Skip to content

vector-gpui: native GPUI client (experiment) - #102

Draft
JSKitty wants to merge 912 commits into
masterfrom
gpui
Draft

JSKitty wants to merge 912 commits into
masterfrom
gpui

Conversation

@JSKitty

@JSKitty JSKitty commented Nov 11, 2026

Copy link
Copy Markdown

Requested by JSKitty · project thread

A long-lived experiment branch: Vector drawn natively with GPUI instead of a webview, aiming for exact parity in layout, palette, fonts and features with the Tauri client. It isn't meant to merge any time soon; the PR is here so the diff and discussion stay in one place.

What's in this first cut

  • crates/vector-gpui: its own workspace and lockfile, so it never touches the Tauri client's dependency graph. The only shared-file change is excluding it from crates/Cargo.toml. GPUI is pinned to Zed f8cfc7e.
  • The widescreen shell: window chrome with client-side decorations, the rail (Messages, New Messages, Communities, Collapse, account), the chat list (virtualised uniform_list) and the open chat (virtualised list with date and NEW dividers). The NEW divider stays put while the chat is open.
  • The composer: real text input with selection, clipboard and IME. Enter sends a local echo; the send pipeline isn't wired up yet.
  • Types come straight from vector-core (SerializableChat, SlimProfile, Message). data.rs mirrors the JS derivations (getName, chatPreviewText, timeAgo, unread counts), and each helper names the function it copies.
  • Icons are compiled in from src/icons, so an icon changed in the Tauri client arrives here on the next build. Rubik is cut from the shipped variable font into static weights; Inter is bundled too.

Parity harness (reference/)

capture.mjs boots the real frontend in Chromium through its debug hot-reload path, using mock-tauri.js and the same fixtures.json. It needs no Rust, account or network. It saves a screenshot plus every element's box, font, colour and border. shot.sh renders vector-gpui under Xvfb, and compare.py puts the two side by side with a diff.

On the 1200×760 chat screen, 1.8% of pixels differ by more than 24/255. Most of that is text anti-aliasing; the remaining offsets are about 1px.

Numbers (release build, Linux, software Vulkan)

  • Window on screen 45 ms after launch.
  • About 42 MB of private anonymous memory. Total RSS is 155 MB here, but ~70 MB of that is Mesa's software renderer (libLLVM, lavapipe), which a real GPU driver doesn't load.
  • Binary is 27.8 MB stripped, including 4.4 MB of fonts.

Known gaps

  • Only the widescreen layout so far, not slim or the Centre Console.
  • No live account yet: it reads fixtures, not vector-core's DB or relays.
  • GPUI has no letter-spacing, so the NEW label spaces its letters by hand.
  • Ellipsis truncation differs slightly from CSS: CSS keeps the trailing space before ….

Generated by Claude Code

@JSKitty JSKitty self-assigned this Nov 11, 2026
@github-actions

github-actions Bot commented Nov 11, 2026 •

Copy link
Copy Markdown

Claude encountered an error —— View job


I'll analyze this and get back to you.

JSKitty commented Nov 11, 2026

Copy link
Copy Markdown
Author

The claude check failed, but the cause isn't this PR. The Claude Code workflow runs on every opened PR, and it stops at its environment validation because ANTHROPIC_API_KEY is empty in the job: no API key, OAuth token or federation rule is set for the repo. The same job failed on the fix-audio-crackling PR 40 minutes earlier for the same reason. This PR doesn't touch .github/, and adding a repo secret is beyond what a PR can do, so there's no fix to port here. It goes green when a maintainer adds ANTHROPIC_API_KEY (or CLAUDE_CODE_OAUTH_TOKEN) to the repository secrets. Alternatively, dropping the bare pull_request trigger from .github/workflows/claude.yml stops it running unasked.


Generated by Claude Code

Comment thread .github/workflows/gpui.yaml Fixed
Comment thread .github/workflows/publish.yaml Fixed

JSKitty commented Nov 13, 2026

Copy link
Copy Markdown
Author

The vector-core failure on b72983b is community::v2::service::tests::a_mint_whose_delivery_lost_a_part_leaves_nobody_out ("the part a relay dropped served nobody"). It comes from master and isn't caused by this PR: the only vector-core file this branch changes is text_color.rs. The test is flaky on master's code: it failed 1 of 6 local runs at the merged master tip. A likely cause is that which recipients end up in the dropped second part depends on iteration order, so sometimes that part holds none of the writers the assertion expects. Leaving it for the master side rather than patching community code on this branch.


Generated by Claude Code

JSKitty commented Nov 14, 2026

Copy link
Copy Markdown
Author

The build (macos-26-intel) failure on 6942f69 has no failed step and no log (the job log returns BlobNotFound), so it looks like the runner was lost rather than a build or test error. That commit is superseded, and CI is running on the current head 3a2d305.


Generated by Claude Code

JSKitty commented Nov 14, 2026

Copy link
Copy Markdown
Author

vector-core on 2560e5b failed in control_sync::a_relay_that_caps_open_sessions_is_never_written_off_and_every_plane_still_compares (its first assert, with 24 concurrent reconciles against an 8-session cap). That commit only touches the vendored GPUI in crates/vector-gpui, and the test and the transport it covers are unchanged from master. It passes 6 out of 6 times locally, so this looks like a timing flake in a master test, not this PR's change.


Generated by Claude Code

JSKitty commented Nov 14, 2026

Copy link
Copy Markdown
Author

build (macos-26-intel) on ce4c294 failed in the Tests step with "The hosted runner lost communication with the server". Build passed, and there is no log (BlobNotFound). This is the same runner loss as on 6942f69, not a code failure. The newer pushes (9b3b16f, e5d3370) run the job again.


Generated by Claude Code

JSKitty commented Nov 15, 2026

Copy link
Copy Markdown
Author

The vector-core test failure on 10ca012 (control_sync::a_relay_that_caps_open_sessions_is_never_written_off_and_every_plane_still_compares) isn't caused by that commit, which only touches crates/vector-gpui (code highlighting). The test hit two "too many open negentropy sessions" refusals under its cap, which looks timing-dependent on a loaded runner. The next push (b9d7ce0) reruns the suite.


Generated by Claude Code

JSKitty commented Nov 15, 2026

Copy link
Copy Markdown
Author

CodeQL's two new high alerts on 752bdc2/56616efd ("Access of invalid pointer", crates/vector-gpui/vendor/gpui_windows/src/platform.rs:1107 and :2001) are in upstream Zed code that came in with the verbatim vendoring of gpui_windows at the pinned rev (a0f3fd0); our only change there is the open dialog's file-type filter. Both dereference a CREDENTIALW that CredReadW just filled: the first after an explicit null check, the second in an upstream test. Neither is reachable with an invalid pointer, so I'm leaving them as they are and not patching upstream code for it.


Generated by Claude Code

JSKitty commented Nov 15, 2026

Copy link
Copy Markdown
Author

The vector-core failure on 7a4ebb0 is community::v2::service::tests::a_mint_whose_delivery_lost_a_part_leaves_nobody_out ("the part a relay dropped served nobody", service.rs:30010), a timing-sensitive relay test. The only vector-core change since the last green run is a defaulted MediaSource::remote() (returns false, used only by the GPUI stream reader), which no community code calls, so this is not from this PR. The next run on 689115a will retry it.


Generated by Claude Code

}

// SAFETY: `CredReadW` succeeded, so this points to a valid `CREDENTIALW` until `CredFree` below.
let result = unsafe { username_and_password(&*credentials) };

let mut credentials: *mut CREDENTIALW = std::ptr::null_mut();
unsafe { CredReadW(target_name, CRED_TYPE_GENERIC, None, &mut credentials) }.unwrap();
let result = unsafe { username_and_password(&*credentials) };

JSKitty commented Nov 15, 2026

Copy link
Copy Markdown
Author

Windows build on b16933a fails one new test, backend::stream::tests::a_reader_in_two_distant_regions_fetches_each_byte_once: a read there timed out after 45 s, while the test passes on macOS. It is a real intermittent stall in the region-aware stream fetcher (a waiting read that never gets served), not a flake to retry. The single-task fetch loop is being replaced with per-connection tasks, and that change will carry the fix along with a repeated run of this test.


Generated by Claude Code

JSKitty commented Nov 16, 2026

Copy link
Copy Markdown
Author

The vector-core failure on a365815 is community::v2::service::tests::control_sync::a_relay_that_caps_open_sessions_is_never_written_off_and_every_plane_still_compares. It isn't from this commit: a365815 changes only crates/vector-gpui/src/emoji.rs, and the same vector-core code passed on the commit before it (7bbd8ea), so it looks like that test is timing-sensitive on the runner. The other 2893 tests passed.


Generated by Claude Code

JSKitty commented Nov 16, 2026

Copy link
Copy Markdown
Author

The Tests failure on d748a88 is community::v2::service::tests::a_mint_whose_delivery_lost_a_part_leaves_nobody_out in vector-core ("the part a relay dropped served nobody"). It isn't from this push: d748a88 changes only crates/vector-gpui, and the same vector-core tree passed Tests on ba90624 an hour earlier. It looks like a timing-dependent test, so I'm leaving it alone here.


Generated by Claude Code

JSKitty commented Nov 17, 2026

Copy link
Copy Markdown
Author

The build (macos-26-intel) failure on ca099ef is infrastructure, not code: GitHub's annotation says the hosted runner lost communication with the server mid-build, after 75 minutes. The run for the current head (46776a2) is in progress and covers the same code.


Generated by Claude Code

JSKitty commented Nov 17, 2026

Copy link
Copy Markdown
Author

The vector-core failure on 96d65fa (a_sibling_that_wins_the_tie_over_the_held_head_holds_a_mint_back) is a rare flake in master's own test, not this PR: it grinds 1000 random rumors for an id below the held head's and unwraps the find, which comes up empty about 1 run in 1000 when the head's id happens to be small. This branch doesn't touch that file. I've rerun the failed job.


Generated by Claude Code

JSKitty commented Nov 18, 2026

Copy link
Copy Markdown
Author

The red vector-core job on bf4107b is one master test, not this PR's code: control_sync::a_relay_that_caps_open_sessions_is_never_written_off_and_every_plane_still_compares failed the same way on master's own run at 32dcb9a (run 44124478117), and this branch's diff to master doesn't touch community/transport.rs, plane_neg.rs or that test. It passed every one of over 100 local runs, many of them in parallel or pinned to one core.

Likely cause, proposed rather than pushed since it changes core timing: plane_neg::reconcile's CloseGuard sends NEG-CLOSE from a spawned task, while LiveTransport::reconcile drops its neg_permits permit as soon as the call returns. The next waiter can send its NEG-OPEN before those closes reach the socket, so under a starved runtime the relay briefly counts more than NEG_SESSIONS_PER_RELAY sessions and answers "too many open negentropy sessions". Proposed fix: acquire with acquire_owned() and move the permit into CloseGuard, dropping it only after the NEG-CLOSE send_msg completes.


Generated by Claude Code

A content mask now carries corner radii: an overflow-hidden element with rounded
corners clips what it holds to its padding box's rounding (its radius less the border),
mirrored with the layout. Quads, shadows, underlines, glyphs and images take the mask's
coverage in their fragment shaders on Metal, DirectX and wgpu (WebGL decoders too).
Menu rows, cards and players no longer poke square corners past a curved edge.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Qq6RetJnzXvEeXAvSsgvs1
A content mask can carry vertical alpha ramps, applied in every primitive's fragment
shader like the rounding, so a scroller fades its ends over any background, as Tauri's
mask-image does. `edge_fade(depth)` on a scrolling div or list fades each end by as much
as is left to scroll that way, up to depth: none at rest against an end. The chat and
channel lists (wide layout) and the Community Rail fade over 24 px, as Tauri's do.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Qq6RetJnzXvEeXAvSsgvs1
…oom under the lists

Opening a chat marked read through the backend's in-memory copy, which can hold only our
own latest message and then marks nothing, so the UI's marker never moved and the NEW
divider came back on every open. Every read now names the newest message from someone
else, as Tauri's markAsRead does.

The chat list and the channel list end with Tauri's bottom padding (8 px wide, 16 px slim).

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Qq6RetJnzXvEeXAvSsgvs1
…ning mask fades

The mask's fade gains left and right ramps, and a box can fade over fixed lengths as well
as by the scroll left (Styled::edge_fades, Styled::fade_out). The lyrics sheet, clamped Nostr
embeds, the collapsed edit history, the kaomoji tabs and the App Store shelves now fade as
Tauri's masks do instead of under painted gradients, and the narrow layout's chat and channel
lists fade too.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Qq6RetJnzXvEeXAvSsgvs1
Hover state read at layout only updated on mouse moves, so a row that came under a still
pointer read unhovered at layout and hovered at paint, flipping its hover fade each frame.
The paint-time hit test now writes the state layout reads.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Qq6RetJnzXvEeXAvSsgvs1
…age, first-launch offer

- vector-i18n: the index format (a kind 30078 event's content; a key hands over by naming
  its successor), English names for packs, en-* packs inherit en-GB for coverage, and the
  languages this build saw in the repository for offering before any download.
- vector-i18n tool (feature sign): publish builds every language and signs the index,
  keygen makes the key, blossom-auth authorises one upload with a key of its own.
- GPUI: the index is fetched on the shared lane and believed only when the pinned key (or
  a successor it named) signed it; packs come from Blossom by hash, then the release, and
  are checked against the index's hash and size. Installed packs update daily and swap in
  live; a device with none never asks.
- Settings > Appearance > Language lists every language with its English name, coverage
  and size; picking one downloads it with progress and switches at once.
- The welcome screen offers the system's language once, downloading only on Switch.
- The en-US pack, and a CI job that publishes once VECTOR_LANGUAGES_KEY is set. The
  pinned key is a throwaway test key until then.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Qq6RetJnzXvEeXAvSsgvs1
…an't reach the release

The app asks the release first and Vector's trusted relays when it can't, through a scratch
client on the shared lane, following each key handover the same way. CI hands the signed
events to those relays after the release upload; a refusal is only a warning.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Qq6RetJnzXvEeXAvSsgvs1
…equests too

A broken message fails the check without holding back the other languages' publish.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Qq6RetJnzXvEeXAvSsgvs1
…ownload

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Qq6RetJnzXvEeXAvSsgvs1
vector_core::search folds every message into one buffer and scans it with
memchr's memmem: case and accent folded, CJK without word breaks, quoted
phrases, and from:, mentions:, has:, in:, before:, after:, during: filters.
GPUI opens it with Cmd/Ctrl+K (every chat) or Cmd/Ctrl+F, the header's
magnifier and the chat menu (the open chat), suggests the filter being
typed, and opens a result at the message. Slim takes the whole window.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Qq6ReurguGC2qWNpRCjenb
Errors, network status, moderation policy wording and the like come from
vector-core in English. coretext matches each line back to the format
string that wrote it and says it through a core.ftl message, with the
values it carried (counts as numbers, so translations can pluralise).
Composite lines (policy rule details, clock skew, nested errors) are
taken apart first. Shown at toasts, popups, dialog errors, network
status, presets, raid and policy reasons, failed sends, embeds and Mini
App permissions.

Two tests keep it honest: a table entry core no longer writes fails, and
new prose in vector-core counts against a per-file allowance until it
gets an entry or is judged developer-only.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Qq6RetJnzXvEeXAvSsgvs1
…ts edge

The system's first search for fonts that draw the kaomoji (the picker's tab label
among them) ran inside the picker's first frame; a background pass shapes their
glyphs once after launch. The grid's overdraw drops from 400 px to two rows.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Qq6RetJnzXvEeXAvSsgvs1
…s in its own

Their entrances and hovers asked the root for frames, so each one redrew the whole
window; now only the pane that moves is drawn again. The attach menu asks its own
view for frames instead of notifying the root.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Qq6RetJnzXvEeXAvSsgvs1
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Qq6RetJnzXvEeXAvSsgvs1
Follow the tail only when the timeline is pinned to it, as Tauri latches
its pin once the user scrolls up.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Qq6RetJnzXvEeXAvSsgvs1
…ice sign-in

A touch catches a fling only while it visibly moves what it scrolls; a
fling stopped by a list's end is over, and Android flings with its own
OverScroller physics. Held picks are read through their own descriptor,
which Android lets an app do, unlike reopening the memfd by path. Long
file names wrap inside the preview card. A phone being signed in opens
on Scan QR Code. Pin to Main Menu shows only in widescreen.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Qq6RetJnzXvEeXAvSsgvs1
… none

A muted subtree keeps drawing but its hitboxes contain no point and its
mouse listeners never run, so a tap during a popup, menu or drawer's
fade-out reaches the app beneath instead of an inert shield.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Qq6RetJnzXvEeXAvSsgvs1
…bar in its tab, media at its width on a chat's first frame

- A message moves only its own chat in the list: 1000 arrivals over 200 chats of 300 messages
  went from 446 ms to 19 ms of main-thread work.
- The UI drains backend events in 4 ms slices, and the Android main queue yields to the looper
  after 4 ms, so touches and vsync are served during a sync.
- The slim tab underline rests inside its tab and only slides from measured bounds.
- The timeline keeps its last width per layout for a new chat's first frame.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Qq6RetJnzXvEeXAvSsgvs1
LineLayoutCache kept only last frame's layouts, so a chat, Community, tab or drawer drawn again
reshaped all its text in its first frame. Lines a frame stops drawing are now retained (bounded
at 3072 lines and 1536 wrapped lines, oldest trimmed). Revisiting a Community: worst frame
9.5 → 4.4 ms, 67 → 0 shape misses; a chat with videos 8.4-11.5 → 4.6-4.9 ms.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Qq6RetJnzXvEeXAvSsgvs1
The root render ticked chat models, so a spinning model's frame requests and engine wakes
redrew the whole window at 30 fps for as long as it played. The timeline pane now asks for
its models' frames; the root only asks while the full-screen viewer is open. Two seconds
after opening a chat with a spinning model: whole-window redraws 98 → 3 cold, 59 → 0 warm.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Qq6RetJnzXvEeXAvSsgvs1
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Qq6RetJnzXvEeXAvSsgvs1
…at list is its own pane

- An inline clip ticks in the timeline pane, so its frames redraw that pane, not the window
  (whole-window redraws over 2.5 s of playback: 301 → 0).
- A playing clip asks for its next redraw when its next decoded frame falls due, not every
  vsync: a 30 fps clip draws 30 times a second (123 → 30/s), drawing time per second of
  playback down 58%, every frame still shown.
- At phone width the chat list draws as a cached pane, so scrolling it leaves the rest of the
  window alone (whole-window redraws during a scroll: 324 → 0).

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Qq6RetJnzXvEeXAvSsgvs1
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Qq6RetJnzXvEeXAvSsgvs1
A new `keyed` element in vendored GPUI draws its subtree once and, while
nothing but its position changes, replays last frame's primitives and hit
areas moved to where it now sits. It records with its own clips only and
takes the clips around it afresh, so edge fades and partial rows stay right.

Views now hear two kinds of notify: data (rebuild) and motion (a scroll or
hover moved things). List and overflow scrolls and hover flips are motion.
A subtree rebuilds when its view's data changes, the window redraws
everything, the pointer is over it, it asked for a frame, or it holds
tooltips, deferred draws or text input. List rows sit on whole device
pixels so a scroll moves them by whole pixels. GPUI_NO_REPLAY=1 turns it off.

Both chat lists use it. Scroll frame on the cloud box (200 chats):
wide 2.10 -> 1.29 ms mean, phone width 2.23 -> 1.83 ms. Screenshots after
a scroll match the no-replay build pixel for pixel; taps right after a
scroll open the same chat in both (8/8).

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Qq6RetJnzXvEeXAvSsgvs1
draw_stats splits each draw into the root's request layout (renders), layout,
prepaint and paint; dev:state reports them as request_layout_prepaint_paint_ms.
dev:pan:x,y,dy,n sends n pixel scrolls, one a frame, as a finger pan does.

Finger-like pans (12 px a frame) over 200 chats, keyed replay on vs off:
wide 2.17 -> 1.20 ms a frame, phone width 2.27 -> 1.67 ms.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Qq6RetJnzXvEeXAvSsgvs1
…e again

As a layout root of its own, a slim row (height only) shrank to its content:
previews stopped truncating and the time ran off the edge. The build is now laid
out inside a box of the slot's size.

Chat list rows key on the minutes since their last message, so a replayed row
never shows a stale "2m", and the list redraws every 30 s as Tauri's list
clock does. Stock emoji picker rows are keyed too. devctl `views:1` times each
view's prepaint and paint (panes by name) into dev:state.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Qq6RetJnzXvEeXAvSsgvs1
…caches

A cached view that misses used to redraw every cached view inside it. When it
misses only because a view inside it was notified, its render is unchanged, so
the views inside may reuse their drawing, as panes under the root already do.
Scrolling the emoji picker no longer redraws the timeline beside it.

refresh_windows now counts as a hard refresh, so keyed rows redraw on a
language switch. GPUI_NO_REPLAY=1 also turns this off, for A/B.

Emoji picker finger pan, chat open: 5.0 -> 3.0 ms a frame, prepaint 2.0 ->
0.78 ms, timeline 1.26 -> 0.03 ms. Typing in the composer: 2.8 -> 2.2 ms.
Screenshots identical on vs off (picker, typing, wide and slim lists).

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Qq6RetJnzXvEeXAvSsgvs1
The code block's 10 px scrollbar strip blocked the mouse, wheel included, and
did so even with nothing to scroll sideways, so the timeline stalled whenever
the pointer rested there. The bar now lets a wheel through, as a page's does,
and without overflow the strip is no bar at all.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Qq6RetJnzXvEeXAvSsgvs1
…s; editor × over the edge

Timeline rows replay their last drawing when scrolled (keyed_measured: any height,
measured on build and kept while replaying). Rows that are loading, animating or
show a relative time mark themselves volatile and redraw. Selection geometry is
re-placed on replay (Window::place). Plain rows keep the same element id so hover
state survives the plain/keyed switch on the first frame.

3000-message chat pan, 1280x800: 5.1 -> 3.2 ms a frame, 73% of rows replayed.
Screenshots and drag selection after panning match GPUI_NO_REPLAY=1 exactly.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Qq6RetJnzXvEeXAvSsgvs1
… overlays [package]

`.backdrop_blur_once(r)` blurs what lies behind an element on the first frame it shows
(quarter size, two 13-tap passes) and reuses that texture while it stays put, so held
frames draw one extra quad and keep the main pass whole. Metal and wgpu read the frame
back; a GL surface without COPY_SRC draws that one frame offscreen and copies it over.
DirectX leaves the backdrop sharp for now. GPUI_NO_BACKDROP=1 turns it off.

Mac GPU: +0.15 ms on the opening frame (1.10 vs 0.94 ms), held frames within noise.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Qq6RetJnzXvEeXAvSsgvs1
…-pointer hovers

The scrollbar's settle redraw of the chat pane is a motion notify, so it no longer
rebuilds every visible row. Content scrolling under a still pointer flips hover
without a fade, so rows carried past it don't animate (and rebuild) for 150 ms.
A keyed record's hover check covers its hit areas, not only its bounds: a hover
toolbar hanging above its row kept a stale highlight on replay.

3000-message chat pan, 1280x800: 3.3-3.4 -> 2.5-3.1 ms a frame, rows replayed 70% -> 85%.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Qq6RetJnzXvEeXAvSsgvs1
… around them

GPUI marks every view holding a notified one dirty, so a timeline scroll redrew the
whole chat pane (header, composer) around it. A cached view now records the cached
views drawn inside it (slots: ranges, dispatch parent, draw stacks). When it is dirty
but not notified and every dirty view inside it lies in one of its slots, it reuses
its last prepaint and paint in pieces around those slots and draws only them afresh
(OpenReuse keeps the dispatch nodes still open between pieces). Anything it cannot
see into (deferred draws in a reused piece, an uncached dirty view, a11y, zoom,
inspector) falls back to a full redraw. Cached view ranges are only reused from the
last draw. GPUI_NO_PUNCH=1 turns it off.

The scrollbar, the way back, the unread pill and the typing line move into their own
pane, redrawn in the same frame as the timeline (Window::redraw_now).

3000-message chat pan, 1280x800: 2.34-2.76 -> 2.22-2.32 ms a frame, prepaint
1.02-1.27 -> 0.80-0.90 ms. Screenshots, drag selection and typing match GPUI_NO_PUNCH=1.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Qq6RetJnzXvEeXAvSsgvs1
…on screen first

A fresh join showed General's messages only after a multi-page catch-up
(35 s on a real join), raced by a sweep of every channel; profiles loaded
one REQ pair at a time in join order, and an empty channel's older-page
fetch spun on a refused claim.

- The opened channel fetches its newest page alone first and paints it;
  sweeps, resyncs, the volley and core's channel rereads wait on a
  session foreground gate, and take the channel on screen first.
- Visible profiles go to the front of their lane in draw order; the
  watcher keeps draw order and holds back anything over a batch rather
  than dropping it; unknown Community members and join rows queue behind.
- The profile processor loads a batch with one REQ for every author's
  metadata and status.
- An empty channel does not fetch older pages; a refused claim backs off.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Qq6RetJnzXvEeXAvSsgvs1
The root view is now cached and punches through like the panes: a frame
dirty only for panes renders those alone. Its per-frame upkeep moves to
Window::on_each_draw, eases built at layout ask for frames through
after_each_prepaint, and the chrome pane (or upkeep that changes what the
root draws) makes it render again via Window::rerender. Deferred draws in
reused pieces are carried with their dispatch parents remapped.

Timeline pan (3000 messages, 1280x800, cloud): 2.22-2.32 -> 2.03-2.10 ms
a frame. Screens, selection, popups, menus and tooltips match
GPUI_NO_PUNCH=1.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Qq6RetJnzXvEeXAvSsgvs1
sync_channel_events pages on for its continuations after painting the
first window, holding the foreground gate (and the background walks
behind it) for seconds more. sync_community_channel_head fetches and
ingests one window, then the catch-up runs as before.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Qq6RetJnzXvEeXAvSsgvs1
Each address of a page was read after the one before it, and a channel's
first paint paid every one of those rounds in series. They are separate
reads whose results fold in the same order, so they now run together.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Qq6RetJnzXvEeXAvSsgvs1
…tes [package]

A neighbour update carries a count and no names; it no longer empties the
card's player list. The pill counts named players, else the mesh's count,
and each avatar names its player on hover.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Qq6RetJnzXvEeXAvSsgvs1
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

3 participants