Local UI to run and train LLMs and diffusion models. Supports GGUF, MLX, Qwen3.8, DeepSeek-V4.1, MiniMax-H3, Gemma 4, FLUX and more.
Python agentaichatgptdeepseekfine-tuninggemmaimage-generationllamallmllmsopenaipythonqwenreinforcement-learningself-hostedstable-diffusiontext-to-speechttsuiunsloth
Studio: open and close dialogs and menus in long chats without a full thread re-layout on Firefox (#17357)
Daniel Han0c46cdbaa28b
diff --git a/studio/frontend/src/index.css b/studio/frontend/src/index.cssindex bdc45d73a04..2e4b7a6d208 100644--- a/studio/frontend/src/index.css+++ b/studio/frontend/src/index.css@@ -2186,6 +2186,20 @@ html[data-chat-font] .aui-root {z-index: 49;pointer-events: auto;}+ /* Firefox only: the cover is always there and the lock only flips its pointer-events. Adding it per+ open made Firefox lay out the thread root's flex column again, measuring every message (about+ 250 ms per modal open and close at 150 turns, 20 ms at 1 turn). Inert when unlocked: no paint,+ pointer-events: none. Not for WebKit: a permanent cover made the Glass browser pane open there+ about 60 ms slower at 150 turns. */+ @supports (-moz-appearance: none) {+ .aui-thread-root::after {+ content: "";+ position: absolute;+ inset: 0;+ z-index: 49;+ pointer-events: none;+ }+ }/* The side pane (browser, research) sets its own pointer-events: auto once shown, so it neverinherited the lock and its scrollers (file view, page) took wheel and touch under a modal menu.The same cover over it blocks them as react-remove-scroll's document listeners did, without adiff --git a/studio/frontend/tests/modal-lock-thread-cover.test.ts b/studio/frontend/tests/modal-lock-thread-cover.test.tsindex f0ba5999bc0..9c3edb0c930 100644--- a/studio/frontend/tests/modal-lock-thread-cover.test.ts+++ b/studio/frontend/tests/modal-lock-thread-cover.test.ts@@ -42,4 +42,24 @@ test("while the lock is up a cover keeps the thread unclickable, below portaled);assert.equal(CSS.split(".aui-thread-root[data-modal-lock-thread]::after").length - 1, 1);assert.equal(CSS.split(" .aui-thread-root::after").length - 1, 1);+ // Plus the Firefox-only permanent cover below.+ assert.equal(CSS.split("\t.aui-thread-root::after {").length - 1, 1);});++test("in Firefox the cover is permanent and inert, so a lock only flips its pointer-events", () => {+ // Adding the cover on each open made Firefox lay out the thread root's flex column again, every+ // message included (about 250 ms per modal open and close at 150 turns). Firefox only: in WebKit a+ // permanent cover slowed the Glass browser pane open.+ const at = CSS.indexOf("@supports (-moz-appearance: none) {\n\t\t.aui-thread-root::after {");+ assert.ok(at >= 0, "missing the Firefox-only permanent cover");+ const block = CSS.slice(at, CSS.indexOf("}", at) + 1);+ assert.match(block, /content: "";/);+ assert.match(block, /position: absolute;/);+ assert.match(block, /inset: 0;/);+ assert.match(block, /z-index: 49;/);+ assert.match(block, /pointer-events: none;/);+ assert.doesNotMatch(block, /background|backdrop-filter|box-shadow/);+ // Lower specificity than the lock rule, which sets pointer-events: auto.+ assert.ok(at > CSS.indexOf("[data-modal-lock-region] .aui-thread-root::after {"));+});+