Now, this is one of these beautiful cases where GitHubs API ist just so
pleasant to work with: Because PRs are treated as issues, assigning an
assignee to a PR suddenly requires issue write permissions, despite the
issue in question being a PR. Not having that permission resulted in
some missing assignees on zed zippy bumps and failures of the workflows
as seen in
https://github.com/zed-industries/zed/actions/runs/31339736929/job/93311493258.
In comparison, labelling PRs requires PR write permissions as seen in
https://github.com/zed-industries/zed/pull/61525🤡
Beautiful API and a pleasure to work with, 10/10 would recommend.
Release Notes:
- N/A
Recommends GPT-5.6 Sol for both OpenAI BYOK and OpenAI subscription.
Also unifies the naming so that we use the same OpenAI model names for
the Zed/OpenAI BYOK and OpenAI subscription providers.
Release Notes:
- N/A
This reverts commit 6297c88f42.
---
fixes#62286fixes#62095https://github.com/zed-industries/zed/pull/61467 fixed its intended bug,
but at the same time introduced an issue where running tasks that would
cause new tasks to be terminated immediately.
https://github.com/zed-industries/zed/pull/62322 tried to fix that
forward, but was unsuccessful. In the mean-time I am going to revert the
original PR.
We can try to re-land the original bugfix in a future PR.
Release Notes:
- N/A
Follow-up to #59838, implementing what was discussed at the end of
#59829: cmd-click navigation now respects `lsp_results_location` when
go-to-definition falls back to find-all-references (invited in
https://github.com/zed-industries/zed/issues/59829#issuecomment-4989966493:
"It would! Feel free to hook that up if you'd like to!").
## Problem
Cmd-clicking a symbol's definition falls back to find-all-references,
but the results always open in a multibuffer even with
`"lsp_results_location": "picker"`. The hover-link click path calls the
editor navigation methods directly, so the action handlers registered by
`lsp_locations` never get a chance to intercept.
## Solution
- `handle_click_hovered_link`'s fallback now dispatches the
`FindAllReferences` action (with `open_results_in: None`, deferring to
the global setting) instead of calling the method, so the
`lsp_locations` handler can intercept it, or propagate to the editor's
built-in handler when the setting is `multi_buffer`, preserving today's
behavior exactly.
- The plain cmd-click arm of `cmd_click_reveal_task` now runs the
definition query via `go_to_definition_of_kind` (no internal references
fallback) instead of `go_to_definition`, so the click path has a single
fallback decision point: the dispatching one. Without this, the method's
baked-in fallback opened a multibuffer before the dispatch could run.
- `go_to_definition_of_kind` visibility widened to `pub(crate)` for the
call from `hover_links.rs`.
Shift/alt click variants (type definition, splits) are untouched.
Keyboard invocations were already intercepted and are unchanged.
## Testing
- New test `test_cmd_click_fallback_honors_lsp_results_location` in
`lsp_locations`, following the module's existing test patterns: fake LSP
returning no definition and two references, `lsp_results_location:
picker`, simulated cmd-click at the cursor's pixel position, asserts the
picker opens. The test fails without this change.
- `cargo nextest run -p lsp_locations`: 6/6.
- `cargo nextest run -p editor -E 'test(hover) or test(fallback) or
test(go_to_definition) or test(references)'`: 54/54.
- `cargo fmt` and `./script/clippy` clean.
- Verified manually in a release build: with the setting on,
cmd-clicking a definition opens the picker; with it off, behavior is
unchanged.
Per the contributing guidelines' note on AI assistance: this change was
developed with heavy AI assistance (Claude Code). I have reviewed and
understand the full diff and the reasoning behind each hunk, and I'm the
one answering review feedback.
Release Notes:
- Fixed cmd-click go-to-definition falling back to a references
multibuffer even when `lsp_results_location` is set to `picker`.
# Objective
The `gpui::img` element always overrides the `aspect_ratio` field, so if
you have an image element that applies its own `.aspect_ratio()` it just
gets wiped out.
## Solution
Only apply the aspect ratio default if one is not already set.
## Testing
It's a very minor change but I did add a small test to ensure it
actually gets overridden.
## Self-Review Checklist:
- [x] I've reviewed my own diff for quality, security, and reliability
- [x] Unsafe blocks (if any) have justifying comments
- [x] The content adheres to Zed's UI standards
([UX/UI](https://github.com/zed-industries/zed/blob/main/CONTRIBUTING.md#uiux-checklist)
and
[icon](https://github.com/zed-industries/zed/blob/main/crates/icons/README.md)
guidelines)
- [x] Tests cover the new/changed behavior
- [x] Performance impact has been considered and is acceptable
## Showcase
This came from an issue where vertical images inside an element (square
in this case, as you'd see in an image gallery) do not behave correctly
with `object_fit` values of `ObjectFit::Contain` or
`ObjectFit::ScaleDown`.
This was simply because despite the image element having a fixed square
size (`img().size(px(200.))`), the aspect ratio would be forced to the
ratio of the image itself, so vertical images weren't being properly
fitted into their containers.
Minimal repro for that issue:
https://github.com/zaknesler/gpui-object-fit
So with this change, you can set `.aspect_square()` and the object fit
will behave as you'd expect:
<img width="1237" height="986" alt="image"
src="https://github.com/user-attachments/assets/5b1045a8-bd71-4b77-8bbd-c3b12112bcb0"
/>
---
Release Notes:
- gpui: Fix image element's aspect ratio overriding existing value
## Summary
- Add `terminal.starts_open` setting so the terminal panel can open
automatically in new workspaces, like project and git panels already can
- Expose the setting through terminal settings docs, default settings,
Settings Editor metadata, and the terminal panel implementation
- I also added a matching settings page item for the existing
`git_panel.start_open` setting
Self-Review Checklist:
- [x] I've reviewed my own diff for quality, security, and reliability
- [x] Unsafe blocks (if any) have justifying comments
- [x] The content is consistent with the [UI/UX
checklist](https://github.com/zed-industries/zed/blob/main/CONTRIBUTING.md#uiux-checklist)
- [ ] Tests cover the new/changed behaviour
- I decided not to add tests as the behaviour seems covered by the
existing settings tests, and similar areas don't seem to have their own
explicit tests.
- [x] Performance impact has been considered and is acceptable
Related to #51542 (issue mentions possibly adding for all panels, closed
with implementation for `git_panel`)
Release Notes:
- Added `terminal.starts_open` to control whether the terminal panel
opens automatically in new workspaces
# Objective
Follow-up to zed-industries/extensions#7062.
## Solution
Add `("windows-batch", &["bat", "cmd"])` to
`SUGGESTIONS_BY_EXTENSION_ID`
## Self-Review Checklist:
- [x] I've reviewed my own diff for quality, security, and reliability
- [x] Unsafe blocks (if any) have justifying comments
- [x] The content adheres to Zed's UI standards
([UX/UI](https://github.com/zed-industries/zed/blob/main/CONTRIBUTING.md#uiux-checklist)
and
[icon](https://github.com/zed-industries/zed/blob/main/crates/icons/README.md)
guidelines)
- [ ] Tests cover the new/changed behavior
- [x] Performance impact has been considered and is acceptable
---
Release Notes:
- Added a suggestion to install the `windows-batch` extension when
opening `.bat` and `.cmd` files.
## Context
This is a rebase of #52302, which was approved by @Veykril in May but
still needed a rebase before it could be merged — it seems the original
author @MostlyKIGuess forgot about it. I rely on this fix daily for REPL
over SSH on an unstable connection and need it rather urgently, so I
cherry-picked the original commit (with authorship preserved) and
resubmitted it as a new PR.
Closes#51834
Supersedes #52302
## How to Review
The code itself was already reviewed and approved in #52302. This PR
only rebases it onto latest `main`, with one simple conflict resolution
against #53014: when all retries are exhausted, the error now uses the
improved message from that PR (suggesting `pip install ipykernel` on the
remote host) instead of the plain one. Comparing this diff against the
original PR's diff should make the review straightforward.
## Self-Review Checklist
- [x] I've reviewed my own diff for quality, security, and reliability
- [x] Unsafe blocks (if any) have justifying comments
- [x] The content adheres to Zed's UI standards
([UX/UI](https://github.com/zed-industries/zed/blob/main/CONTRIBUTING.md#uiux-checklist)
and
[icon](https://github.com/zed-industries/zed/blob/main/crates/icons/README.md)
guidelines)
- [x] Tests cover the new/changed behavior
- [x] Performance impact has been considered and is acceptable
Release Notes:
- repl: Fixed iopub connection failures over SSH on slow or unstable
connections by retrying with exponential backoff.
Co-authored-by: mostlykiguess <bruvistrue93@gmail.com>
the buffer search and project search bars both already load the `regex`
language and put it on their query buffer when the regex filter is on.
the text finder never got that, so a regex you type in there is just
plain text.
did the same thing here. the only slightly annoying part was that the
picker's head editor is `pub(crate)`, so text_finder couldn't get at it.
added a small `query_editor()` accessor to `Picker` for that.
`adjust_query_regex_language` mirrors the two existing ones.
test asserts the query buffer's language is regex with the filter on,
and gone once it's off.
Closes#59945.
Release Notes:
- Improved the text finder by highlighting the query as a regex when the
regex filter is on
---------
Co-authored-by: Kirill Bulatov <kirill@zed.dev>
# Objective
- Zed can hang (and eventually get force-killed) when opening certain
binary files, because `analyze_byte_content`'s UTF-16 heuristic
misclassifies them as UTF-16LE/BE text.
- Reproduced with a real-world case: a ~92 MB OTBM game map file (the
binary map format used by OpenTibia/Tibia servers), which interleaves
short ASCII strings with small u16 length/type fields. Its byte pattern
(mostly-zero high bytes, very few control characters) passed the
existing check, so Zed read the entire file, decoded it as UTF-16, and
opened it as an editable buffer with tens of millions of characters and
effectively no line breaks — a pathological case for the text
layout/renderer that hangs or crashes the app (most noticeably on
Windows).
## Solution
`is_plausible_utf16_text` in `crates/language/src/file_content.rs`
previously only rejected the UTF-16 hypothesis when too many code units
were control characters (> 2%). That's not sufficient on its own: binary
formats that interleave short ASCII fragments with small numeric fields
can have a very low control-character ratio while still not being real
text — most of their "characters" land on stray symbol/high-byte values
rather than letters, digits, or spaces.
This PR adds a second, independent requirement: at least 30% of the
analyzed code units must be letters, digits, or spaces (the bulk of any
real UTF-16 text sample). Both conditions now have to hold for a byte
sequence to be classified as UTF-16 text — otherwise it falls through to
`ByteContent::Binary`, and file loading is rejected early, as intended
for binary files, instead of decoding the whole file as garbled text.
## Testing
- Added `test_length_prefixed_binary_not_misdetected_as_utf16le` in
`crates/worktree/src/worktree.rs`, using a synthetic byte pattern that
reproduces the same statistical shape as the real file (null high bytes,
low control-character ratio, no word-like low bytes) — asserts it is now
classified `Binary`.
- Verified against the real 92 MB `.otbm` file that triggered the bug
(not committed, since it's user data): before the fix it was classified
`Utf16Le`, after the fix it's classified `Binary`.
- Ran the full existing `analyze_byte_content` /
`is_plausible_utf16_text` test suite (`cargo test -p worktree --lib
tests::`) — all 7 tests pass, including the pre-existing positive
UTF-16LE/UTF-16BE detection tests, so legitimate UTF-16 files are
unaffected.
- Built a full `--release` binary on Windows and confirmed opening the
real file now shows "Binary files are not supported" immediately instead
of hanging.
## Self-Review Checklist:
- [x] I've reviewed my own diff for quality, security, and reliability
- [x] Unsafe blocks (if any) have justifying comments — N/A, no unsafe
code
- [x] The content adheres to Zed's UI standards — N/A, no UI change
- [x] Tests cover the new/changed behavior
- [x] Performance impact has been considered and is acceptable — only
affects classification of the first 1 KB of a file, negligible cost
---
Release Notes:
- Fixed: Zed no longer hangs when opening certain binary files (e.g.
game asset/map formats) that were previously misdetected as UTF-16 text.
---------
Co-authored-by: Kirill Bulatov <kirill@zed.dev>
RPC log grouping only tracked whether consecutive messages had the same
transport direction. This made an untimed request appear beneath the
duration header of an unrelated response when both were sent in the same
direction.
Before:
// Send (took 53.0ms):
{"jsonrpc":"2.0","id":"##CodeLensRefreshRequest#1226","result":null}
{"jsonrpc":"2.0","id":53,"method":"textDocument/diagnostic"}
After:
// Send (took 53.0ms):
{"jsonrpc":"2.0","id":"##CodeLensRefreshRequest#1226","result":null}
// Send:
{"jsonrpc":"2.0","id":53,"method":"textDocument/diagnostic"}
Apply the same boundary handling to received messages while continuing
to group consecutive untimed messages.
---
Release Notes:
- N/A or Added/Fixed/Improved ...
Closes#62007
Adds Gemini 3.6 Flash to the Google AI provider so it can be selected in
the agent, alongside the existing Gemini 3.5 Flash.
Google released Gemini 3.6 Flash on July 21, 2026. It keeps the 1
million token context window, has a 64k output limit, and supports
thinking. The model is added with the same capabilities and thinking
levels as 3.5 Flash (Minimal, Low, Medium, High, defaulting to Medium).
The provider builds its model list from `google_ai::Model::iter()`, so
adding the enum variant is enough for it to show up in the model
dropdown. The thinking behavior in `completion.rs` keys off the
`gemini-3` id prefix, so it already applies to this model.
Release Notes:
- Added Gemini 3.6 Flash to the Google AI models
Co-authored-by: Claude Opus 4.8 <noreply@anthropic.com>
GPUI's declarative hover styles are recomputed from the current frame's
hit test, but `on_hover` listeners previously updated only in response
to mouse movement or the pointer leaving the window. When layout moved a
different element beneath a stationary pointer, the new element looked
hovered without receiving its hover-start callback.
This change reconciles each hover listener against the current hit test
during painting and defers transitions until after the paint cycle, when
application callbacks can safely update state. Reconciliation pauses
while a mouse press is pending so hover-only controls remain mounted
between mouse-down and mouse-up.
The regression coverage moves an element beneath and away from a
stationary pointer and verifies both transitions. It also verifies that
repainting during a stationary mouse press does not generate a spurious
hover exit.
Testing:
- `cargo test -p gpui --lib`
- `cargo fmt --check -p gpui`
- `cargo check -p delta --bin delta` with Delta temporarily patched to
the local GPUI checkout
Release Notes:
- Fixed hover interactions not updating when interface elements move
beneath a stationary pointer.
Spinners rendered near each other (like a sidebar full of running
threads) each start rotating from the moment their element first
renders, so they spin out of phase and the group looks jittery.
This adds `Animation::repeat_synced()`, which derives the phase from a
clock shared by the whole app instead of the element's mount time, and
switches `with_rotate_animation` over to it, so every spinner with the
same period now renders the same frame at the same time. The epoch is
read off the scheduler clock at app creation, so the phase stays
deterministic under the test scheduler.
Release Notes:
- Improved loading spinners to rotate in phase with one another
Currently, GPUI on web calls `requestAnimationFrame` every frame, which
causes it to do meaningful GPU work every frame, even if it was
identical.
This PR makes GPUI only call `requestAnimationFrame` when it has new
content to display.
---
Release Notes:
- N/A or Added/Fixed/Improved ...
The markdown preview used `vertical_scrollbar_for()` which always sets
the scrollbar to auto-hide mode, ignoring the user's `scrollbar.show`
setting.
This fix uses
`Scrollbars::for_settings::<EditorSettingsScrollbarProxy>()` so the
markdown preview respects the same editor scrollbar visibility setting
as regular editor panes (always, auto, system, never).
Changes:
- Replaced `vertical_scrollbar_for()` with `custom_scrollbars()` +
`Scrollbars::for_settings::<EditorSettingsScrollbarProxy>()` in the
markdown preview render method
Fixes#60380
Release Notes:
- Fixed `scrollbar.show` setting not being respected in Markdown
previews
---------
Co-authored-by: Akshit Jain <akshitj11@users.noreply.github.com>
Co-authored-by: MrSubidubi <finn@zed.dev>
# Objective
- Fixes#61925.
- GPUI already carries AccessKit author IDs through its platform
adapters, but applications cannot set one on an ordinary element.
## Solution
- Add `accessibility_id(...)` beside the other semantic builders.
- Store the value separately from GPUI's element ID and write it to
`Node::set_author_id`.
- Document that applications should keep the value stable and unique
within the accessibility tree.
- Document the platform mapping accurately: UIA `AutomationId` on
Windows, `AXIdentifier` on macOS, and AT-SPI `AccessibleId` on Linux
stacks whose deployed adapter exposes it.
## Testing
Exact candidate `d099c529`:
- `cargo fmt --all -- --check`
- `git diff --check`
- `cargo test -p gpui test_accessibility_id_builder_writes_author_id
--offline`: 1 passed, 218 filtered out
The test exercises the public builder and verifies that the author ID
reaches the AccessKit node. Validation is unit-level; this branch has
not run live platform UIA, AX, or AT-SPI end-to-end tests.
## Self-Review Checklist:
- [ ] I've reviewed my own diff for quality, security, and reliability
- [x] Unsafe blocks (if any) have justifying comments ; no unsafe block
added
- [x] The content adheres to Zed's UI standards ; no visual change
- [x] Tests cover the new/changed behavior
- [x] Performance impact has been considered and is acceptable
Release Notes:
- N/A
---------
Co-authored-by: Friday <260232009+buildfriday@users.noreply.github.com>
Label | Description
-- | --
area:integrations/git/panel | Feedback for the Git panel UI and
behavior.
area:gpui/graphics/wgpu | Feedback for GPUI's WGPU graphics backend,
including initialization and rendering.
area:breadcrumbs | Feedback for editor path and symbol breadcrumbs.
area:title bar | Issues suitable for title bar.
area:ai/agent thread/checkpoints | Feedback for Agent checkpoint
creation, comparison, restoration, and related Git behavior.
area:ai/agent thread/tools | Feedback for built-in Agent tools,
including execution, permissions, inputs, and outputs.
Release Notes:
- N/A
While building Zed with nightly rustc I've noticed it doesn't compile
because of good old pathfinder_simd. It also emits a bunch of warnings
about use of f64 literals where f32 is expected, so I've fixed them - it
should make future upgrades more straightforward.
# Objective
Prevent the thread from automatically scrolling to the end when
expanding a "Context Compacted" message at the end of the thread, as we
suspect the user would likely be trying to read it from top to bottom.
## Solution
Update how the `ThreadView` behaves when expanding a "Context Compacted"
message, namely by updating `ThreadView::toggle_compaction_expansion` in
order to anchor the scroll position, which disables the tail following
behavior, ensuring that the scroll offset is kept.
In order to better support this, a new method is introduced to
`gpui::elements::list::ListState`, `anchor_scroll_position`, which takes
the current scroll offset and scrolls to it which, in turns, disables
the following behavior.
Another option would be to simply introduce
`gpui::elements::list::ListState::stop_following` but I'm not sure
whether we'd want to expose that implementation detail, even though we
already have `gpui::elements::list::ListState::is_following_tail` .
## Testing
Tested both manually as well as introduced a new test for
`ListState::anchor_scroll_position`.
## Self-Review Checklist:
- [x] I've reviewed my own diff for quality, security, and reliability
- [ ] Unsafe blocks (if any) have justifying comments
- [ ] The content adheres to Zed's UI standards
([UX/UI](https://github.com/zed-industries/zed/blob/main/CONTRIBUTING.md#uiux-checklist)
and
[icon](https://github.com/zed-industries/zed/blob/main/crates/icons/README.md)
guidelines)
- [x] Tests cover the new/changed behavior
- [x] Performance impact has been considered and is acceptable
## Showcase
<details>
<summary>Before</summary>
https://github.com/user-attachments/assets/cf8ee10f-9c74-429d-afed-7a39ccd71c67
</details>
<details>
<summary>After</summary>
https://github.com/user-attachments/assets/78edfca9-099a-4fdb-9c4e-75f790096957
</details>
---
Release Notes:
- Improved expanding "Context Compacted" messages in the Agent Panel to
ensure that the scroll position is preserved, instead of automatically
scrolling to the end of the thread.
---------
Co-authored-by: Joseph T. Lyons <JosephTLyons@gmail.com>
Auto-hiding scrollbars currently reveal whenever their prepaint geometry
changes. That matches browser behavior, but it causes continuously
updating views to restart the visibility timer on every content change
even when the user has not scrolled.
Add a per-instance `ScrollbarRevealPolicy`. The default
`ScrollOrContentChange` policy preserves the existing behavior for all
current callers. Views with continuously growing content can opt into
`ScrollOnly`, which reveals only when the offset changes independently
of content size, including when a view remains anchored to the bottom.
Tests pin the default behavior, the opt-in behavior, ordinary scrolling,
stationary content growth, bottom-anchored growth, and scrolling while
content is growing.
Release Notes:
- N/A
# Objective
Fixes FR-149
- Make word movement and selection predictable around punctuation.
- Prior art: [#58882](https://github.com/zed-industries/zed/pull/58882),
[#61916](https://github.com/zed-industries/zed/pull/61916), and their
revert [#62213](https://github.com/zed-industries/zed/pull/62213).
## Solution
- Treat words and punctuation runs as separate movement units in both
directions.
- This adds a stop between leading punctuation and its following word,
such as `.▏foo`.
## Testing
- Added table-driven coverage for reported examples, delimiters,
operators, punctuation runs, and Unicode punctuation.
- Tested editor movement, selection, and deletion behavior.
## Self-Review Checklist:
- [x] I've reviewed my own diff for quality, security, and reliability
- [x] Unsafe blocks (if any) have justifying comments
- [x] The content adheres to Zed's UI standards
([UX/UI](https://github.com/zed-industries/zed/blob/main/CONTRIBUTING.md#uiux-checklist)
and
[icon](https://github.com/zed-industries/zed/blob/main/crates/icons/README.md)
guidelines)
- [x] Tests cover the new/changed behavior
- [x] Performance impact has been considered and is acceptable
---
Release Notes:
- Fixed word movement and selection stopping on the wrong side of
punctuation.
In the current Python Tree-sitter `highlights.scm`, dunder variables
(e.g., `__name__`, `__file__`) are not highlighted.
This PR adds a regex `#match?` so all dunder variables are highlighted
under `@attribute.special`.
An alternative solution would be to use `#any-of?` and list all dunder
variables.
| Before | After |
|
:---------------------------------------------------------------------------------------------------------------------------------:
|
:--------------------------------------------------------------------------------------------------------------------------------:
|
|<img width="623" height="895" alt="image"
src="https://github.com/user-attachments/assets/ad7b67ba-541c-40ac-a38a-cc863e417fcc"
/> | <img width="612" height="903" alt="image"
src="https://github.com/user-attachments/assets/a91f5efb-230a-4214-b16b-04278b3ef9ee"
/>|
Self-Review Checklist:
- [x] I've reviewed my own diff for quality, security, and reliability
- [x] Unsafe blocks (if any) have justifying comments
- [x] The content is consistent with the [UI/UX
checklist](https://github.com/zed-industries/zed/blob/main/CONTRIBUTING.md#uiux-checklist)
- [ ] Tests cover the new/changed behavior
- [x] Performance impact has been considered and is acceptable
Release Notes:
- Added Python dunder variables highlighting via `attribute.special` in
themes.
## Objective
Add functional horizontal scrollbars to wide Markdown tables, matching
the existing code-block scrollbar pattern.
Fixes#61437
PR was talked about in https://github.com/zed-industries/zed/pull/61698
## Implementation:
- Added a `BTreeMap<usize, ScrollHandle>` to `Markdown` for table scroll
handles, keyed by source-range start.
- During `MarkdownElement` layout, each table container is connected to
a `Scrollbars` widget configured for the horizontal axis, using a stable
ID derived from `("markdown-table-scrollbar", range.start)`.
- The inner table div uses `overflow_x_scroll()` + `track_scroll()` with
`restrict_scroll_to_axis`.
- Handles for removed tables are discarded after re-rendering.
- No public API, setting, schema, or migration changes required.
## Recording
https://github.com/user-attachments/assets/cd0f635b-a485-4de8-9059-40a05f256fb1
## Release Notes:
- Improved navigation of wide Markdown tables with horizontal
scrollbars.
---------
Co-authored-by: MrSubidubi <finn@zed.dev>
# Objective
Zed's `wgpu` backend currently allocates memory for fonts, even if they
are statically available in memory. Zed embeds 8 fonts, which accounts
for about 1.6 megabytes of memory. This change should reduce memory
usage by that amount on the `wgpu` backend. (Linux and Web)
For more context, this is the function signature for `load_font_data`:
```rust
/// Loads a font data into the `Database`.
///
/// Will load all font faces in case of a font collection.
pub fn load_font_data(&mut self, data: Vec<u8>) {
self.load_font_source(Source::Binary(alloc::sync::Arc::new(data)));
}
```
## Solution
Instead of calling `fontdb::Database::load_font_data`, which needs an
owned `Vec`, use `load_font_source`, which allows us to create an `Arc`
ourselves, which doesn't require ownership of the underlying data.
## Testing
All tests pass
## Self-Review Checklist:
- [x] I've reviewed my own diff for quality, security, and reliability
- [x] Unsafe blocks (if any) have justifying comments
- [x] The content adheres to Zed's UI standards
([UX/UI](https://github.com/zed-industries/zed/blob/main/CONTRIBUTING.md#uiux-checklist)
and
[icon](https://github.com/zed-industries/zed/blob/main/crates/icons/README.md)
guidelines)
- [x] Tests cover the new/changed behavior
- [x] Performance impact has been considered and is acceptable
Release Notes:
- Slightly improved memory usage on Linux
# Objective
This PR aims to fix these related deficiencies with the terminal tool's
working directory path resolution:
- Fix#60014
- Fix#60040
- Fix#60043
## Solution
- Use `project.path_style(cx).is_absolute(dir)` instead of
`Path::is_absolute`. The latter follows the path semantics of the host
which fails to recognize `/home/...` etc as an absolute path on Windows,
we want to follow the path semantics of the *project*.
- Splits out path resolution into a separate pure function
`resolve_cd_in_worktrees` to enable comprehensive cross-PathStyle
testing
- The function unifies path prefix checking across both absolute and
relative modes, using the `path_style.strip_prefix` (which uses the
RelPath util internally) which fails when subdirs try to escape with
`..`.
- Add comprehensive tests for absolute/relative modes, path styles, and
potential `..` escapes
- Enables targeting subdirectories with both absolute and relative path
modes (separate commit)
- Also makes both path modes use `project.worktrees`. Before, the
relative path mode used `project.visible_worktrees`.
## Testing
Here's how I tested these changes:
- Added a new test: `test_resolve_cd_uses_project_path_style` with a
comprehensive collection of assertions for `..` edge cases
- Run `cargo test -p agent` and `./script/clippy -p agent` on:
- Linux
- Windows in Ubuntu WSL
- Windows native
- Build and run Zed on:
- Linux
- Windows
- Verify that the three issues are fixed:
- `cd` param path traversal escapes are blocked
- Linux local project
- Windows local project
- Windows remote WSL project
- Allows absolute paths when running on Windows in a WSL project
- `cd` param may target subdirectories in both absolute and relative
path modes
- Linux local project
- Windows local project
- Windows WSL project
- Check that different models find the updated tool descriptions to be
understandable and usable the first time. Used the prompt below to get
the following models to test path resolution on each of: Windows local,
Windows WSL, Linux. All models except GPT-5-nano one-shot the below
prompt.
- Claude Sonnet 4.6
- Claude Opus 4.8
- Claude Haiku 4.5
- GPT-5.5 pro
- GPT-5-nano - got there eventually but is confused by the design of the
cd param, it expects that if you set cd to `my-project` that it will
target the directory `my-project/my-project`.
- Gemini 3.1 Pro
- Gemini 3.5 Flash
- Grok 4.3
- Grok 0.1 build
<details>
<summary>Agent test prompt</summary>
<blockquote>
List the current directory contents and pick one subdir.
Then use that information and the project path details you were provided
to execute `pwd` with the terminal tool's cd parameter set to the
following values:
- current project absolute path
- current project name
- current project absolute path + an existing subdir
- current project name + an existing subdir
- absolute path to user's .ssh dir
- current project absolute path + whatever necessary path traversal .. +
.ssh segments to target the user's .ssh dir
Then tell me what happened.
Then stop.
</blockquote>
</details>
## Self-Review Checklist:
- [x] I've reviewed my own diff for quality, security, and reliability
- [x] Unsafe blocks (if any) have justifying comments
- [x] The content adheres to Zed's UI standards
([UX/UI](https://github.com/zed-industries/zed/blob/main/CONTRIBUTING.md#uiux-checklist)
and
[icon](https://github.com/zed-industries/zed/blob/main/crates/icons/README.md)
guidelines)
- [x] Tests cover the new/changed behavior
- [x] Performance impact has been considered and is acceptable
---
Release Notes:
- Fixed terminal tool targeting absolute directories on Windows host
connected to a remote SSH / WSL project
- Fixed terminal tool not blocking path traversal escapes
- Improved terminal tool targeting project subdirectories
---------
Co-authored-by: Richard Feldman <richard@zed.dev>
Co-authored-by: Richard Feldman <oss@rtfeldman.com>
# Objective
Fix several Git panel focus issues:
- Directional navigation could not move focus into the History tab.
- History focus state was not updated when the panel received focus.
- Switching back to Changes could leave focus on the panel instead of
the commit editor.
Fixes#62211
## Solution
- Use the Git panel root as the activation focus target when History is
active.
- Reuse the panel's focus handle for focus events.
- Use the appropriate activation target when switching tabs.
## Testing
- Add a regression test
`test_history_tab_pane_navigation_focuses_rendered_panel` for navigating
between the editor and History.
## Self-Review Checklist:
- [x] I've reviewed my own diff for quality, security, and reliability
- [x] Unsafe blocks (if any) have justifying comments
- [x] The content adheres to Zed's UI standards
([UX/UI](https://github.com/zed-industries/zed/blob/main/CONTRIBUTING.md#uiux-checklist)
and
[icon](https://github.com/zed-industries/zed/blob/main/crates/icons/README.md)
guidelines)
- [x] Tests cover the new/changed behavior
- [x] Performance impact has been considered and is acceptable
## Showcase
after fix:
[2026-08-05
22-45-38.webm](https://github.com/user-attachments/assets/fdb3f87a-f1af-4ae1-95f7-0d61acfb3095)
---
Release Notes:
- Fixed directional focus navigation into the Git panel's History tab.
The renderer benchmarks drain the main queue to quiescence before each
measured frame, so a task that yields and re-queues itself (an idle
sweep, a poller) runs to completion inside a single measured frame — or
keeps the benchmark spinning forever. Real platforms don't work that
way: macOS's main-queue drain only runs the items present when the drain
starts, Linux runs one runnable per calloop idle callback, and Windows
drains under a 10ms budget before servicing paint messages — frames
always get a chance to preempt. So `run_ready_main_tasks` now only runs
the tasks that were queued when servicing began, letting self-re-queuing
work advance one batch per frame like it would under a real run loop,
and `BenchAppContext` grows a `run_until` (wrapping the dispatcher
primitive task benchmarks already use) so setup code can pump the app
until a condition holds without settling to idle first.
Release Notes:
- N/A
# Objective
Document how to use Poolside in Zed through the ACP Registry, Poolside
Agent CLI, manual Custom Agent configuration, and Terminal Threads.
## Solution
- Add Poolside paths to the AI by Company guide.
- Document ACP Registry installation and in-thread login.
- Document CLI-assisted and manual Custom Agent configuration.
- Explain the settings-path and PATH requirements.
- Link to the public Poolside Agent CLI repository and Poolside’s Zed
documentation.
## Testing
- Ran `./script/prettier`.
- Ran `git diff --check`.
- Verified Poolside’s live ACP Registry entry.
- Verified the commands, configuration, and authentication behavior
against the Poolside and Zed implementations.
## Self-Review Checklist:
- [x] I've reviewed my own diff for quality, security, and reliability
- [x] Unsafe blocks are not applicable
- [x] The content follows Zed’s documentation and UI conventions
- [x] Code tests are not applicable to this documentation-only change
- [x] Performance impact is not applicable
---
Release Notes:
- N/A
# Objective
Prevent auto-indent from using stale syntax when text changes during a
background parse.
## Solution
Keep auto-indent pending and parsing active until the current parse
completes.
## Self-Review Checklist:
- [x] I've reviewed my own diff for quality, security, and reliability
- [x] Unsafe blocks (if any) have justifying comments
- [x] The content adheres to Zed's UI standards
([UX/UI](https://github.com/zed-industries/zed/blob/main/CONTRIBUTING.md#uiux-checklist)
and
[icon](https://github.com/zed-industries/zed/blob/main/crates/icons/README.md)
guidelines)
- [x] Tests cover the new/changed behavior
- [x] Performance impact has been considered and is acceptable
---
Release Notes:
- Editor: Fixed auto-indent occasionally using stale syntax after rapid
edits.
Enables zooming of mermaid diagrams by supporting horizontal scroll
---
Release Notes:
- improved: Mermaid diagrams can now be zoomed and horizontally
scrolled, in both the markdown preview and the agent panel
Resolves#61888.
Multi-modifier gestures could leave GPUI’s standalone-modifier state
armed after one modifier was released. For example, releasing `Alt`
before `Shift` during an `Alt+Shift` gesture caused the final `Shift`
release to be synthesized as a standalone shift keystroke.
This diff invalidates standalone-modifier synthesis whenever multiple
modifiers are active, while preserving genuine modifier-only key
bindings.
Release Notes:
- Fixed Alt+Shift gestures potentially triggering standalone Shift key
bindings.
`ThreadedDispatcher::run_until` - the completion mechanism under
`BenchAppContext` async task benchmarks (`bench_task` /
`bench_batched_task`, added in #62180) - drained the entire main queue
before checking its readiness predicate. Main-thread work that re-queues
itself (an idle-time sweep, a polling loop) keeps the queue non-empty
until it finishes every iteration, so a task benchmark's measured
interval silently extended past the awaited task's completion until all
such deferred work settled. Benchmarks of UI workloads that schedule
follow-up idle work could report several times their true completion
latency.
This changes `run_until` to step main-thread runnables one at a time and
check readiness before each, so it returns at the completion it awaits
rather than at queue quiescence, making async task benchmarks more
accurate. The regression test spawns a task that re-queues itself 10,000
times plus a one-shot completion task, and asserts `run_until` returns
without draining the re-queued work; it fails against the previous
drain-first implementation.
Release Notes:
- N/A
Removes an unnecessary dependency on settings as
`settings::OpenAiReasoningEffort` just points to
`language_model::ReasoningEffort`
Release Notes:
- N/A
# Objective
Closes#41744
When a multi-key binding switches from Insert mode to Helix normal mode,
subsequent editing actions can place the cursor incorrectly.
While a printable multi-key binding is pending, Zed temporarily inserts
the pending keys into the buffer. Once the binding is matched, Zed
dispatches the associated action and then deletes the pending text.
Because the buffer uses a CRDT, the deleted text remains as tombstoned
fragments. The cursor can still resolve to the correct visible offset
while its selection anchor remains associated with one of those
fragments, affecting the ordering of subsequent edits.
Among the explicit Vim mode-switch actions, `SwitchToHelixNormalMode` is
the only one that preserves the existing selections:
ce6f3af5f7/crates/vim/src/vim.rs (L713-L719)
As a result, the tombstone-associated anchor is carried into Helix
normal mode, where later actions such as `o` can place the cursor on the
wrong side of the inserted newline.
The affected path is narrow: a multi-key binding must invoke
`SwitchToHelixNormalMode` from Insert mode. This creates pending text
and then preserves the resulting selection anchors when entering Helix
normal mode.
There are two possible layers at which to address this. At the Editor
layer, selections could be refreshed whenever pending text is removed so
that they no longer reference deleted fragments. At the Vim layer, the
Helix mode transition can explicitly guard against preserving those
anchors.
PR #50918 attempted the broader Editor-layer solution by refreshing
selections after resolving a multi-key binding. However, as discussed in
this review comment:
https://github.com/zed-industries/zed/pull/50918#issuecomment-4411319469,
the more appropriate scope for this reported issue is the Vim
mode-switching layer.
## Solution
Based on PR #50918 and its review feedback, this PR limits the fix to
`SwitchToHelixNormalMode`.
When that action is triggered by pending Insert-mode keystrokes, Vim
refreshes the preserved selection anchors after the pending text is
removed. This keeps the fix scoped to the affected Helix transition
without changing other multi-key bindings.
## Testing
Added a GPUI regression test.
## Self-Review Checklist:
- [x] I've reviewed my own diff for quality, security, and reliability
- [x] Unsafe blocks (if any) have justifying comments
- [x] The content adheres to Zed's UI standards
([UX/UI](https://github.com/zed-industries/zed/blob/main/CONTRIBUTING.md#uiux-checklist)
and
[icon](https://github.com/zed-industries/zed/blob/main/crates/icons/README.md)
guidelines)
- [x] Tests cover the new/changed behavior
- [x] Performance impact has been considered and is acceptable
---
Release Notes:
- Fixed incorrect cursor placement after using a multi-key binding to
leave insert mode in Vim or Helix mode.
---------
Co-authored-by: dino <dinojoaocosta@gmail.com>
# Objective
Fixes#61767.
Double-clicking a service name after a `systemctl` tree branch also
selects the `└─` prefix.
## Solution
Add `─` to Alacritty's semantic escape characters for both PTY and
display-only terminals. The regression test checks that selecting
`└─zms-demo.target` returns only `zms-demo.target`.
## Testing
- `cargo fmt --all -- --check`
- `git diff --check`
- Added `semantic_selection_stops_at_tree_branch`
The regression test could not run locally because this Windows
environment does not have MSVC's `link.exe`. It can be run with:
`cargo test -p terminal semantic_selection_stops_at_tree_branch`
## Self-Review Checklist:
- [x] I've reviewed my own diff for quality, security, and reliability
- [x] Unsafe blocks (if any) have justifying comments
- [x] The content adheres to Zed's UI standards
- [x] Tests cover the new/changed behavior
- [x] Performance impact has been considered and is acceptable
---
Release Notes:
- Fixed terminal word selection around tree branch glyphs.
# Objective
This is a follow-up to my own PR #59270 which fixed the hang it targeted
(#55042) only partially and broke `ShellBuilder::build` for fish.
**The hang is only half fixed.** The agent's terminal tool wraps
commands as `sh -i -c 'exec </dev/null; <command>'`. The redirect sits
on the same line as the command, but POSIX shells parse an entire line
before executing any of it. So when `<command>` has a syntax error, the
whole line is discarded — including the `exec` that was supposed to
close stdin. Interactive `dash` then recovers from the error, finds
stdin still attached to the PTY, and sits at a `$` prompt until the user
presses Ctrl+D.
That means #59270 only helped when the redirect had already run, i.e.
multi-line commands whose error is on a later line. A single-line
command still hangs, including `cat <(echo hi)`, the example from that
PR's own description. Probably prompts or tool definitions changed since
that PR landed which now make it more likely again to run into this
issue, at least with Anthropic models.
**fish runs no command at all.** #59270 also applied the `exec
</dev/null; …` form to fish, replacing its `begin; …; end </dev/null`
wrapper. But fish's `exec` requires a command to replace the shell with,
so a bare `exec </dev/null` is a usage error: fish prints `exec`'s help
text, skips the rest of the command, and exits 0.
## Solution
For POSIX shells, put the redirect on its own line:
```sh
sh -i -c 'exec </dev/null
<command>'
```
That makes it a separate parse-and-execute unit, so it runs before the
command line is parsed. Any syntax error is then reported by a shell
that already has `/dev/null` on stdin, and it exits instead of
prompting.
For fish, go back to `begin; <command>; end </dev/null`. fish exits on a
parse error even when interactive, so it never needed the `exec` form to
begin with.
Only the `Posix` and `Fish` arms change; the other shells' redirects are
untouched.
Note this is still a targeted fix for `dash`. The underlying reason a
syntax error becomes an interactive prompt is that agent terminals are
built with `-i` at all — non-interactive `dash` exits cleanly even with
no redirect. Dropping `-i` would remove the whole class of hang, but it
would also stop sourcing users' interactive shell config in agent
terminals.
## Testing
`cargo test -p util --lib shell_builder` covers the changed command
assembly.
Reviewers can check the shell behavior directly on a PTY, without
building Zed:
```sh
# hangs at a `$` prompt until Ctrl+D
script -qec "sh -i -c 'exec </dev/null; echo hi ;;'" /dev/null
# exits immediately with just the syntax error
script -qec "sh -i -c 'exec </dev/null
echo hi ;;'" /dev/null
```
```sh
fish -i -c 'exec </dev/null; echo test' # prints exec's help, no "test", exit 0
fish -i -c 'begin; echo test; end </dev/null' # prints "test"
fish -i -c 'if true' # exit 127, no hang
```
macOS is unaffected in practice because `/bin/sh` there is bash, which
exits on a syntax error even with `-i`; Windows goes through the
`Cmd`/`PowerShell` arms, which this doesn't touch. A `zsh`-as-`/bin/sh`
setup is likewise unaffected.
## Self-Review Checklist:
- [x] I've reviewed my own diff for quality, security, and reliability
- [x] Unsafe blocks (if any) have justifying comments
- [x] The content adheres to Zed's UI standards
([UX/UI](https://github.com/zed-industries/zed/blob/main/CONTRIBUTING.md#uiux-checklist)
and
[icon](https://github.com/zed-industries/zed/blob/main/crates/icons/README.md)
guidelines)
- [x] Tests cover the new/changed behavior
- [x] Performance impact has been considered and is acceptable
---
Release Notes:
- Fixed agent terminal commands hanging at a shell prompt after a syntax
error
# Objective
Fixes#62084
Fix the Wayland IME candidate window flickering between two positions
when an inactive Zed window has a terminal producing continuous output.
Terminal wakeup events in the inactive window can trigger IME position
updates, causing its cursor bounds to alternate with those from the
active window.
90d024b88a/crates/terminal_view/src/terminal_view.rs (L1130-L1136)90d024b88a/crates/gpui/src/window.rs (L5617-L5627)
## Solution
Ignore IME position updates from inactive Wayland windows.
This ensures that only the active window can update the cursor rectangle
of the shared Wayland text-input object.
## Testing
To test manually on Linux with Wayland and Fcitx5:
1. Open two Zed windows.
2. Start continuous terminal output in the first window.
3. Use the Chinese IME in an editor in the second window.
4. Verify that the IME candidate window remains at a stable position.
## Self-Review Checklist:
- [X] I've reviewed my own diff for quality, security, and reliability
- [X] Unsafe blocks (if any) have justifying comments
- [X] The content adheres to Zed's UI standards
([UX/UI](https://github.com/zed-industries/zed/blob/main/CONTRIBUTING.md#uxui-checklist)
and
[icon](https://github.com/zed-industries/zed/blob/main/crates/icons/README.md)
guidelines)
- [ ] Tests cover the new/changed behavior
- [X] Performance impact has been considered and is acceptable
## Showcase
after fix:
[2026-08-02
22-43-35.webm](https://github.com/user-attachments/assets/ff2e3bc7-7772-481e-82f8-92e6e3fa7275)
---
Release Notes:
- Fixed the IME candidate window flickering when another Zed window has
continuous terminal output
# Objective
Follow-up to #62028. When archiving removed a thread's workspace, the
replacement
was derived from whichever sidebar row sat adjacent to the archived one.
Rows are
a flat list with project headers as separators, so archiving the oldest
thread in
a project selected the next project's newest thread and moved the user
out of the
project they were working in — and when that neighbor was remote,
connected to
its host just to pick a fallback.
# Solution
- `MultiWorkspace::remove` now chooses the replacement itself from a
`RemovalIntent` (keep the project vs close it): a live workspace in the
same
project, the project's own roots when kept, the nearest retained
neighbor,
an adjacent local project, then an empty workspace. The workspaces being
removed are excluded from that search in one place instead of five.
- Sidebar neighbor selection stays within the entry's project section.
- The three archive paths now share one removal orchestration, and the
three
"open the closed workspace first" helpers collapse into one.
# Testing
New tests cover: the keep-project fallback, neighbor selection staying
inside the
project section, and archiving with a mock-remote neighboring project
never
connecting to it. Each was verified to fail against the previous
behavior.
## Self-Review Checklist:
- [x] I've reviewed my own diff for quality, security, and reliability
- [x] Unsafe blocks (if any) have justifying comments
- [x] The content adheres to Zed's UI standards
([UX/UI](https://github.com/zed-industries/zed/blob/main/CONTRIBUTING.md#uiux-checklist)
and
[icon](https://github.com/zed-industries/zed/blob/main/crates/icons/README.md)
guidelines)
- [x] Tests cover the new/changed behavior
- [x] Performance impact has been considered and is acceptable
---
Release Notes:
- Fixed archiving a thread switching the window to a different project.
- Fixed a possible crash when archiving a thread while its window was
closing.
### Summary
This PR improves GPUI's Criterion benchmark integration by adding
task-based measurement APIs to `BenchAppContext`. Benchmarks can now
await asynchronous GPUI operations whose completion depends on
foreground work, background work, timers, or external wakes, while
keeping setup and destruction outside the measured interval.
#### Added `BenchAppContext` APIs
- Add `bench_task` to measure a GPUI `Task` through completion in
Criterion's iteration loop.
- Add `bench_batched_task` to perform per-iteration setup outside the
timed interval before measuring a task.
- Preserve support for non-`Send` foreground task outputs and destroy
outputs after timing stops.
- Collect frame timings produced during asynchronous benchmark
iterations without including collection in the measurement.
- Drive task completion across temporary dispatcher quiescence and
externally delivered wakes.
Release Notes:
- N/A
