not run |
| who caused the GC > heap_sites keys each allocating stack, names the nearest component; slice_delta is the growth between reads | Unit | pass | not run | not run | not run |
packages/devtools/test/profiler-dev-tools.test.ts | |||||
| a dev tool’s origin, by its package | Unit | pass | not run | not run | not run |
| mostly waiting is read off one render, not the window the profiler idled in | Unit | pass | not run | not run | not run |
| the app’s lines and the libraries a build keeps are not | Unit | pass | not run | not run | not run |
packages/devtools/test/profiler-dev.test.ts | |||||
| the dev server > a `.ts` module's map comes from the dev server (any extension), shifted by the runner's own lines | Unit | pass | not run | not run | not run |
| the dev server > a call site's file: its sourceURL when it has no file name (code made from a string) | Unit | pass | not run | not run | not run |
| the dev server > a heap sample of a module made from a string has no url: its script id names it | Unit | pass | not run | not run | not run |
| the dev server > a pre-bundled dependency is its package: Svelte's own runtime is Svelte, not `.vite` | Unit | pass | not run | not run | not run |
packages/devtools/test/profiler-doc-diff.test.ts | |||||
| doc_diff > an attribute stamped per render (one group, whatever tags), an island (region and props as one), comment markers | Unit | pass | not run | not run | not run |
| doc_diff > past a small insertion, the rest still lines up | Unit | pass | not run | not run | not run |
| doc_diff > the review cases: a dropped piece, an added one at the end, a dropped attribute, single quotes, a whole attribute name | Unit | pass | not run | not run | not run |
| doc_diff > the same document: nothing | Unit | pass | not run | not run | not run |
| doc_diff > where your markup sets a changing attribute, on the tags the diff saw it on | Unit | pass | not run | not run | not run |
| doc_diff reads tags as the browser does > a comment holding a tag changes as a comment marker | Unit | pass | not run | not run | not run |
| doc_diff reads tags as the browser does > a script whose text holds `<` changes as the script’s text | Unit | pass | not run | not run | not run |
| doc_diff reads tags as the browser does > an attribute: <a data-track='1'> | Unit | pass | not run | not run | not run |
| doc_diff reads tags as the browser does > an attribute: <a data-track="1"> | Unit | pass | not run | not run | not run |
| doc_diff reads tags as the browser does > an attribute: <a data-track=1> | Unit | pass | not run | not run | not run |
| doc_diff reads tags as the browser does > an attribute: <a title='a<b' data-track='1'> | Unit | pass | not run | not run | not run |
| doc_diff reads tags as the browser does > an attribute: <a title='a<b' data-track="1"> | Unit | pass | not run | not run | not run |
| doc_diff reads tags as the browser does > an attribute: <a title='a<b' data-track=1> | Unit | pass | not run | not run | not run |
| doc_diff reads tags as the browser does > an attribute: <a title=""=>" data-track='1'> | Unit | pass | not run | not run | not run |
| doc_diff reads tags as the browser does > an attribute: <a title=""=>" data-track="1"> | Unit | pass | not run | not run | not run |
| doc_diff reads tags as the browser does > an attribute: <a title=""=>" data-track=1> | Unit | pass | not run | not run | not run |
| doc_diff reads tags as the browser does > an attribute: <a title="say 'hi'" data-track='1'> | Unit | pass | not run | not run | not run |
| doc_diff reads tags as the browser does > an attribute: <a title="say 'hi'" data-track="1"> | Unit | pass | not run | not run | not run |
| doc_diff reads tags as the browser does > an attribute: <a title="say 'hi'" data-track=1> | Unit | pass | not run | not run | not run |
| doc_diff reads tags as the browser does > an attribute: <a title="x>y" data-track='1'> | Unit | pass | not run | not run | not run |
| doc_diff reads tags as the browser does > an attribute: <a title="x>y" data-track="1"> | Unit | pass | not run | not run | not run |
| doc_diff reads tags as the browser does > an attribute: <a title="x>y" data-track=1> | Unit | pass | not run | not run | not run |
| doc_diff reads tags as the browser does > text after a `<` that opens no tag is the text of the tag before it | Unit | pass | not run | not run | not run |
packages/devtools/test/profiler-drill.test.ts | |||||
| build_drill > a $lib helper a load ran sits under that load file (its stack says so), not loose in the phase | Unit | pass | not run | not run | not run |
| build_drill > a component row carries its render count, and "renders per item" when that pattern names it | Unit | pass | not run | not run | not run |
| build_drill > a CPU owner that is one row carries its low–high over the renders | Unit | pass | not run | not run | not run |
| build_drill > a gap with a queued callback takes the line that queued it, and its load file | Unit | pass | not run | not run | not run |
| build_drill > a helper's line goes one level further: the lines of yours that called it | Unit | pass | not run | not run | not run |
| build_drill > a library roots in your line that calls it; your function in its lines | Unit | pass | not run | not run | not run |
| build_drill > a package called from one line of yours splits into the package's own functions | Unit | pass | not run | not run | not run |
| build_drill > a wait row carries its endpoint's waiting over the renders | Unit | pass | not run | not run | not run |
| build_drill > a wait row knows how long it held the render alone: with other endpoints in flight it did not | Unit | pass | not run | not run | not run |
| build_drill > adds up at every level, and leaves the profiler out | Unit | pass | not run | not run | not run |
| build_drill > an HTTP wait splits each call's own clock once: their side (Server-Timing, nested entries not double counted), the rest, the body | Unit | pass | not run | not run | not run |
| build_drill > an owner splits by what its samples ran inside it (its arrows, itself), each by its own lines | Unit | pass | not run | not run | not run |
| build_drill > garbage collection goes to the lines that made the garbage, by the pauses each caused | Unit | pass | not run | not run | not run |
| build_drill > tags each row with the patterns found on its line, however the path is spelled, and lifts them to its parents | Unit | pass | not run | not run | not run |
| build_drill > the load phase splits by load file; ids of one endpoint are one wait row, shared waits split, rooted at the calling line | Unit | pass | not run | not run | not run |
| build_drill > the rows an "N more" folds stay inside it, adding up to it | Unit | pass | not run | not run | not run |
| drill_deltas > a line renumbered by an edit above it is the same line, not one fixed and one new | Unit | pass | not run | not run | not run |
| drill_deltas > a moved line's parent is weighed against the pair: its own move is not lost | Unit | pass | not run | not run | not run |
| drill_deltas > a row folded into "N more" on one side is never called fixed: it is at most the fold; small moves are noise | Unit | pass | not run | not run | not run |
| drill_deltas > names the deepest row that moved, not every level above it | Unit | pass | not run | not run | not run |
| drill_deltas > with both sides' spreads over renders, a move is ranges that do not overlap | Unit | pass | not run | not run | not run |
| drill_deltas > with the code on both sides: the same code is one line (its own change kept); other code is a fix and a new cost | Unit | pass | not run | not run | not run |
| drill: the review cases > a line folded into "N more" still speaks for its row's keys; a CPU row reads its lines, not its function's start | Unit | pass | not run | not run | not run |
| drill: the review cases > identical calls from one line are each a call | Unit | pass | not run | not run | not run |
| drill: the review cases > v2 is a word, P12 an id | Unit | pass | not run | not run | not run |
| order_of > no warning across servers (a restart between them) | Unit | pass | not run | not run | not run |
| order_of > warns when the later report ran on the heap an earlier leaky one filled, on the same server | Unit | pass | not run | not run | not run |
| owner_of_stack > with no named frame of yours, it names the place: a module loading, or a nameless callback | Unit | pass | not run | not run | not run |
packages/devtools/test/profiler-fake-progress.test.ts | |||||
| a disposed progress bar stops ticking | Unit | pass | not run | not run | not run |
packages/devtools/test/profiler-forecast.test.ts | |||||
| forecast_of > a fix inside a call another fix moves out of the request saves nothing more: one group | Unit | pass | not run | not run | not run |
| forecast_of > a fix inside a function the deleted line calls (inlined onto it) joins the deletion | Unit | pass | not run | not run | not run |
| forecast_of > a line's CPU and its wait are different time: both count | Unit | pass | not run | not run | not run |
| forecast_of > a whole-document cache is a separate lever; waits are capped at the waiting; never the whole render | Unit | pass | not run | not run | not run |
| forecast_of > deleting one call of a chain that runs together anyway saves nothing more: one group | Unit | pass | not run | not run | not run |
| forecast_of > each resource is capped by what the render used: waiting (deleted waits too) and CPU | Unit | pass | not run | not run | not run |
| forecast_of > keeping the answers, measured: the render served them, less the code fixes the caching does not cover | Unit | pass | not run | not run | not run |
| forecast_of > patterns joined through a caller line (a helper) are one group | Unit | pass | not run | not run | not run |
| forecast_of > separate fixes add up; fixes on the same line count their biggest once | Unit | pass | not run | not run | not run |
| forecast_of > the same unread data in several rows is one deletion: its rows add up | Unit | pass | not run | not run | not run |
| forecast_of > two fixes with one title name their places (one title per fix) | Unit | pass | not run | not run | not run |
| forecast_of > two unread calls under one shared helper line are two deletions (both count); a fix on that line still joins them | Unit | pass | not run | not run | not run |
| forecast_of > work for data nothing reads is deleted, and a pattern inside it is moot | Unit | pass | not run | not run | not run |
packages/devtools/test/profiler-hole-slots.test.ts | |||||
| hole_slots > a failed render is no render time; a hole with no page row is named by the endpoint | Unit | pass | not run | not run | not run |
| hole_slots > names who held the slots the queued holes waited for, render counted without the wait | Unit | pass | not run | not run | not run |
| hole_slots > no hole requests: nothing to say | Unit | pass | not run | not run | not run |
| hole_slots > one request per instance (nothing queued): the table without a summary | Unit | pass | not run | not run | not run |
| hole_slots > several copies of one hole (different props): the component alone | Unit | pass | not run | not run | not run |
packages/devtools/test/profiler-hooks-phase.test.ts | |||||
| handle_parts: the handle functions a hooks file borrows from other modules > reads sequence(...) and follows the imports to their files | Unit | pass | not run | not run | not run |
| handle_parts: the handle functions a hooks file borrows from other modules > takes a handle that is one imported function | Unit | pass | not run | not run | not run |
| handle_parts: the handle functions a hooks file borrows from other modules > the phase: that function in its own file reads as hooks, the same name elsewhere does not | Unit | pass | not run | not run | not run |
packages/devtools/test/profiler-hot-line.test.ts | |||||
| a leaf’s own loop: the line, no callee (Math.max, new Date, a method are its own work) | Unit | pass | not run | not run | not run |
| a load blamed for an inlined callee: the line, and the call to look inside | Unit | pass | not run | not run | not run |
| no single line holds the time, or it is the first line, or no source: nothing said | Unit | pass | not run | not run | not run |
packages/devtools/test/profiler-hydration-hazards.test.ts | |||||
| hydration_hazards: the lines that draw differently in the browser > a top-level `if (typeof window …)` is a guard: a lead, marked, not a prediction | Unit | pass | not run | not run | not run |
| hydration_hazards: the lines that draw differently in the browser > a top-level await, a browser value in a derived and in the markup | Unit | pass | not run | not run | not run |
| hydration_hazards: the lines that draw differently in the browser > what runs after the wake is never one: callbacks, effects, handlers, the module script | Unit | pass | not run | not run | not run |
| the report says them > each line by file and line, with what it does; the fix per kind | Unit | pass | not run | not run | not run |
packages/devtools/test/profiler-island-files.test.ts | |||||
| island_files_diff > a build from before content hashing (the file is the identity) never counts as moved | Unit | pass | not run | not run | not run |
| island_files_diff > an island new in the second build, or copies of one island, count once or not at all | Unit | pass | not run | not run | not run |
| island_files_diff > every island moved (three or more): the build’s names are not stable | Unit | pass | not run | not run | not run |
| island_files_diff > moved and kept, the moved files weighed from the second build | Unit | pass | not run | not run | not run |
| island_files_diff > no islands in common: nothing to say | Unit | pass | not run | not run | not run |
packages/devtools/test/profiler-key-sources.test.ts | |||||
| constant_work > a helper of the file that reads the disk, a clock or a random, or changes the file's state, answers per call | Unit | pass | not run | not run | not run |
| constant_work > a per-call module's value (node:crypto, $app/server) is never the same every request | Unit | pass | not run | not run | not run |
| constant_work > a statement wrapped over several lines is read whole | Unit | pass | not run | not run | not run |
| constant_work > a value the load changes afterwards is its own, and so is what it was made from | Unit | pass | not run | not run | not run |
| constant_work > a value written straight into the return counts too; an object key is not a read | Unit | pass | not run | not run | not run |
| constant_work > hoist_plan: no move when a top-level binding comes after the load, or a moved name clashes | Unit | pass | not run | not run | not run |
| constant_work > hoist_plan: the costly lines and what they read, placed above the load | Unit | pass | not run | not run | not run |
| constant_work > load lines that read only the module: computed the same on every request | Unit | pass | not run | not run | not run |
| constant_work > review fixes: a per-call name imported over several lines; a helper with no braces; a load name that shadows the module | Unit | pass | not run | not run | not run |
| each_blocks > the loops in markup, the key each walks, and the components inside (nested too) | Unit | pass | not run | not run | not run |
| formatter_rewrite > a component: the addition goes in <script module>, since the instance script runs per render | Unit | pass | not run | not run | not run |
| formatter_rewrite > arguments that read variables: a cached factory, called in place of `new Intl.X(` | Unit | pass | not run | not run | not run |
| formatter_rewrite > every match on the line; none inside a string or a comment | Unit | pass | not run | not run | not run |
| formatter_rewrite > factory types: a locale list type, required options where Intl requires them; no factory for Locale | Unit | pass | not run | not run | not run |
| formatter_rewrite > literal arguments: one formatter at the top, used on the line | Unit | pass | not run | not run | not run |
| formatter_rewrite > localeCompare with literal locale and options: one collator | Unit | pass | not run | not run | not run |
| formatter_rewrite > names: never one the file uses; different arguments get different names; the same is reused | Unit | pass | not run | not run | not run |
| formatter_rewrite > the receiver is the member chain only; a comma inside a call is not an argument break | Unit | pass | not run | not run | not run |
| key_sources > a callback’s parameter is its own: `(id) => …` in a key does not tie it to the loop’s `id` | Unit | pass | not run | not run | not run |
| key_sources > a local in a callback is that callback’s own: two `res` do not tie session to flags | Unit | pass | not run | not run | not run |
| key_sources > a typed declaration, then filled by index or by push | Unit | pass | not run | not run | not run |
| key_sources > follows a key back through the variables it reads to the lines that make them | Unit | pass | not run | not run | not run |
| key_sources > only the load’s return, not a helper’s after it | Unit | pass | not run | not run | not run |
| late_island_keys > keys used only as props of islands that wake late | Unit | pass | not run | not run | not run |
| line_rw > what a line writes and what it reads | Unit | pass | not run | not run | not run |
| load_inputs > what a load takes from the request | Unit | pass | not run | not run | not run |
| parent_use > what the page took from parent(), and the first line that reads it | Unit | pass | not run | not run | not run |
| promise_all_rewrite > a trailing comment, a let, and names that only look like verbs are fine | Unit | pass | not run | not run | not run |
| promise_all_rewrite > an await spread over several lines: no rewrite | Unit | pass | not run | not run | not run |
| promise_all_rewrite > another statement between them: no rewrite (it could feed a later call) | Unit | pass | not run | not run | not run |
| promise_all_rewrite > any other shape of line: no rewrite | Unit | pass | not run | not run | not run |
| promise_all_rewrite > plain one-line awaits next to each other (comments between are fine) become one Promise.all | Unit | pass | not run | not run | not run |
| source reads: the review cases > a `let` declared with no value, then assigned in branches | Unit | pass | not run | not run | not run |
| source reads: the review cases > a helper's local is not a module value: a line reading the request stays per request | Unit | pass | not run | not run | not run |
| source reads: the review cases > a loop adding up its answers can run its calls together | Unit | pass | not run | not run | not run |
| source reads: the review cases > a prop on the second line of a multi-line island tag is the island’s | Unit | pass | not run | not run | not run |
| source reads: the review cases > a return type's braces are the type's, not the body | Unit | pass | not run | not run | not run |
| source reads: the review cases > page.data.x is the page’s data; post.data.x is not | Unit | pass | not run | not run | not run |
| tag_fills > a helper inlined into the load: the owner mixes lines, and the line row still says what it fills | Unit | pass | not run | not run | not run |
| tag_fills > a helper's CPU under a load file, with no load line of its own, goes by the name the key's lines use | Unit | pass | not run | not run | not run |
| tag_fills > a wait whose calling line feeds only unread keys says so | Unit | pass | not run | not run | not run |
packages/devtools/test/profiler-key-web-crypto.test.ts | |||||
| with no node:crypto, login seats a session, the session opens the profiler, a wrong key is refused | Unit | pass | not run | not run | not run |
packages/devtools/test/profiler-kit-hash.test.ts | |||||
| Kit's ETag hash in a bundled chunk > an app’s own `hash` (any other caller) stays the app’s | Unit | pass | not run | not run | not run |
| Kit's ETag hash in a bundled chunk > under render_response: Kit’s, not the app’s | Unit | pass | not run | not run | not run |
packages/devtools/test/profiler-ledger.test.ts | |||||
| grows_between > a check cut to 5 renders never calls a 4-entry cache a leak (the slow-patterns flake) | Unit | pass | not run | not run | not run |
| grows_between > confirming takes the larger per-render estimate: a low one-render draw cannot make a cache a leak | Unit | pass | not run | not run | not run |
| grows_between > halfway against the end of one sampler: a leak doubles over 3→6, a 4-entry cache does not | Unit | pass | not run | not run | not run |
| line ledger > a library entry V8 inlined away: the inlining-off render says which app line entered the package | Unit | pass | not run | not run | not run |
| line ledger > a map with no mappings (a plugin broke the chain) still resolves by the line’s own text, inside its region | Unit | pass | not run | not run | not run |
| line ledger > core’s own source, under either package name, is no app path | Unit | pass | not run | not run | not run |
| line ledger > joins CPU, allocations, GC and retained bytes on one app line, with its code | Unit | pass | not run | not run | not run |
| line ledger > library time lands on the app line that calls it, through an import alias, split over repeat calls | Unit | pass | not run | not run | not run |
| line ledger > memory a library made after an await (no app caller) goes to the app line that calls that library | Unit | pass | not run | not run | not run |
| line ledger > memory charged to a function’s first line joins that function’s hottest line | Unit | pass | not run | not run | not run |
| line ledger > merged calls split back: an inlined function’s time on the next line goes home, by the un-inlined render | Unit | pass | not run | not run | not run |
| line ledger > ogygia’s own source through a workspace link is the library, not a line the app edits | Unit | pass | not run | not run | not run |
| line ledger > short paths | Unit | pass | not run | not run | not run |
| line ledger > skips makers with no app line and caps the list | Unit | pass | not run | not run | not run |
| line ledger > the framework’s own calls (Kit’s json / resolve) are never library time on a line | Unit | pass | not run | not run | not run |
| line ledger > the resolver reads a source from the map’s embedded copy when the file is not on this machine | Unit | pass | not run | not run | not run |
| per-line growth > a line holding about N renders’ worth after N renders grows; one holding about one render’s is flat | Unit | pass | not run | not run | not run |
packages/devtools/test/profiler-left-each-run.test.ts | |||||
| one render proves nothing | Unit | pass | not run | not run | not run |
| the same line left open in every run: marked; one started once, a short or unref’d timeout: not | Unit | pass | not run | not run | not run |
packages/devtools/test/profiler-login-guard.test.ts | |||||
| profiler login: the secret > a production secret shorter than 16 bytes opens nothing, and says why | Unit | pass | not run | not run | not run |
| profiler login: the session > a session ends after its lifetime | Unit | pass | not run | not run | not run |
| profiler login: the session > a session signed under another secret is refused | Unit | pass | not run | not run | not run |
| profiler login: the session > a session with any part changed is refused | Unit | pass | not run | not run | not run |
| profiler login: the session > each login is a new random session, not a value the secret alone decides | Unit | pass | not run | not run | not run |
| profiler login: the session > logout ends the session it carried; another session stays | Unit | pass | not run | not run | not run |
| profiler login: where it goes after > a POST without a session goes to the login with the dashboard as next, a GET keeps its page | Unit | pass | not run | not run | not run |
| profiler login: wrong keys per address > after ten wrong keys an address is refused, even with the right key; another address is not | Unit | pass | not run | not run | not run |
| profiler login: wrong keys per address > the right key spends nothing: a user who logs in often is never refused | Unit | pass | not run | not run | not run |
| profiler login: wrong keys per address > the x-profiler-key header shares the count | Unit | pass | not run | not run | not run |
packages/devtools/test/profiler-maps-browser.test.ts | |||||
| reads no profiler maps in the browser | Unit | pass | not run | not run | not run |
packages/devtools/test/profiler-measures.test.ts | |||||
| profiler: measures > a flag of 64 characters but more bytes (a decoded %C3%A9 each) is refused, not thrown | Unit | pass | not run | not run | not run |
| profiler: measures > a flag signed under another secret is not | Unit | pass | not run | not run | not run |
| profiler: measures > a flag with any part changed is not | Unit | pass | not run | not run | not run |
| profiler: measures > a logged-in browser measures with the site-wide flag alone (its session stays on the profiler path) | Unit | pass | not run | not run | not run |
| profiler: measures > logout ends the flag with its session; another session’s flag stays | Unit | pass | not run | not run | not run |
| profiler: measures > the beacon flag alone is not the profiler’s user | Unit | pass | not run | not run | not run |
| profiler: measures > the flag ends when its session would have | Unit | pass | not run | not run | not run |
| profiler: measures > the old flag, one value per secret for every browser, is not | Unit | pass | not run | not run | not run |
packages/devtools/test/profiler-mount-once.test.ts | |||||
| the profiler mount > a mount that fails is said once, with its error | Unit | pass | not run | not run | not run |
| the profiler mount > a mount whose import failed is tried again by the next request | Unit | pass | not run | not run | not run |
| the profiler mount > two first requests at once mount one profiler | Unit | pass | not run | not run | not run |
packages/devtools/test/profiler-net-capture.test.ts | |||||
| install_net_capture > after the hooks run again, a call in the new profiler’s request reaches the new profiler (#285)#285 | Unit | pass | not run | not run | not run |
| install_net_capture > the copy of an endless body it counts stops being read when its request ends | Unit | pass | not run | not run | not run |
packages/devtools/test/profiler-ogp-recover.test.ts | |||||
| recover_ogp_bytes > decodes a base64-encoded file (round-trips byte-for-byte) | Unit | pass | not run | not run | not run |
| recover_ogp_bytes > decodes base64 with wrapping newlines | Unit | pass | not run | not run | not run |
| recover_ogp_bytes > leaves genuine garbage alone (import still rejects it) | Unit | pass | not run | not run | not run |
| recover_ogp_bytes > passes a real .ogp through unchanged | Unit | pass | not run | not run | not run |
| recover_ogp_bytes > strips a prepended UTF-8 BOM | Unit | pass | not run | not run | not run |
packages/devtools/test/profiler-on-demand.test.ts | |||||
| profiler_demand_skip > demand-only + not mounted + a normal page request → SKIP (profiler never imported) | Unit | pass | not run | not run | not run |
| profiler_demand_skip > demand-only + not mounted + a request TO the profiler path → do NOT skip (mount it now) | Unit | pass | not run | not run | not run |
| profiler_demand_skip > honours a custom profiler path | Unit | pass | not run | not run | not run |
| profiler_demand_skip > not demand-only (default) → never skip, profiler behaves as today | Unit | pass | not run | not run | not run |
| profiler_demand_skip > once mounted, never skip again — later requests are instrumented like normal mode | Unit | pass | not run | not run | not run |
packages/devtools/test/profiler-other-requests.test.ts | |||||
| another request in the window > its route's code, and everything under it, is set aside; the profiled route's, its layouts' and shared code stay | Unit | pass | not run | not run | not run |
| another visitor on the same page > the samples of a render they overlapped are set aside by time, the clean render counts | Unit | pass | not run | not run | not run |
| route files > a page or layout file, by its folder, in source or built form | Unit | pass | not run | not run | not run |
| route files > a route's own folder and the ones above it (their layouts are its) | Unit | pass | not run | not run | not run |
| route files > another route's page file is another visitor's; this route's, its layouts, endpoints and shared code never | Unit | pass | not run | not run | not run |
packages/devtools/test/profiler-page-beacon-parity.test.ts | |||||
| attr_changes() is core's | Unit | pass | not run | not run | not run |
| attr_changes() is core's over serialized markup | Unit | pass | not run | not run | not run |
| bare() is core's without_runtime_marks(without_comments()) | Unit | pass | not run | not run | not run |
| copy_of() is core's hole_copy_of | Unit | pass | not run | not run | not run |
packages/devtools/test/profiler-parallelizable.test.ts | |||||
| long waits with a few ms of CPU between: one chain, every run | Unit | pass | not run | not run | not run |
| short waits keep the strict line: 8 ms of CPU between 30 ms waits is work between them | Unit | pass | not run | not run | not run |
| two calls of the same label (one route, two queries) are two links, not one call continuing | Unit | pass | not run | not run | not run |
packages/devtools/test/profiler-patterns.test.ts | |||||
| a whole function inlined into a line that runs once > its cost on the caller line is read inside the callee, whose own loop gives the context | Unit | pass | not run | not run | not run |
| audit false positives (normal pages) > a Date from a number is not date parsing; a Date from text is | Unit | pass | not run | not run | not run |
| audit false positives (normal pages) > a Date parsed from text and formatted on one line: the formatting is the time, not the parse | Unit | pass | not run | not run | not run |
| audit false positives (normal pages) > a table built from the call’s own values is not a fixed table; one built from its own index is | Unit | pass | not run | not run | not run |
| audit false positives (normal pages) > a template literal’s ${} parts are tokenized | Unit | pass | not run | not run | not run |
| cache_rules > a same-answer site carries what the service said about keeping it | Unit | pass | not run | not run | not run |
| cache_rules > one helper line fetching several endpoints: requests from lines, not places | Unit | pass | not run | not run | not run |
| cache_rules > what a Cache-Control allows: shared max-age wins, no-store, private | Unit | pass | not run | not run | not run |
| compare: an unchanged render makes CPU moves "shifted" > within the runs’ spread, fixed/worse CPU rows become shifted; waits and kept memory keep their verdicts | Unit | pass | not run | not run | not run |
| find_cold_caches > a cache that answers most calls, a helper called from elsewhere too, or a cheap memo: quiet | Unit | pass | not run | not run | not run |
| find_cold_caches > the miss path’s helper ran as often as the memo: the cache never hits | Unit | pass | not run | not run | not run |
| find_patterns > a build without sourcemaps: a compiled component line builds its table per render; a split spread is one statement | Unit | pass | not run | not run | not run |
| find_patterns > a fixed table built in a component is flagged, but not in <script module> | Unit | pass | not run | not run | not run |
| find_patterns > a formatter built only when a cache misses is the fix, not the problem (the written change has that shape) | Unit | pass | not run | not run | not run |
| find_patterns > a library called once per item is named after the library function | Unit | pass | not run | not run | not run |
| find_patterns > a locale method counts only with a locale or options: with none V8 reuses a cached formatter | Unit | pass | not run | not run | not run |
| find_patterns > a search after a loop header on the same line counts; the header call itself does not | Unit | pass | not run | not run | not run |
| find_patterns > a search inside a loop and a spread accumulator are named | Unit | pass | not run | not run | not run |
| find_patterns > a table sized by the input is not a fixed table; a timer awaited per turn is its own pattern | Unit | pass | not run | not run | not run |
| find_patterns > a thin wrapper’s site lists the caller lines that call it, through an instrument() alias, in a loop | Unit | pass | not run | not run | not run |
| find_patterns > a true pattern worth under a millisecond is not reported | Unit | pass | not run | not run | not run |
| find_patterns > copied on each step means the ACCUMULATOR spread: an item or call-argument spread is not it | Unit | pass | not run | not run | not run |
| find_patterns > groups one pattern in several places into one item | Unit | pass | not run | not run | not run |
| find_patterns > library calls per item group per library function | Unit | pass | not run | not run | not run |
| find_patterns > names the rescan-per-item loop and the per-call formatter, with numbers, biggest win first | Unit | pass | not run | not run | not run |
| find_patterns > Node's own functions are the runtime, not a library called per item | Unit | pass | not run | not run | not run |
| find_patterns > stays quiet where the shape is there but the context or the cost is not | Unit | pass | not run | not run | not run |
| find_patterns > string growth needs text added: a literal, a call answering text, or a name made text above; not a sum | Unit | pass | not run | not run | not run |
| find_patterns > the USE of a formatter a cache built above is not 'the first use of one built here' | Unit | pass | not run | not run | not run |
| find_wait_patterns > a loop whose turn feeds the next (a cursor) cannot run its calls together | Unit | pass | not run | not run | not run |
| find_wait_patterns > a Promise.all batch right after a chain is not a link in it | Unit | pass | not run | not run | not run |
| find_wait_patterns > calls from one line that ran one after another: the waits add up, the saving is all but the longest | Unit | pass | not run | not run | not run |
| find_wait_patterns > calls one after another from ONE line that runs no loop are nested, not a missed Promise.all | Unit | pass | not run | not run | not run |
| find_wait_patterns > calls that overlapped (already together), or from different pages, are no chain | Unit | pass | not run | not run | not run |
| find_wait_patterns > calls whose each asks with the answer before it are the program, not a chain — independent ones are | Unit | pass | not run | not run | not run |
| find_wait_patterns > one slow call holding a batch: named with the gap it costs; an even batch is quiet | Unit | pass | not run | not run | not run |
| find_wait_patterns > review: a call from another line starting alongside is not this chain’s batch; a straggler needs a Promise.all | Unit | pass | not run | not run | not run |
| find_wait_patterns > two awaits of a helper written on two lines are a chain; its callers show as where to fix | Unit | pass | not run | not run | not run |
| findings link to the pattern that explains them > by a shared line, a caller line, the row they open, or the calls they name | Unit | pass | not run | not run | not run |
| fix_impact > adds CPU and waiting savings against the render, capped at 90 % | Unit | pass | not run | not run | not run |
| fix_impact caps the wait fixes at the render’s waiting > caching and starting together overlap: together they never take off more than was waited | Unit | pass | not run | not run | not run |
| fix_impact units > CPU savings add up every profiled render, so they are divided per render; waits are one render | Unit | pass | not run | not run | not run |
| heap growth check > a leak past doubt after three renders: steady steps of 16 MB or more; a smaller leak or a cache slowing down, never | Unit | pass | not run | not run | not run |
| heap growth check > a steady climb is a leak; a climb that stops is a bounded cache | Unit | pass | not run | not run | not run |
| heap growth check > flat after three renders: the check may stop (a leak, or a cache still filling, never) | Unit | pass | not run | not run | not run |
| heap growth check > the memory card says which, and a plateau is neither severe nor counted down | Unit | pass | not run | not run | not run |
| in_loop > sees for / while / callback bodies, a same-line loop header, and {#each} | Unit | pass | not run | not run | not run |
| inferno round: context fixes > `<script module>` in a comment is not a module script | Unit | pass | not run | not run | not run |
| inferno round: context fixes > a callback defined on the costly line is read in the function around it | Unit | pass | not run | not run | not run |
| inferno round: context fixes > a dedupe by indexOf inside filter is a list search, not a string rescan | Unit | pass | not run | not run | not run |
| inferno round: context fixes > a formatter built on one line and used (and charged) on the next | Unit | pass | not run | not run | not run |
| inferno round: context fixes > a line with no loop of its own, called from a caller’s loop, runs per item | Unit | pass | not run | not run | not run |
| inferno round: context fixes > a RegExp built on one line and charged to its first use on the next > both lines, either order: one site, both costs | Unit | pass | not run | not run | not run |
| inferno round: context fixes > a RegExp built on one line and charged to its first use on the next > passed to replace: not moved (making the new text is work that stays) | Unit | pass | not run | not run | not run |
| inferno round: context fixes > a RegExp built on one line and charged to its first use on the next > the use alone: named, at the build line | Unit | pass | not run | not run | not run |
| inferno round: context fixes > walking entries to find one key | Unit | pass | not run | not run | not run |
| kept-per-render > a line holding a sliver of the total is a passenger: left out of the places, counted in the text | Unit | pass | not run | not run | not run |
| kept-per-render > lines that leave memory alive after a warm render make one memory card, beside any CPU card for the same line | Unit | pass | not run | not run | not run |
| kept-per-render > memory made inside a library the line calls: the card says the library keeps it, and gives its fixes | Unit | pass | not run | not run | not run |
| kept-per-render > under a quarter megabyte in all is no card | Unit | pass | not run | not run | not run |
| kept-per-render countdown > on a serverless function the ceiling is its memory size, and what happens there is said | Unit | pass | not run | not run | not run |
| kept-per-render countdown > says how many more requests fill the heap at this rate | Unit | pass | not run | not run | not run |
| line_deltas > matches lines by their code (an edit above moves the number), per render; a changed line is fixed + new | Unit | pass | not run | not run | not run |
| pattern_deltas > fix check with the older report's forecast: its parts that got done, counted its way | Unit | pass | not run | not run | not run |
| pattern_deltas > fix check: wait patterns over the same calls promise their largest move, not their sum; a whole-page cache promises nothing | Unit | pass | not run | not run | not run |
| pattern_deltas > fix check: what the fixed patterns promised against what the median render measured | Unit | pass | not run | not run | not run |
| pattern_deltas > per render: a CPU pattern over 3 renders against 6 is not "better" for having fewer renders; a wait is one render | Unit | pass | not run | not run | not run |
| pattern_deltas > review fixes: code-less sites give no verdict; unread work folded away in B is still there; a clean B has none; a covered fix is worth its own saving | Unit | pass | not run | not run | not run |
| pattern_deltas > says what a change fixed, brought in, and moved; library patterns match per library | Unit | pass | not run | not run | not run |
| pattern_deltas > still_there: the share of an older pattern still flagged, by its own sites, matched by code | Unit | pass | not run | not run | not run |
| repeat-request > the exact same read made again in one render; its callers are listed; writes and other pages are not repeats | Unit | pass | not run | not run | not run |
| rescan needs the rebuilt string > a short value cleaned per item is not a whole-string rescan; the accumulator scanned then reassigned is | Unit | pass | not run | not run | not run |
| same_document_pattern: what keeps a shared cache out is named first > a cookie on the answer | Unit | pass | not run | not run | not run |
| same_document_pattern: what keeps a shared cache out is named first > nothing in the way: the plain advice | Unit | pass | not run | not run | not run |
| same_document_pattern: what keeps a shared cache out is named first > private / no-store, and both | Unit | pass | not run | not run | not run |
| seed-whole-read > points at the line in each island that takes page.data whole, with what the seed ships | Unit | pass | not run | not run | not run |
| site context > each site carries two lines each side of it, when the source is there | Unit | pass | not run | not run | not run |
| sync I/O too quick to sample, from V8 call counts > a counted *Sync call inside a function the render ran is a site; start-up code, comments, helpers that never ran are not | Unit | pass | not run | not run | not run |
| sync I/O too quick to sample, from V8 call counts > no *Sync call counted in Node, or none in the route files: nothing | Unit | pass | not run | not run | not run |
| the same answer every render > find_same_answers: one place per kind of request and line, sized, the per-user ones flagged | Unit | pass | not run | not run | not run |
| the same answer every render > stable_answers: a GET in every run with one body fingerprint; a changed body, a missed run or a write is not | Unit | pass | not run | not run | not run |
| tokens > keeps identifiers and punctuation, drops strings, numbers and comments | Unit | pass | not run | not run | not run |
| waits on the critical path > a chain saves its sum minus its longest only when nothing as long runs beside it | Unit | pass | not run | not run | not run |
packages/devtools/test/profiler-permalink.test.ts | |||||
| is_webkit_capped > every iOS browser and desktop Safari are capped; Chromium and Firefox are not | Unit | pass | not run | not run | not run |
| permalink > a blob that is not an ogygia share link is refused by name | Unit | pass | not run | not run | not run |
| permalink > a share link is fragment-safe text (no +, / or =) and opens again | Unit | pass | not run | not run | not run |
| permalink > a share link made earlier still opens | Unit | pass | not run | not run | not run |
| permalink > a share link opens only with its password | Unit | pass | not run | not run | not run |
packages/devtools/test/profiler-pick-fields.test.ts | |||||
| pick_fields (a compact profile) > only the sections asked for, and what names the report | Unit | pass | not run | not run | not run |
| pick_fields (a compact profile) > whole when nothing is asked | Unit | pass | not run | not run | not run |
packages/devtools/test/profiler-pure-bits.test.ts | |||||
| app_asset_rel > never walks out of the client folder | Unit | pass | not run | not run | not run |
| contention > a request to the islands endpoint (its one path under the app's base) is one of the page's holes; a page whose path ends the same is not | Unit | pass | not run | not run | not run |
| contention > caps a request's CPU at its overlap with the window | Unit | pass | not run | not run | not run |
| contention > counts every request by kind, though the list stops at 40 | Unit | pass | not run | not run | not run |
| contention > the busy share counts two requests over the same time once | Unit | pass | not run | not run | not run |
| html_diff > a one-word change in a long text run is one word | Unit | pass | not run | not run | not run |
| range_hot > counts a sample that straddles the start by its overlap | Unit | pass | not run | not run | not run |
| refresh_target > a body over 2 KB is a page, not a stub | Unit | pass | not run | not run | not run |
| refresh_target > reads the content attribute on either side of http-equiv, quotes off | Unit | pass | not run | not run | not run |
| render_steps > folds the smallest steps into a neighbour, counted, until at most `max` | Unit | pass | not run | not run | not run |
| render_steps > joins touching CPU segments of one owner, and a gap is a wait named by what was pending | Unit | pass | not run | not run | not run |
packages/devtools/test/profiler-recorder.test.ts | |||||
| the recorder has one owner > a page whose body breaks off mid-render is a 500 that says profiling failed, and why | Unit | pass | not run | not run | not run |
| the recorder has one owner > a recording wedged past the hard cap is cut: the site's requests stop paying for its context | Unit | pass | not run | not run | not run |
| the recorder has one owner > a reset during a page profile: that profile ending leaves the window started after the reset recording | Unit | pass | not run | not run | not run |
| the recorder has one owner > a reset during a window clears it: a page profile after it attributes each of its renders | Unit | pass | not run | not run | not run |
| the recorder has one owner > a second stop of a window that is still finishing its report never ends the next window | Unit | pass | not run | not run | not run |
| the recorder has one owner > two profilers in one process (a dev server that re-ran the hooks): the first window to end keeps the other's spans | Unit | pass | not run | not run | not run |
| the recorder has one owner > two profilers in one process: the first page profile to end keeps the other's call stacks on | Unit | pass | not run | not run | not run |
packages/devtools/test/profiler-recursive-total.test.ts | |||||
| a recursive function in one 10 ms sample > a component that renders itself: the component shows 10 ms in every component view | Unit | pass | not run | not run | not run |
| a recursive function in one 10 ms sample > a recursive caller of two hot functions: its path group shows it at 20 ms over two 10 ms samples, never 60 | Unit | pass | not run | not run | not run |
| a recursive function in one 10 ms sample > f calls f calls f: every view that groups by function shows f at 10 ms, never 30 | Unit | pass | not run | not run | not run |
| a recursive function in one 10 ms sample > f calls g calls f (mutual recursion): f and g each show 10 ms | Unit | pass | not run | not run | not run |
| a recursive function in one 10 ms sample > range_hot gives each of its three frames the 10 ms, and the range 10 ms of cpu | Unit | pass | not run | not run | not run |
packages/devtools/test/profiler-redis-connect.test.ts | |||||
| a Redis store whose node-redis cannot connect > fails with node-redis's own error, not a missing ioredis | Unit | pass | not run | not run | not run |
packages/devtools/test/profiler-redis-fallback.test.ts | |||||
| a Redis store with no node-redis installed > connects through ioredis | Unit | pass | not run | not run | not run |
packages/devtools/test/profiler-redis-store.test.ts | |||||
| redis store > io: an expired report's id leaves the index when the list is read | Unit | pass | not run | not run | not run |
| redis store > io: prune keeps the newest, and visits are capped per page | Unit | pass | not run | not run | not run |
| redis store > io: the prefix names every key, and a TTL write is one SET | Unit | pass | not run | not run | not run |
| redis store > node: an expired report's id leaves the index when the list is read | Unit | pass | not run | not run | not run |
| redis store > node: prune keeps the newest, and visits are capped per page | Unit | pass | not run | not run | not run |
| redis store > node: the prefix names every key, and a TTL write is one SET | Unit | pass | not run | not run | not run |
packages/devtools/test/profiler-refresh-stub.test.ts | |||||
| a `>` in the URL is part of it; a refresh in a script or a comment is no tag | Unit | pass | not run | not run | not run |
| a stub in any spelling the browser reads: quotes, no quotes, spaces, a > in a value | Unit | pass | not run | not run | not run |
| not a stub: no refresh, another origin, no url, a real page | Unit | pass | not run | not run | not run |
| the build’s redirect stub: the path it sends the visitor to | Unit | pass | not run | not run | not run |
packages/devtools/test/profiler-report-gzip-cap.test.ts | |||||
| a gzipped report that inflates past the cap is refused with 413, not inflated whole | Unit | pass | not run | not run | not run |
packages/devtools/test/profiler-routes.test.ts | |||||
| the profiler routes > a report file is GET by id from the store, or POST with the report in the body on any instance | Unit | pass | not run | not run | not run |
| the profiler routes > a report the store no longer holds is a thrown 404 | Unit | pass | not run | not run | not run |
| the profiler routes > every route is behind the guard: a deny answers for the beacon and the recording window | Unit | pass | not run | not run | not run |
| the profiler routes > the bare report page gives server data only to a logged-in visitor | Unit | pass | not run | not run | not run |
| the profiler routes that change state > /logout: a GET changes nothing, a POST does | Unit | pass | not run | not run | not run |
| the profiler routes that change state > /page: a GET changes nothing, a POST does | Unit | pass | not run | not run | not run |
| the profiler routes that change state > /replay/r1: a GET changes nothing, a POST does | Unit | pass | not run | not run | not run |
| the profiler routes that change state > /reset: a GET changes nothing, a POST does | Unit | pass | not run | not run | not run |
packages/devtools/test/profiler-session.test.ts | |||||
| describe_cookie_diagnosis > tells a dropped cookie, a duplicate, a wrong secret and no header apart | Unit | pass | not run | not run | not run |
| raw_cookie_values > decodes what decodes, keeps the raw value where it does not, strips quotes | Unit | pass | not run | not run | not run |
| raw_cookie_values > every same-named pair, in order; only that exact name | Unit | pass | not run | not run | not run |
| the profiler session in a large cookie jar > a valid session is found behind a stale same-named cookie the parser kept | Unit | pass | not run | not run | not run |
| the profiler session in a large cookie jar > a valid session the parser missed entirely (a malformed neighbour) still counts | Unit | pass | not run | not run | not run |
| the profiler session in a large cookie jar > an ordinary visit without a session is sent to the login page, no diagnosis | Unit | pass | not run | not run | not run |
| the profiler session in a large cookie jar > right after login with no session: the login page explains, and the server logs facts, not values | Unit | pass | not run | not run | not run |
| the profiler session in a large cookie jar > the login response sends the browser on with the after-login marker | Unit | pass | not run | not run | not run |
| the run page, from the dashboard form > opened from another site, the run waits for a click | Unit | pass | not run | not run | not run |
| the run page, from the dashboard form > takes a path typed without its slash, and still refuses another site | Unit | pass | not run | not run | not run |
packages/devtools/test/profiler-site-fixes.test.ts | |||||
| site_fixes > a line two pages share, its saving per page (by the site share) and in all; one-page lines left to their report | Unit | pass | not run | not run | not run |
| site_fixes > keeping answers or the whole page is a per-page decision; memory is not a time; a built line goes by its module | Unit | pass | not run | not run | not run |
packages/devtools/test/profiler-slow-regex.test.ts | |||||
| backtracks > a fixed part in the group pins where each repeat ends: not claimed | Unit | pass | not run | not run | not run |
| backtracks > a repeat inside a repeated group | Unit | pass | not run | not run | not run |
| backtracks > one level of repeats, a repeated group of fixed parts, a choice, escapes and classes | Unit | pass | not run | not run | not run |
| regex_literals > the literals on a line, not a division, a string or a comment | Unit | pass | not run | not run | not run |
| slow-regex > a cheap pattern run thousands of times stays quiet: no rewrite of it helps | Unit | pass | not run | not run | not run |
| slow-regex > a short value cleaned per item is not "the whole string rescanned", however much it adds up to | Unit | pass | not run | not run | not run |
| slow-regex > names the line, says it can backtrack, and leaves a quick pattern alone | Unit | pass | not run | not run | not run |
| slow-regex > no RegExp frame (V8 charged the pattern to the line): a literal that can backtrack still names it | Unit | pass | not run | not run | not run |
| slow-regex > read into an inlined callee, the caller's one call is not the callee's: its bytes are every item's | Unit | pass | not run | not run | not run |
| slow-regex > the ledger charges a pattern’s own code to the line that writes it | Unit | pass | not run | not run | not run |
packages/devtools/test/profiler-span.test.ts | |||||
| instrument without a recording > an instrumented method keeps its `this` | Unit | pass | not run | not run | not run |
| span under the profiler host > a span opened by hand while a span runs is its child: the parent keeps only its own time | Unit | pass | not run | not run | not run |
| span under the profiler host > a span's caller is the app line that called span(), not the profiler's own file | Unit | pass | not run | not run | not run |
packages/devtools/test/profiler-sqlite-drivers.test.ts | |||||
| a SQLite store without node:sqlite > opens its file with better-sqlite3 | Unit | pass | not run | not run | not run |
| a SQLite store without node:sqlite > with neither driver, says what to install | Unit | pass | not run | not run | not run |
packages/devtools/test/profiler-sqlite-open.test.ts | |||||
| a SQLite store whose file cannot be opened > fails with node:sqlite's own error, not a missing driver | Unit | pass | not run | not run | not run |
packages/devtools/test/profiler-standalone-attrs.test.ts | |||||
| the standalone export reads a module script the way a browser does > a repeated src is dropped in every copy, not just the first | Unit | pass | not run | not run | not run |
| the standalone export reads a module script the way a browser does > TYPE = "module" SRC = "/r.js?a=1&b=2", kept attributes at 0 | Unit | pass | not run | not run | not run |
| the standalone export reads a module script the way a browser does > TYPE = "module" SRC = "/r.js?a=1&b=2", kept attributes at 1 | Unit | pass | not run | not run | not run |
| the standalone export reads a module script the way a browser does > TYPE = "module" SRC = "/r.js?a=1&b=2", kept attributes at 2 | Unit | pass | not run | not run | not run |
| the standalone export reads a module script the way a browser does > type='module' src='/r.js?a=1&b=2', kept attributes at 0 | Unit | pass | not run | not run | not run |
| the standalone export reads a module script the way a browser does > type='module' src='/r.js?a=1&b=2', kept attributes at 1 | Unit | pass | not run | not run | not run |
| the standalone export reads a module script the way a browser does > type='module' src='/r.js?a=1&b=2', kept attributes at 2 | Unit | pass | not run | not run | not run |
| the standalone export reads a module script the way a browser does > type="module" src="/r.js?a=1&b=2", kept attributes at 0 | Unit | pass | not run | not run | not run |
| the standalone export reads a module script the way a browser does > type="module" src="/r.js?a=1&b=2", kept attributes at 1 | Unit | pass | not run | not run | not run |
| the standalone export reads a module script the way a browser does > type="module" src="/r.js?a=1&b=2", kept attributes at 2 | Unit | pass | not run | not run | not run |
| the standalone export reads a module script the way a browser does > type="module" src="/r.js?a=1&b=2", kept attributes at 0 | Unit | pass | not run | not run | not run |
| the standalone export reads a module script the way a browser does > type="module" src="/r.js?a=1&b=2", kept attributes at 1 | Unit | pass | not run | not run | not run |
| the standalone export reads a module script the way a browser does > type="module" src="/r.js?a=1&b=2", kept attributes at 2 | Unit | pass | not run | not run | not run |
| the standalone export reads a module script the way a browser does > type=module src=/r.js?a=1&b=2, kept attributes at 0 | Unit | pass | not run | not run | not run |
| the standalone export reads a module script the way a browser does > type=module src=/r.js?a=1&b=2, kept attributes at 1 | Unit | pass | not run | not run | not run |
| the standalone export reads a module script the way a browser does > type=module src=/r.js?a=1&b=2, kept attributes at 2 | Unit | pass | not run | not run | not run |
packages/devtools/test/profiler-standalone-key.test.ts | |||||
| the standalone export sends the profiler key to no asset host | Unit | pass | not run | not run | not run |
packages/devtools/test/profiler-storage.test.ts | |||||
| memory store > caps at its size | Unit | pass | not run | not run | not run |
| memory store > satisfies the contract | Unit | pass | not run | not run | not run |
| sqlite store > satisfies the contract (in-memory db) | Unit | pass | not run | not run | not run |
packages/devtools/test/profiler-store-init-retry.test.ts | |||||
| a store whose init failed once is tried again by its next use, and the report reaches it | Unit | pass | not run | not run | not run |
packages/devtools/test/profiler-store-setup.test.ts | |||||
| the Postgres store > a setup that fails once is set up again by the next call | Unit | pass | not run | not run | not run |
| the Postgres store > two first calls at once open one pool | Unit | pass | not run | not run | not run |
| the Redis store > two first calls at once open one client | Unit | pass | not run | not run | not run |
packages/devtools/test/profiler-stores-kept.test.ts | |||||
| the Postgres store > `schema` and `table` name the tables, and the dump column is jsonb | Unit | pass | not run | not run | not run |
| the Postgres store > a visit write trims its page to the newest MAX_VISITS_PER_KEY | Unit | pass | not run | not run | not run |
| the Postgres store > after close, the next call opens a new pool and the old one is ended | Unit | pass | not run | not run | not run |
| the Postgres store > the default tables are ogygia_reports and ogygia_visits | Unit | pass | not run | not run | not run |
| the Redis store > a visit that expired leaves its page's index when the visits are listed | Unit | pass | not run | not run | not run |
| the Redis store > after close, the next call connects a new client | Unit | pass | not run | not run | not run |
| the Redis store > keeps at most 1000 reports with no prune from the host, and trims only past that | Unit | pass | not run | not run | not run |
| the SQLite store > keeps the newest MAX_VISITS_PER_KEY visits per page | Unit | pass | not run | not run | not run |
| the SQLite store > makes its file's parent folders | Unit | pass | not run | not run | not run |
packages/devtools/test/profiler-ui-browser-build.test.ts | |||||
| the browser build of the profiler’s islands imports no server module | Unit | pass | not run | not run | not run |
packages/devtools/test/profiler-ui-store-close.test.ts | |||||
| a store call closes its connection > when the transaction cannot start (it throws) | Unit | pass | not run | not run | not run |
| a store call closes its connection > when the work in it succeeds | Unit | pass | not run | not run | not run |
packages/devtools/test/profiler-vitals-moved.test.ts | |||||
| vitals_moved > INP: the phase that grew is named (its handlers) | Unit | pass | not run | not run | not run |
| vitals_moved > LCP grew: the part that grew is named (its download) | Unit | pass | not run | not run | not run |
| vitals_moved > the noise floor: per vital (TTFB 50, FCP/LCP 100, INP 40 ms) and a fifth of the old value; CLS 0.05; a missing side says nothing | Unit | pass | not run | not run | not run |
| vitals_moved > the parts are the explanation’s own numbers (one split, never two that drift) | Unit | pass | not run | not run | not run |
packages/devtools/test/profiler-weigh-urls.test.ts | |||||
| a 404 is that for good: not asked again | Unit | pass | not run | not run | not run |
| a 5xx is weighed again by the next report | Unit | pass | not run | not run | not run |
| a timeout is weighed again by the next report | Unit | pass | not run | not run | not run |
| two reports at once weigh a file once | Unit | pass | not run | not run | not run |
packages/devtools/test/profiler-window-cap.test.ts | |||||
| a window that reaches its cap and then fails to stop leaves no unhandled rejection, and frees the recorder | Unit | pass | not run | not run | not run |
packages/devtools/test/profiler-window-id.test.ts | |||||
| an id ends after its lifetime | Unit | pass | not run | not run | not run |
| an id it minted checks; the same id under another secret does not | Unit | pass | not run | not run | not run |
| an id with any part changed does not check | Unit | pass | not run | not run | not run |
| dev with no secret signs under a key of its own process: another one’s id does not check | Unit | pass | not run | not run | not run |
| production with no secret has no window ids | Unit | pass | not run | not run | not run |
packages/devtools/test/profiler-window-key.test.ts | |||||
| a production profiler with no secret writes no request under a window id | Unit | pass | not run | not run | not run |
| a window id signed the old way (a bare HMAC of the nonce, no lifetime) is not taken | Unit | pass | not run | not run | not run |
packages/devtools/test/profiler.test.ts | |||||
| a merged component without line ticks > reads at the line of the frame that held most of its self time (an $effect arrow), not the wrapper | Unit | pass | not run | not run | not run |
| a report carried with the request (serverless: another instance answers) > a fresh instance answers .json / .dump / .ogp from a gzipped kept copy, and opens its own .ogp | Unit | pass | not run | not run | not run |
| a visit posted in halves > the runtime half (islands) and the mark half (marks) of one visit fold into one picture | Unit | pass | not run | not run | not run |
| analyze > builds a flame tree with (root) lifted and idle pruned | Unit | pass | not run | not run | not run |
| analyze > counts recursive frames once in total time | Unit | pass | not run | not run | not run |
| analyze > joins call counts per function by name+url, so same-named functions stay separate | Unit | pass | not run | not run | not run |
| analyze > merges a component wrapper frame with its inline anonymous work | Unit | pass | not run | not run | not run |
| analyze > recognizes components and buckets time by package | Unit | pass | not run | not run | not run |
| analyze > splits busy / idle / gc and attributes self+total per frame | Unit | pass | not run | not run | not run |
| analyze_heap > aggregates sampled allocations per function | Unit | pass | not run | not run | not run |
| call stacks + locations > a call path reads as if the profiler were not there: no span wrapper, no AsyncLocalStorage run it called in | Unit | pass | not run | not run | not run |
| call stacks + locations > a component is one frame on a stack: the bundled wrapper + the renderer calls between fold into the .svelte body | Unit | pass | not run | not run | not run |
| call stacks + locations > a nested occurrence (recursion, or a component’s inner frame under its own wrapper) is one stack, never its own caller | Unit | pass | not run | not run | not run |
| call stacks + locations > joins a sourcemap’s relative sources onto the chunk directory (an openable path) | Unit | pass | not run | not run | not run |
| call stacks + locations > keeps the full path + column so a row can be opened in an editor | Unit | pass | not run | not run | not run |
| call stacks + locations > ranks a function’s call paths by the time that flowed through each, nearest caller first | Unit | pass | not run | not run | not run |
| call stacks + locations > ships stacks + locations in the JSON report | Unit | pass | not run | not run | not run |
| call stacks + locations > the island host wrapper reads as its island, not as the hash Svelte named its virtual file with | Unit | pass | not run | not run | not run |
| categorize > classifies by url and falls back to Svelte naming for bundles | Unit | pass | not run | not run | not run |
| compare + history > every row carries a unique key, even two anonymous functions in one file (a view keys by it) | Unit | pass | not run | not run | not run |
| compare + history > groups page reports into a per-page history, oldest first, by median | Unit | pass | not run | not run | not run |
| compare + history > reports signed deltas per component / function and what changed in the findings | Unit | pass | not run | not run | not run |
| compare: the garbage makers diff > matches makers by line, signs the change, and marks what is only on one side | Unit | pass | not run | not run | not run |
| component confirmation (structural) > keeps a candidate that reaches svelte internals, drops one that does not | Unit | pass | not run | not run | not run |
| component confirmation (structural) > names an anonymous inline frame after its .svelte file the way Svelte does (kebab-case too) | Unit | pass | not run | not run | not run |
| findings name the row and the fix > a repeated component gets a warn with its row anchor, file and a fix; a heavy single render names what burns inside | Unit | pass | not run | not run | not run |
| findings name the row and the fix > N+1: the same endpoint once per item, folded to a template | Unit | pass | not run | not run | not run |
| findings name the row and the fix > ogygia’s own cost: the seed and props bytes ride the request log into the findings | Unit | pass | not run | not run | not run |
| I/O wait attribution > buckets async resource types into friendly kinds | Unit | pass | not run | not run | not run |
| I/O wait attribution > includes the waiting aggregate and callers in the JSON | Unit | pass | not run | not run | not run |
| I/O wait attribution > renders "Waiting by function" from network callers + async I/O ops | Unit | pass | not run | not run | not run |
| inline_threshold_tune: the inlineStyleThreshold that takes the small sheets off the paint > one small sheet, or none: nothing to tune | Unit | pass | not run | not run | not run |
| inline_threshold_tune: the inlineStyleThreshold that takes the small sheets off the paint > the smallest build sheets first, each 8 KB or less, 24 KB in all; just over the largest, in whole KB | Unit | pass | not run | not run | not run |
| instrument (ogygia/profiler): every call of a function becomes a span > the function form: a wrapped function, attrs from the result and the arguments, this kept | Unit | pass | not run | not run | not run |
| instrument (ogygia/profiler): every call of a function becomes a span > the method form: patched in place, restored by the returned function | Unit | pass | not run | not run | not run |
| lake_saving: what making islands lakes saves > each chunk once; a chunk an awake island loads stays | Unit | pass | not run | not run | not run |
| more from the same snapshot, on the host > a page profile counts the promises per render and measures what a render leaves behind | Unit | pass | not run | not run | not run |
| page-mode honesty findings (redirect / not-a-render / budget) > a window over several requests: the slowest named with their split, no one-request claims | Unit | pass | not run | not run | not run |
| page-mode honesty findings (redirect / not-a-render / budget) > does NOT cry low-confidence on a real render | Unit | pass | not run | not run | not run |
| page-mode honesty findings (redirect / not-a-render / budget) > each island's own JS: shared chunks counted once, where no other island uses them | Unit | pass | not run | not run | not run |
| page-mode honesty findings (redirect / not-a-render / budget) > echoes the serverless budget note | Unit | pass | not run | not run | not run |
| page-mode honesty findings (redirect / not-a-render / budget) > flags a low-confidence window (no components, few samples) | Unit | pass | not run | not run | not run |
| page-mode honesty findings (redirect / not-a-render / budget) > memory kept per render: the runway to the heap limit, and a package named as its owner | Unit | pass | not run | not run | not run |
| page-mode honesty findings (redirect / not-a-render / budget) > reports the redirect it followed | Unit | pass | not run | not run | not run |
| page-mode honesty findings (redirect / not-a-render / budget) > spots an app that caches the page after the first render | Unit | pass | not run | not run | not run |
| page-mode honesty findings (redirect / not-a-render / budget) > warns when the body was too small to be a real page | Unit | pass | not run | not run | not run |
| page-mode honesty findings (redirect / not-a-render / budget) > warns when the profiled renders were a 3xx (unfollowed redirect), not a page | Unit | pass | not run | not run | not run |
| paths to fix: hot functions grouped by the caller they share > a path needs two hot functions and real time; a lone hot function is no path | Unit | pass | not run | not run | not run |
| paths to fix: hot functions grouped by the caller they share > groups walk, format and sortRows under buildTree, never under the route root; the tree keeps the chain | Unit | pass | not run | not run | not run |
| paths to fix: hot functions grouped by the caller they share > the owner is the most specific frame that keeps the time: down through the root and an island host, not the layout glue | Unit | pass | not run | not run | not run |
| paths to fix: hot functions grouped by the caller they share > the profiler’s own span/within wrappers fold out of a path chain like Svelte’s glue | Unit | pass | not run | not run | not run |
| paths: the owner is a named place > an anonymous arrow holding the hot functions never owns the path; its named parent does, and the app’s own hot functions stay in the set beside a dependency’s many | Unit | pass | not run | not run | not run |
| profiler handle > beacon: the tag rides an authed document, /beacon is bounded, the report joins by fingerprint | Unit | pass | not run | not run | not run |
| profiler handle > hides the UI in prod without a secret and accepts the right key | Unit | pass | not run | not run | not run |
| profiler handle > production: an idle request is not attributed; a header-profiled one is | Unit | pass | not run | not run | not run |
| profiler handle > runs a lean path with network off: still logs requests, no Server-Timing | Unit | pass | not run | not run | not run |
| profiler handle > serializes recordings: a second profile while one is running renders un-profiled | Unit | pass | not run | not run | not run |
| profiler handle > times requests, attributes outbound fetches, serves the dashboard | Unit | pass | not run | not run | not run |
| profiler handle > wraps Kit's event.fetch: a same-origin call dispatched in-process is recorded like any other | Unit | pass | not run | not run | not run |
| report data (JSON, findings, budget, treemap, waiting) > emits a curated JSON report an agent can read | Unit | pass | not run | not run | not run |
| report data (JSON, findings, budget, treemap, waiting) > renders budget bar + treemap and counts idle exactly once | Unit | pass | not run | not run | not run |
| report data (JSON, findings, budget, treemap, waiting) > round-trips a dump through JSON and renders identically (the serverless path) | Unit | pass | not run | not run | not run |
| report data (JSON, findings, budget, treemap, waiting) > shares one findings source between HTML and JSON | Unit | pass | not run | not run | not run |
| report data (JSON, findings, budget, treemap, waiting) > sorts the components table by self desc by default (ancestors sink) | Unit | pass | not run | not run | not run |
| route_prerendered: the nearest `export const prerender` decides > the page file, a layout above it, or nothing (the root layout says no) | Unit | pass | not run | not run | not run |
| sequential_ms > sums back-to-back calls but merges overlapping ones | Unit | pass | not run | not run | not run |
| sourcemaps > decodes segments and resolves a generated position to the original file | Unit | pass | not run | not run | not run |
| sourcemaps > recovers original identifiers for anonymous frames via `names` | Unit | pass | not run | not run | not run |
| span + tag (ogygia/profiler) > cold-instance: on AWS Lambda the time to the first request is the cold start; elsewhere it says nothing | Unit | pass | not run | not run | not run |
| span + tag (ogygia/profiler) > fair_shares: the wall split among whoever is open, adding up to the union | Unit | pass | not run | not run | not run |
| span + tag (ogygia/profiler) > findings quote one render’s worth: a 3-render profile’s 30 ms hot function burns 10 ms per render | Unit | pass | not run | not run | not run |
| span + tag (ogygia/profiler) > findings: a span repeated per item, a cold cache, a failing span | Unit | pass | not run | not run | not run |
| span + tag (ogygia/profiler) > is a pass-through without a recorder: sync stays sync, a promise stays a promise, nothing recorded | Unit | pass | not run | not run | not run |
| span + tag (ogygia/profiler) > key-unread: a load’s wait counts once, however many of its keys nobody reads | Unit | pass | not run | not run | not run |
| span + tag (ogygia/profiler) > records name, attrs (static or from the result), errors, nesting and tags while recording | Unit | pass | not run | not run | not run |
| span + tag (ogygia/profiler) > span_rows folds a recording per name: count, p50, max, errors, cache tallies, callers | Unit | pass | not run | not run | not run |
| span + tag (ogygia/profiler) > the timeline treats a span as the overlay: a gap inside it reads "in <span>", a call inside it names the span | Unit | pass | not run | not run | not run |
| standalone HTML export > a build’s island: its `src` is the file inlined (its `entry` is a name, no file); both name the key | Unit | pass | not run | not run | not run |
| standalone HTML export > inlines styles, the runtime and every reachable chunk as an import map of data: URLs | Unit | pass | not run | not run | not run |
| standalone HTML export > reads the page’s tags as a browser does: a quoted `>`, a comment, rel tokens, any case | Unit | pass | not run | not run | not run |
| the accuracy round: hot lines, server-timing, call paths, cold start, run spread, vitals > a function's hot lines add up to its own time, not to its ticks at the profile's mean interval | Unit | pass | not run | not run | not run |
| the accuracy round: hot lines, server-timing, call paths, cold start, run spread, vitals > a newer profiler in the process takes over: the older one’s sampler stands down | Unit | pass | not run | not run | not run |
| the accuracy round: hot lines, server-timing, call paths, cold start, run spread, vitals > analyze: V8 line ticks become the function’s hot lines; runs split a component’s time per render | Unit | pass | not run | not run | not run |
| the accuracy round: hot lines, server-timing, call paths, cold start, run spread, vitals > app_call_chain: the first-party frames above an I/O call, nearest first | Unit | pass | not run | not run | not run |
| the accuracy round: hot lines, server-timing, call paths, cold start, run spread, vitals > beacon marks: per page, per name, joined to the report as "marked in the browser" | Unit | pass | not run | not run | not run |
| the accuracy round: hot lines, server-timing, call paths, cold start, run spread, vitals > beacon: a vitals-only post lands per page; the report joins them; junk is bounded | Unit | pass | not run | not run | not run |
| the accuracy round: hot lines, server-timing, call paths, cold start, run spread, vitals > cold start: the warm-up render against the warm ones, per file | Unit | pass | not run | not run | not run |
| the accuracy round: hot lines, server-timing, call paths, cold start, run spread, vitals > heap-filled-before: a page that keeps memory, profiled on a heap earlier requests filled | Unit | pass | not run | not run | not run |
| the accuracy round: hot lines, server-timing, call paths, cold start, run spread, vitals > parse_server_timing: tolerant of quotes, missing dur, junk; capped | Unit | pass | not run | not run | not run |
| the accuracy round: hot lines, server-timing, call paths, cold start, run spread, vitals > run_spread + the variance findings: a cold first run vs a run that just varies | Unit | pass | not run | not run | not run |
| the accuracy round: hot lines, server-timing, call paths, cold start, run spread, vitals > run_trend: a climb across the runs, not one slow run, a flat page or a cold start | Unit | pass | not run | not run | not run |
| the accuracy round: hot lines, server-timing, call paths, cold start, run spread, vitals > spans nest outside a request context too: a handle in front of the profiled request keeps parent links | Unit | pass | not run | not run | not run |
| the accuracy round: hot lines, server-timing, call paths, cold start, run spread, vitals > the always-on sampler folds short windows into a hot-functions table on the dashboard | Unit | pass | not run | not run | not run |
| the accuracy round: hot lines, server-timing, call paths, cold start, run spread, vitals > the browser’s vitals against the server: the LCP gap, the TTFB gap, or just the numbers | Unit | pass | not run | not run | not run |
| the accuracy round: hot lines, server-timing, call paths, cold start, run spread, vitals > the caller skip list: Kit's bundled runtime frames are the framework, not the app | Unit | pass | not run | not run | not run |
| the accuracy round: hot lines, server-timing, call paths, cold start, run spread, vitals > the trap: a coarse background window keeps only a request over the threshold | Unit | pass | not run | not run | not run |
| the accuracy round: hot lines, server-timing, call paths, cold start, run spread, vitals > upstream-split: the slowest call’s own Server-Timing says whose the wait is | Unit | pass | not run | not run | not run |
| the browser CPU profile and the request replay > a browser frame with a source map: the minified name and the chunk give way to the real function and its .svelte file | Unit | pass | not run | not run | not run |
| the browser CPU profile and the request replay > a minified client chunk: the confirmed component is named from the .svelte files the build put in the chunk; helpers are not | Unit | pass | not run | not run | not run |
| the browser CPU profile and the request replay > a prerendered page: Kit throws on url.search, and the request log keeps the entry without it | Unit | pass | not run | not run | not run |
| the browser CPU profile and the request replay > beacon: a cpu trace lands per page and the report shows it as browser components | Unit | pass | not run | not run | not run |
| the browser CPU profile and the request replay > beacon: a script from another host, or a chunk with no app file in it, is a dependency whatever its names | Unit | pass | not run | not run | not run |
| the browser CPU profile and the request replay > client files: app assets map to the built client dir, other hosts and crafted paths never do | Unit | pass | not run | not run | not run |
| the browser CPU profile and the request replay > name_at_source: the function name written in the original source at a mapped position | Unit | pass | not run | not run | not run |
| the browser CPU profile and the request replay > replay: a header-profiled request is rendered again in page mode with its query and kept headers | Unit | pass | not run | not run | not run |
| the browser CPU profile and the request replay > self_profile_to_cpuprofile: stacks become nodes, samples become deltas, no stack is idle; junk is null | Unit | pass | not run | not run | not run |
| the causality graph, pending resources, memory per component, path diff > a gap is named after the resources pending across it | Unit | pass | not run | not run | not run |
| the causality graph, pending resources, memory per component, path diff > await_graph: an edge where a call started only once another finished, none for calls in flight together | Unit | pass | not run | not run | not run |
| the causality graph, pending resources, memory per component, path diff > compare: the paths to fix diff by owner | Unit | pass | not run | not run | not run |
| the causality graph, pending resources, memory per component, path diff > heap_by_component: bytes credited to the nearest component, nested components take their own | Unit | pass | not run | not run | not run |
| the deep pictures on the host: visits, keep, the byte strip, nested traces, the sink, the site page > /page?format=keep answers the dump for the browser to keep, with the byte strip of the rendered document | Unit | pass | not run | not run | not run |
| the deep pictures on the host: visits, keep, the byte strip, nested traces, the sink, the site page > beacon: a visit lands per page, bounded, and the report JSON carries its picture | Unit | pass | not run | not run | not run |
| the deep pictures on the host: visits, keep, the byte strip, nested traces, the sink, the site page > compare with a report this server no longer holds renders the browser-store fallback, not a 404 | Unit | pass | not run | not run | not run |
| the deep pictures on the host: visits, keep, the byte strip, nested traces, the sink, the site page > nested traces: a request that asks gets our picture in a header; our outbound calls ask while recording | Unit | pass | not run | not run | not run |
| the deep pictures on the host: visits, keep, the byte strip, nested traces, the sink, the site page > the sink: a profiler that stands down during a post still sends the rows left | Unit | pass | not run | not run | not run |
| the deep pictures on the host: visits, keep, the byte strip, nested traces, the sink, the site page > the sink: request rows post as NDJSON with the bearer key; a failed post keeps the rows | Unit | pass | not run | not run | not run |
| the deep pictures on the host: visits, keep, the byte strip, nested traces, the sink, the site page > the sink: the rows of a failed post go before the rows read while it was out | Unit | pass | not run | not run | not run |
| the deep pictures on the host: visits, keep, the byte strip, nested traces, the sink, the site page > the sink: while a post is out, the rows read wait in the buffer, not in a post per tick | Unit | pass | not run | not run | not run |
| the ogygia / svelte / kit round: islands, seed, holes, lanes, markup vs logic > a cross-origin-isolated page in Safari: the first byte hidden, the visit kept, never a 0 ms first byte | Unit | pass | not run | not run | not run |
| the ogygia / svelte / kit round: islands, seed, holes, lanes, markup vs logic > a page the build wrote as a file (no render ran): no render findings, the note says why | Unit | pass | not run | not run | not run |
| the ogygia / svelte / kit round: islands, seed, holes, lanes, markup vs logic > a render reading a prerender function's answer over HTTP: its own call, named, not another visitor | Unit | pass | not run | not run | not run |
| the ogygia / svelte / kit round: islands, seed, holes, lanes, markup vs logic > a Safari visit’s hidden other-origin sizes take the report’s own weights, said whose they are | Unit | pass | not run | not run | not run |
| the ogygia / svelte / kit round: islands, seed, holes, lanes, markup vs logic > an island that reads page.data whole is the reason everything ships | Unit | pass | not run | not run | not run |
| the ogygia / svelte / kit round: islands, seed, holes, lanes, markup vs logic > an island that threw its server DOM away and drew the same markup back: its own finding, no line of it blamed | Unit | pass | not run | not run | not run |
| the ogygia / svelte / kit round: islands, seed, holes, lanes, markup vs logic > analyze: a component’s own time splits into markup (svelte internals under it) and logic; its parent is who rendered it | Unit | pass | not run | not run | not run |
| the ogygia / svelte / kit round: islands, seed, holes, lanes, markup vs logic > another engine’s visit: what Safari saw that the latest Chrome visit did not, named with the browser | Unit | pass | not run | not run | not run |
| the ogygia / svelte / kit round: islands, seed, holes, lanes, markup vs logic > findings: the seed explained, devalue culprits, the wake advisor, hole economics, the browser side | Unit | pass | not run | not run | not run |
| the ogygia / svelte / kit round: islands, seed, holes, lanes, markup vs logic > fp-unstable: an island whose fingerprint moved between the profile’s own renders, its prop named; ogygia’s own ids never | Unit | pass | not run | not run | not run |
| the ogygia / svelte / kit round: islands, seed, holes, lanes, markup vs logic > group_islands: one row per island — fingerprints merge, the first culprit and name stand | Unit | pass | not run | not run | not run |
| the ogygia / svelte / kit round: islands, seed, holes, lanes, markup vs logic > kit-uneval on a csr=false page: Kit’s copy of the server load data, built and dropped; ogygia’s own serialization apart | Unit | pass | not run | not run | not run |
| the ogygia / svelte / kit round: islands, seed, holes, lanes, markup vs logic > kit-uneval: the client router’s serialization is named, with the csr=false answer | Unit | pass | not run | not run | not run |
| the ogygia / svelte / kit round: islands, seed, holes, lanes, markup vs logic > never-hydrated: the beacon reported other islands but not this one | Unit | pass | not run | not run | not run |
| the ogygia / svelte / kit round: islands, seed, holes, lanes, markup vs logic > one module, two copies, both on this page: named with its files and why; a copy the page never loads is no finding | Unit | pass | not run | not run | not run |
| the ogygia / svelte / kit round: islands, seed, holes, lanes, markup vs logic > production: the beacon tag needs the signed flag cookie the login sets (site-wide), or the key header | Unit | pass | not run | not run | not run |
| the ogygia / svelte / kit round: islands, seed, holes, lanes, markup vs logic > promises by the tens of thousands: Svelte's async renderer is named with what it renders most; an app's own is the loop | Unit | pass | not run | not run | not run |
| the ogygia / svelte / kit round: islands, seed, holes, lanes, markup vs logic > report JSON: island rows joined with weights, the browser and the components; seed + holes; kit costs | Unit | pass | not run | not run | not run |
| the ogygia / svelte / kit round: islands, seed, holes, lanes, markup vs logic > the browser a visit came from: named from its user agent, and on its findings when it is not a Chromium one | Unit | pass | not run | not run | not run |
| the ogygia / svelte / kit round: islands, seed, holes, lanes, markup vs logic > the browser's vitals: a latest visit that loaded with the dock open says its paints may be the dock's | Unit | pass | not run | not run | not run |
| the ogygia / svelte / kit round: islands, seed, holes, lanes, markup vs logic > the Islands table: an island reading a browser-only value while rendering is told so, by its line | Unit | pass | not run | not run | not run |
| the ogygia / svelte / kit round: islands, seed, holes, lanes, markup vs logic > the process's memory growth: said plainly, and not beside the render measured to keep memory | Unit | pass | not run | not run | not run |
| the ogygia / svelte / kit round: islands, seed, holes, lanes, markup vs logic > the screen the visit had: named on the findings it decides, and what a phone saw that a wide visit did not | Unit | pass | not run | not run | not run |
| the ogygia / svelte / kit round: islands, seed, holes, lanes, markup vs logic > timeline: a load lane waiting only inside a span (a driver the hooks cannot see) still counts, and chains | Unit | pass | not run | not run | not run |
| the ogygia / svelte / kit round: islands, seed, holes, lanes, markup vs logic > timeline: loads that overlap are no chain; a universal load with waits gets the info | Unit | pass | not run | not run | not run |
| the ogygia / svelte / kit round: islands, seed, holes, lanes, markup vs logic > timeline: one lane per Kit load, the page lane behind the layout lane is the parent() chain | Unit | pass | not run | not run | not run |
| the profiler’s own cost is taken out > a page report carries the overhead it took out, and the runs as the app would have paid them | Unit | pass | not run | not run | not run |
| the profiler’s own cost is taken out > analyze: samples on the profiler’s own frames are overhead, not busy time | Unit | pass | not run | not run | not run |
| timeline (critical path + phases) > a library working on after an await takes the phase of whoever last called into it | Unit | pass | not run | not run | not run |
| timeline (critical path + phases) > classifies a stack by the deepest phase marker; ogygia beats render beats kit | Unit | pass | not run | not run | not run |
| timeline (critical path + phases) > coalesce: a run of small slivers becomes one block named after what dominated it | Unit | pass | not run | not run | not run |
| timeline (critical path + phases) > lists the steps that set the time, in order, and folds the slivers | Unit | pass | not run | not run | not run |
| timeline (critical path + phases) > rides on analyze: the report carries the timeline and the findings read it | Unit | pass | not run | not run | not run |
| timeline (critical path + phases) > splits the window by phase, cpu and wait apart | Unit | pass | not run | not run | not run |
| timeline (critical path + phases) > spots awaits in a row (different calls, no CPU between) and what running them together saves | Unit | pass | not run | not run | not run |
| timeline (critical path + phases) > the profiler’s own CPU inside a wait does not chop it: one wait, as if the profiler were not there | Unit | pass | not run | not run | not run |
| timeline (critical path + phases) > walks the window into cpu / wait / gap segments with owners and phases | Unit | pass | not run | not run | not run |
| who caused the GC, on the host > a render that churns memory gets its pauses attributed to the allocating function | Unit | pass | not run | not run | not run |
packages/devtools/test/property-login-next.test.ts | |||||
| the login's next > a protocol-relative or backslash next goes to the profiler, at the root too | Unit | pass | not run | not run | not run |
| the login's next > any next stays on the origin and under the base; one already there is kept | Unit | pass | not run | not run | not run |
packages/devtools/test/property-visit.test.ts | |||||
| beacon visit ingest > any body: never throws; keeps plain, bounded data; leaves Object.prototype alone | Unit | pass | not run | not run | not run |
| beacon visit ingest > posts of one visit, folded, stay inside the bounds of one post | Unit | pass | not run | not run | not run |
| beacon visit ingest > what it kept parses back to itself | Unit | pass | not run | not run | not run |
packages/devtools/test/proto-key-bodies.test.ts | |||||
| a report carried with the request > a dump with a `__proto__` or `constructor` key at any depth is opened as data, or refused | Unit | pass | not run | not run | not run |
| the beacon > takes a `__proto__` or `constructor` key at any depth as plain data, or refuses it | Unit | pass | not run | not run | not run |
| the unlock > a key under `__proto__` is no key | Unit | pass | not run | not run | not run |
packages/devtools/test/regions-scan.test.ts | |||||
| cannot be fooled by tag-like text > a `>` inside an attribute value does not end the tag early | Unit | pass | not run | not run | not run |
| cannot be fooled by tag-like text > a region’s OWN tag can carry a `<ogygia-region>` in an attribute without spawning a phantom | Unit | pass | not run | not run | not run |
| cannot be fooled by tag-like text > a stray `<` in text (a < b) is not a tag | Unit | pass | not run | not run | not run |
| cannot be fooled by tag-like text > ignores <ogygia-region> and stray </ogygia-region> inside an HTML comment | Unit | pass | not run | not run | not run |
| cannot be fooled by tag-like text > ignores <ogygia-region> inside a CSS comment in an inlined <style> (the reported bug) | Unit | pass | not run | not run | not run |
| cannot be fooled by tag-like text > ignores <ogygia-region> inside ANOTHER element’s attribute value (double-quoted) | Unit | pass | not run | not run | not run |
| cannot be fooled by tag-like text > ignores a stray </ogygia-region> inside a single-quoted attribute value | Unit | pass | not run | not run | not run |
| cannot be fooled by tag-like text > ignores tag-like text inside a <script> body | Unit | pass | not run | not run | not run |
| classification > a hole is a hole even with a wake attribute (endpoint wins) | Unit | pass | not run | not run | not run |
| classification > reads island / lake / hole off the attributes | Unit | pass | not run | not run | not run |
| nesting > does NOT yield regions strictly inside an island (its subtree is atomic) | Unit | pass | not run | not run | not run |
| nesting > yields a hole inside a lake (recurse into lakes and holes) | Unit | pass | not run | not run | not run |
| nesting > yields a login island nested inside a header lake so the caller can protect it | Unit | pass | not run | not run | not run |
| robustness > a document with no regions yields nothing | Unit | pass | not run | not run | not run |
| robustness > a self-closing region yields without opening a subtree | Unit | pass | not run | not run | not run |
| robustness > an unterminated region is surfaced (span is just its opening tag) | Unit | pass | not run | not run | not run |
| robustness > does not match <ogygia-region-ish> (a longer tag name) | Unit | pass | not run | not run | not run |
| spans > start / innerStart / innerEnd / end bracket the element and its inner HTML | Unit | pass | not run | not run | not run |
| tag names are case-insensitive, and whole names only > a longer name that starts with a known one is not that tag | Unit | pass | not run | not run | not run |
| tag names are case-insensitive, and whole names only > an upper-case region is a region; an upper-case raw-text element is skipped | Unit | pass | not run | not run | not run |
packages/devtools/test/request-stats-detail.test.ts | |||||
| hold_request_stats_detail > a hold given an owner goes with it: a window that ends without its release lets go | Unit | pass | not run | not run | not run |
| hold_request_stats_detail > a release called twice lets go once: it never ends another window’s hold | Unit | pass | not run | not run | not run |
| hold_request_stats_detail > stays on while any window holds it; overlapping windows end in any order | Unit | pass | not run | not run | not run |
packages/devtools/test/request-stats-standalone.test.ts | |||||
| a standalone profiler > hands core its request-stats store, so recordings keep per-island rows and batch holes | Unit | pass | not run | not run | not run |
packages/devtools/test/server-beacon-cache.test.ts | |||||
| a document of a browser that did not switch devtools on keeps the app's caching | Unit | pass | not run | not run | not run |
| a document the beacon is added to (devtools switched on by the cookie) is private and never stored | Unit | pass | not run | not run | not run |
| a document the beacon is added to (devtools switched on in the URL) is private and never stored | Unit | pass | not run | not run | not run |