[Bug]: ComposerBanner's `+ :has()` variant makes banner changes restyle the whole page on every thread switch
I maintainer di solito rispondono entro 1 giorno
Nessuno ha ancora preso questa issue.
Valutazione
- Difficoltà
- 2/5
- Tempo stimato
- 1-3 ore
- Idoneità per principianti
- 88/100
- Tipo di issue
- Bug
- Chiarezza
- Specificata chiaramente
- Stato di attività
- Attiva
- Stack tecnologico
- tailwindcss, typescript
- Ambito
- frontend, performance
Direzione di ricerca
Inizia in apps/web/src/components/chat/ComposerBanner.tsx alla riga 114 e confronta il selettore emesso con la riproduzione minima di Chromium nell’issue. Verifica la modifica con i casi di equivalenza dei selettori elencati e una traccia Performance mentre attivi o disattivi il banner oppure passi da un thread all’altro. Il lavoro è completato quando lo stile del banner rimane equivalente, mentre viene ridotta l’invalidazione ampia degli stili.
Scritto dal modello di indicizzazione a partire dal testo della issue.
Descrizione
Before submitting
- I searched existing issues and did not find a duplicate.
- I included enough detail to reproduce or investigate the problem.
Area
apps/web
Steps to reproduce
- Build
main(measured at95030dc674) and open the web app in Chromium with a populated sidebar. - Record a DevTools Performance trace while switching between threads.
- Look at the "Recalculate Style" events during each switch.
The minimal reproduction below needs no T3 Code. Open it in Chromium, record a Performance trace, and click the button:
<!doctype html>
<style>
/* ComposerBanner.tsx:114, as Tailwind compiles it */
.X + :has([data-chat-composer-form]) [data-chat-composer-form] > [data-slot=composer-banner-attachment]:first-child [data-composer-banner-surface=attached]::before { border-radius: 0 }
/* any group-has-* utility; its group does not even need to be on the page */
:is(:where(.group\/surface):has([data-banner=attached]) *) { box-shadow: none }
</style>
<button id="toggle">toggle banner</button>
<div id="list"></div>
<div id="stack"><div id="banner" class="X">banner</div><div><form data-chat-composer-form><div data-slot="composer-banner-attachment"><div data-composer-banner-surface="attached"></div></div></form></div></div>
<script>
document.getElementById("list").innerHTML = "<div><span>row</span></div>".repeat(1500);
const stack = document.getElementById("stack");
const banner = document.getElementById("banner");
document.getElementById("toggle").onclick = () => (banner.isConnected ? banner.remove() : stack.prepend(banner));
</script>
Expected behavior
Adding or removing a banner restyles the banner's neighbourhood, as it does when either rule is present alone.
Actual behavior
The Tailwind arbitrary variant in apps/web/src/components/chat/ComposerBanner.tsx:114 compiles to a bare :has() after a sibling combinator:
.X + :has([data-chat-composer-form]) [data-chat-composer-form] > [data-slot=composer-banner-attachment]:first-child [data-composer-banner-surface=attached]::before
When a group-has-* utility is in the same stylesheet (main ships many), adding or removing a sibling in that area makes Chromium restyle almost the whole document. The reproduction above, 20 banner toggles, counted with the trace's UpdateLayoutTree.elementCount:
| Stylesheet in the reproduction | Elements restyled per banner toggle (of 3,013) |
|---|---|
no :has() rules |
1 |
group-has rule only |
1 |
| ComposerBanner rule only | 6 |
| both, as in the file above | 2,861 |
| both, with the suggested rewrite below | 6 |
The group-has rule's group element is not on the page at all; the rule only has to be in the stylesheet.
A thread switch adds and removes these siblings, so every switch pays for it. Measured on a main build (95030dc674, default settings), 13 scripted thread switches in headless Chromium, three rounds each:
| As shipped | Only the two ComposerBanner rules rewritten | |
|---|---|---|
| Recalcs touching ≥1,000 elements | 17 | 1 |
| Elements restyled | 32.3k | 12.4k (−62%) |
| Style recalculation time | 315 ms | 156 ms (−50%) |
| Main thread blocked in long tasks, rapid switching | 728 ms | 461 ms (−37%) |
| Click → next paint, p50 | 74 ms | 59 ms |
| Keydown → next paint, p95 | 54–64 ms | 25–26 ms |
The element counts come from Chromium's own trace and do not depend on machine load; they repeated exactly across rounds.
Suggested fix: remove the :has(), which is redundant. The rest of the selector already requires a [data-chat-composer-form] descendant of that sibling, so + * matches the same elements:
- "[&+:has([data-chat-composer-form])_[data-chat-composer-form]>[data-slot=composer-banner-attachment]:first-child_[data-composer-banner-surface=attached]]:before:rounded-none [&+:has([data-chat-composer-form])_[data-chat-composer-form]>[data-slot=composer-banner-attachment]:first-child_[data-composer-banner-surface=attached]]:before:border-t-0",
+ "[&+*_[data-chat-composer-form]>[data-slot=composer-banner-attachment]:first-child_[data-composer-banner-surface=attached]]:before:rounded-none [&+*_[data-chat-composer-form]>[data-slot=composer-banner-attachment]:first-child_[data-composer-banner-surface=attached]]:before:border-t-0",
Checked equivalence: computed ::before styles are identical for the old and new selector with the form host after the attachment, with no attachment before it, with the attachment not first, and with the form as the sibling itself. With the rewrite, the reproduction's banner toggle restyles 6 elements instead of 2,861.
Impact
Major degradation or frequent failure
Version or commit
main @ 95030dc674 (rule introduced in #8693)
Environment
Linux, Chromium (headless, for the measurements) and the Electron desktop app. The effect is in Chromium's :has() invalidation, so every Chromium-based client is affected.
Logs or stack traces
The trace shows ScheduleStyleInvalidationTracking with changedPseudo: "has" walking every ancestor up to <html>, followed by StyleInvalidatorInvalidationTracking "Invalidation set invalidates subtree" with allDescendantsMightBeInvalid: true on each of them.
Screenshots, recordings, or supporting files
The reproduction above.
Workaround
None for users. Removing the two classes, or applying the suggested rewrite, removes the effect.
- Lingua principale
- TypeScript
- Stelle
- 24k
- Fork
- 6.3k
- Merge medio
- 10h 46m
- PR unite (30g)
- 299
Preparare l'ambiente
Avvia il container di sviluppo del progetto nel browser, con il tuo account GitHub.
- Nessun Dockerfile né file Docker Compose
- Ha un modello di pull request
- Leggi la guida per i contributori
Come iniziare
- Leggi tutta la issue e poi la guida ai contributi del progetto.
- Commenta sulla issue per dire che te ne occupi tu — evita che due persone facciano lo stesso lavoro.
- Fai un fork del repository e lavora su un branch.
- Apri una pull request che faccia riferimento al numero della issue.
Altre issue di pingdotgg/t3code
-
enhancement via-triage
Difficoltà 2/5 1-3 ore Idoneità per principianti 72/100
pingdotgg/t3code#14541 · 1 commento ·
I maintainer di solito rispondono entro 1 giorno
-
Difficoltà 2/5 1-3 ore Idoneità per principianti 68/100
pingdotgg/t3code#14536 · 2 commenti ·
I maintainer di solito rispondono entro 1 giorno
-
bug via-triage
Difficoltà 2/5 1-3 ore Idoneità per principianti 78/100
pingdotgg/t3code#14452 · 1 commento ·
I maintainer di solito rispondono entro 1 giorno
-
Difficoltà 2/5 1-3 ore Idoneità per principianti 84/100
I maintainer di solito rispondono entro 1 giorno
-
Difficoltà 2/5 1-3 ore Idoneità per principianti 84/100
I maintainer di solito rispondono entro 1 giorno
Tutte le issue di pingdotgg/t3code
Issue simili
-
Difficoltà 2/5 1-3 ore Idoneità per principianti 85/100
siyuan-note/siyuan#20040 ·
I maintainer di solito rispondono entro 1 giorno
-
enhancement
Difficoltà 2/5 1-3 ore Idoneità per principianti 72/100
RunestoneInteractive/rs#1574 ·
I maintainer di solito rispondono entro 2 giorni
-
Difficoltà 2/5 1-3 ore Idoneità per principianti 68/100
remotion-dev/remotion#11901 ·
I maintainer di solito rispondono entro 1 giorno
-
Poll constructor throws for an uncached channel while resolving a message context-menu interactionApertabug need repro packages:discord.js
Difficoltà 2/5 1-3 ore Idoneità per principianti 82/100
discordjs/discord.js#11645 ·
I maintainer di solito rispondono entro 3 giorni
-
🐞 bug
Difficoltà 2/5 1-3 ore Idoneità per principianti 72/100
Sitecore/content-sdk#641 ·
I maintainer di solito rispondono entro 2 giorni