This PR improves the `edit prediction: Capture Example` in several ways:
* fixed bugs in how the uncommitted diff was calculated
* added a `edit_predictions.examples_dir` setting that can be set in
order to have the action automatically save examples into the given
folder
* moved the action into the `edit_predictions` crate, in preparation for
collecting this data passively from end users, when they have opted in
to data sharing, similar to what we did for Zeta 1
Release Notes:
- N/A
This PR adds the ability to favorite models for external agents—writing
to the settings in the `agent_servers` key—as well as a handful of other
improvements:
- Make the cycling keybinding `alt-enter` work for the inline assistant
as well as previous user messages
- Better organized the keybinding files removing some outdated
agent-related keybinding definitions
- Renamed the inline assistant key context to "InlineAssistant" as
"PromptEditor" is old and confusing
- Made the keybindings to rate an inline assistant response visible in
the thumbs up/down button's tooltip
- Created a unified component for the model selector tooltip given we
had 3 different places creating the same element
- Make the "Cycle Favorited Models" row in the tooltip visible only if
there is more than one favorite models
Release Notes:
- agent: External agents also now support the favoriting model feature,
which comes with a handy keybinding to cycle through the favorite list.
Closes#44720
Release Notes:
- Fixed dock panel button tooltips not being dismissed when toggling
panels via keyboard shortcut
**Problem:** When hovering over a dock panel button and using a keyboard
shortcut to toggle the panel, the tooltip remains visible with stale
content. This is inconsistent with mouse click behavior, where the
tooltip is dismissed on mouse down.
**Solution:** Include the panel's active state in the button's element
ID. When the state changes, the element ID changes (e.g., `"DebugPanel"`
→ `"DebugPanel-active"`), which causes GPUI to discard the old element
state including the cached tooltip.
**Testing:** Manually verified:
1. Hover over a dock panel button, wait for tooltip
2. Press keyboard shortcut to toggle the panel
3. Tooltip is now dismissed (consistent with mouse click behavior)
https://github.com/user-attachments/assets/ed92fb6c-6c22-44e2-87e3-5461d35f7106
---------
Co-authored-by: MrSubidubi <finn@zed.dev>
Draft as a base for continuing the discussion in #8008 : adds a
`SplitOperation` enum to support bindings like `["pane::SplitLeft",
{"operation": "Clear"}]`
To be discussed @MrSubidubi and others:
- Naming: Generally not happy with names yet and specifically `Empty` is
unclear, e.g., what does this mean for terminal panes? Added placeholder
code to split without cloning, but unsure what users would expect in
this case.
- ~~I removed `SplitAndMoveXyz` actions but I guess we should keep them
for backwards compatibility?~~
- May have missed details in the move implementation. Will check the
code again for opportunities to refactor more code after we agree on the
approach.
- ~~Tests should go to `crates/collab/src/tests/integration_tests.rs`?~~
Closes#8008
Release Notes:
- Add `pane::Split` mode (`{ClonePane,EmptyPane,MovePane}`) to allow
creating an empty buffer.
---------
Co-authored-by: Finn Evers <finn.evers@outlook.de>
Co-authored-by: MrSubidubi <finn@zed.dev>
* Abs path trust should transitively trust all single file worktrees on
the same host
* Init worktree trust on the client side even when devcontainers are
run: remote host unconditionally checks trust, hence the client has to
keep track of it and respond with approves/declines.
Do trust all devcontainers' remote worktrees, as containers are isolated
and "safe".
Release Notes:
- N/A
The function returns the previous focus handle, which may be null if
there is no previous focus. Unfortunately that also overlaps with the
error return value, so winapi will hand us a error 0 back in those cases
which we log ...
Release Notes:
- N/A *or* Added/Fixed/Improved ...
Closes#18287
Release Notes:
- Added autocomplete for lsp initialization_options
## Description
This MR adds the following code-changes:
- `initialization_options_schema` to the `LspAdapter` to get JSON
Schema's from the language server
- Adds a post-processing step to inject schema request paths into the
settings schema in `SettingsStore::json_schema`
- Adds an implementation for fetching the schema for rust-analyzer which
fetches it from the binary it is provided with
- Similarly for ruff
<img width="857" height="836" alt="image"
src="https://github.com/user-attachments/assets/3cc10883-364f-4f04-b3b9-3c3881f64252"
/>
## Open Questions(Would be nice to get some advice here)
- Binary Fetching:
- I'm pretty sure the binary fetching is suboptimal. The main problem
here was getting access to the delegate but i figured that out
eventually in a way that i _hope_ should be fine.
- The toolchain and binary options can differ from what the user has
configured potentially leading to mismatches in the autocomplete values
returned(these are probably rarely changed though). I could not really
find a way to fetch these in this context so the provided ones are for
now just `default` values.
- For the trait API it is just provided a binary, since i wanted to use
the potentially cached binary from the CachedLspAdapter. Is that fine
our should the arguments be passed to the LspAdapter such that it can
potentially download the LSP?
- As for those LSPs with JSON schema files in their repositories i can
add the files to zed manually e.g. in
languages/language/initialization_options_schema.json, which could cause
mismatches with the actual binary. Is there a preferred approach for Zed
here also with regards to updating them?
Continuing of #44334
I removed disabling of vsync which was causing jitter on some external
displays
cc: @maxbrunsfeld @Anthony-Eid
Release Notes:
- Mark metal layers opaque for non-transparent windows to allow
direct-to-display when supported
Signed-off-by: Marco Mihai Condrache <52580954+marcocondrache@users.noreply.github.com>
Release Notes:
- Fixed spurious "no checkpoint" error in agent panel
---
## Summary
`update_last_checkpoint` would call `last_user_message()` twice - once
at the start to capture the checkpoint, and again in an async closure
after the checkpoint comparison completed. If a new user message without
a checkpoint was added between these two calls, the second call would
find the new message and fail with "no checkpoint".
## Fix
Capture the user message ID at the start and use `user_message_mut(&id)`
in the async closure to find the specific message.
cc @mikayla-maki
This PR fixes two issues with regards to markdown codeblocks rendered in
tool call input and output content display:
- the JSON code snippets weren't properly indented
- codeblocks weren't being rendered in unique containers; e.g., if you
hovered one scrollbar, all of them would also be hovered, even though
horizontal scrolling itself worked properly
Here's the end result:
https://github.com/user-attachments/assets/3d6daf64-0f88-4a16-a5a0-94998c1ba7e2
Release Notes:
- agent: Fix scrollbar and JSON indentation for tool call input/output
content's markdown codeblocks.
- Make the buttons capable of changing the editor's content
(incrementing or decrementing the value)
- Make arrow key up and down increment and decrement the editor value
- Tried to apply a bit of DRY here by creating some functions that can
be reused across the buttons and editor given they all essentially do
the same thing (change the value)
- Fixed an issue where the editor would not allow focus to move
elsewhere, making it impossible to open a dropdown, for example, if your
focus was on the number field's editor
Release Notes:
- N/A
Closes https://github.com/zed-industries/zed/issues/43208
This PR essentially unblocks the editable number field. The function
that shapes editor lines was hard-coding text alignment to the left,
meaning that whatever different alignment we'd pass through
`EditorStyles`would be ignored. To solve this, I just added a text align
and align width fields to the line paint function and updated all call
sites keeping the default configuration. Had to also add an
`alignment_offset()` helper to make sure the cursor positioning, the
selection background element, and the click-to-focus functionality were
kept in-sync with the non-left aligned editor.
Then... the big star of the show here is being able to add the `mode`
method to the number field, which uses `TextAlign::Center`, thus making
it work as we designed it to work.
https://github.com/user-attachments/assets/3539c976-d7bf-4d94-8188-a14328f94fbf
Next up, is turning the number filed to edit mode where applicable.
Release Notes:
- Fixed a bug where different text alignment configurations (i.e.,
center and right-aligned) wouldn't take effect in editors.
Currently on x11, gpui PopUp windows only rely on the "notification"
type in order to indicate that they should spawn as floating window.
Several window managers (leftwm in my case, but it also seems to be the
case for dwm and ratpoison) do not this property into account thus not
spawning them as float. On the other hand, using Floating instead of
PopUp do make those windows spawn as floating, as these window manager
do take into account the (older) "dialog" type.
The [freedekstop
documentation](https://specifications.freedesktop.org/wm/1.5/ar01s05.html#id-1.6.7)
does seem to suggest that these windows should also have the override
redirect property :
> This property is typically used on override-redirect windows.
Note that this also disables pretty much all interactions with the
window manager (such as moving the window, resizing etc...)
Release Notes:
- Fix popup windows not spawning floating sometime on x11
This fixes an X11 scrolling issue where Zed may jump by a large amount
due to the scroll valuator state not being reset when the window loses
focus. If you Alt-Tab away from Zed, scroll in another application, then
return, the first scroll event in Zed applies the entire accumulated
delta instead of a single step.
The missing FocusOut reset was originally identified in issue #34901.
Resetting scroll positions on FocusOut matches the behavior already
implemented in the XinputLeave handler and prevents this jump.
Closes#34901Closes#40538
Release Notes:
- Fixed an X11 issue where Alt-Tabbing to another application,
scrolling, and returning to Zed could cause the next scroll event to
jump by a large amount.
I've just enriched the existing tab_stop.rs example for GPUI with a
demonstration of tab_group. I don't think tab groups existed when the
original example was written.
(I didn't understand the behaviour for tab_group from the doccomments
and the example was missing, so I think this is a productive PR)
Release Notes:
- N/A
The failure would happen if the current version of the file was open as
an editor. This happened because the git blob and current version of the
buffer would have the same `ProjectPath`.
The fix was adding a new `DiskState::Historic` variant to represent
buffers that are past versions of a file (usually a snapshot from
version control). Historic buffers don't return a `ProjectPath` because
the file isn't real, thus there isn't and shouldn't be a `ProjectPath`
to it. (At least with the current way we represent a project path)
I also change the display name to use the local OS's path style instead
of being hardcoded to Posix, and cleaned up some code too.
Release Notes:
- N/A
---------
Co-authored-by: Cole Miller <cole@zed.dev>
Co-authored-by: cameron <cameron.studdstreet@gmail.com>
Co-authored-by: xipengjin <jinxp18@gmail.com>
Reverts zed-industries/zed#44334
From my testing, this PR introduced screen tearing, or some kind of
strange visual artifact, when scrolling at medium speed on a large
display.
Release notes:
- N/A
#40291 made floating windows always stay on top, which made the settings
ui window always on top of Zed. To maintain the old behavior, this PR
changes the setting window to be a normal window.
Release Notes:
- N/A
## Problem
When users paste or drag large images into the agent panel, the encoded
payload can exceed upstream provider limits (e.g., Anthropic's 5MB
per-image limit), causing API errors.
## Solution
Enforce a default 5MB limit on encoded PNG bytes in
`LanguageModelImage::from_image`:
1. Apply existing Anthropic dimension limits first (1568px max in either
dimension)
2. Iteratively downscale by ~15% per pass until the encoded PNG is under
5MB
3. Return `None` if the image can't be shrunk within 8 passes
(fail-safe)
The limit is enforced at the `LanguageModelImage` conversion layer,
which is the choke point for all image ingestion paths (agent panel
paste/drag, file mentions, text threads, etc.).
## Future Work
The 5MB limit is a conservative default. Provider-specific limits can be
introduced later by adding a `from_image_with_constraints` API.
## Testing
Added a regression test that:
1. Generates a noisy 4096x4096 PNG (guaranteed >5MB)
2. Converts it via `LanguageModelImage::from_image`
3. Asserts the result is ≤5MB and was actually downscaled
---
**Note:** This PR builds on #45312 (prompt store fail-open fix). Please
merge that first.
cc @rtfeldman
---------
Co-authored-by: Zed Zippy <234243425+zed-zippy[bot]@users.noreply.github.com>
In the case of large vertical_scroll_margin, we could scroll up such
that the assistant was out of view. Now, keep it no lower than the
center of the editor.
Closes#18058
Release Notes:
- N/A
This commit fixes an issue where saving UTF-16 files resulted in UTF-8
bytes due to `encoding_rs` default behavior. It also introduces a
heuristic to detect BOM-less UTF-16 and binary files.
Changes:
- Manually implement UTF-16LE/BE encoding during file save to avoid
implicit UTF-8 conversion.
- Add `analyze_byte_content` to guess UTF-16LE/BE or Binary based on
null byte distribution.
- Prevent loading binary files as text by returning an error when binary
content is detected.
Special thanks to @CrazyboyQCD for pointing out the `encoding_rs`
behavior and providing the fix, and to @ConradIrwin for the suggestion
on the detection heuristic.
Closes#14654
Release Notes:
- (nightly only) Fixed an issue where saving files with UTF-16 encoding
incorrectly wrote them as UTF-8. Also improved detection for binary
files and BOM-less UTF-16.
Closes#42586
This includes a rewrite of `calculate_relative_line_numbers()`. Now it's
linear-time with respect to the number of rows displayed, instead of
linear time with respect to the number of rows displayed _plus_ the
distance to the base row.
Release Notes:
- Improved performance when using relative line numbers in large files
- Fixed relative line numbers not appearing in sticky headers
Closes#17467
Release Notes:
- On macOS, buffer search now syncs with the system find pasteboard,
allowing <kbd>⌘E</kbd> and <kbd>⌘G</kbd> to work seamlessly across Zed
and other apps.
Closes #ISSUE
Release Notes:
- Fixed an issue where the `pane: close all items` action would give up
if you hit "Cancel" on the prompt for what to do with a dirty buffer
It's been a little that we've noticed some flickering and other weird
resizing behavior with text truncation in Zed:
https://github.com/user-attachments/assets/4d5691a3-cd3d-45e0-8b96-74a4e0e273d2https://github.com/user-attachments/assets/d1d0e587-7676-4da0-8818-f4e50f0e294e
Initially, we suspected this could be due to how we calculate the length
of a line to insert truncation, which is based first on the length of
each individual character, and then second goes through a pass
calculating the line length as a whole. This could cause mismatch and
culminate in our bug.
However, even though that felt like a reasonable suspicion, I realized
something rather simple at some point: the `truncate` and
`truncate_start` methods in the `Label` didn't use `whitespace_nowrap`.
If you take Tailwind as an example, their `truncate` utility class takes
`overflow: hidden; text-overflow: ellipsis; white-space: nowrap;`. This
pointed out to a potential bug with `whitespace_nowrap` where that was
blocking truncation entirely, even though that's technically part of
what's necessary to truncate as you don't want text that will be
truncated to wrap.
Ultimately, what was happening was that the text element was caching its
layout based on its `wrap_width` but not considering its
`truncate_width`. The truncate width is essentially the new definitive
width of the text based on the available space, which was never being
computed. So the fix here was to add `truncate_width.is_none()` to the
cache validation check, so that it only uses the cached text element
size _if the truncation width is untouched_. But if that changes, we
need to account for the new width. Then, in the Label component, we
added `min_w_0` to allow the label div to shrink below its original
size, and finally, we added `whitespace_nowrap()` as the cache check
fundamentally fixed that method's problem.
In a future PR, we can basically remove the `single_line()` label method
because: 1) whenever you want a single label, you most likely want it to
truncate, and 2) most instances of `truncate` are already followed by
`single_line` in Zed today, so we can cut that part.
Result is no flickering with truncated labels!
https://github.com/user-attachments/assets/ae17cbde-0de7-42ca-98a4-22fcb452016b
Release Notes:
- Fixed a bug in GPUI where truncated text would flicker as you resized
the container in which the text was in.
Co-authored-by: Lukas Wirth <me@lukaswirth.dev>
Closes#5089
Release notes:
- Markdown lists now continue automatically when you press Enter
(unordered, ordered, and task lists). This can be configured with
`extend_list_on_newline` (default: true).
- You can now indent list markers with Tab to quickly create nested
lists. This can be configured with `indent_list_on_tab` (default: true).
---------
Co-authored-by: Claude <noreply@anthropic.com>
Co-authored-by: Smit Barmase <heysmitbarmase@gmail.com>
From Zed's title bar, you can click on buttons to open three modal
pickers: remote projects, projects, and branches. All of these pickers
use the modal layer, which by default, renders them centered on the UI.
However, a UX issue we've been bothered by is that when you _click_ to
open them, they show up just way too far from where your mouse likely is
(nearby the trigger you just clicked). So, this PR introduces a
`ModalPlacement` enum to the modal layer, so that we can pick between
the "centered" and "anchored" options to render the picker. This way, we
can make the pickers use anchored positioning when triggered through a
mouse click and use the default centered positioning when triggered
through the keybinding.
One thing to note is that the anchored positioning here is not as
polished as regular popovers/dropdowns, because it simply uses the x and
y coordinates of the click to place the picker as opposed to using
GPUI's `Corner` enum, thus making them more connected to their triggers.
I chose to do it this way for now because it's a simpler and more
contained change, given it wouldn't require a tighter connection at the
code level between trigger and picker. But maybe we will want to do that
in the near future because we can bake in some other related behaviors
like automatically hiding the button trigger tooltip if the picker is
open and changing its text color to communicate which button triggered
the open picker.
https://github.com/user-attachments/assets/30d9c26a-24de-4702-8b7d-018b397f77e1
Release Notes:
- Improved the UX of title bar modal pickers (remote projects, projects,
and branches) by making them open closer to the trigger when triggering
them with the mouse.
This adds the following Vim commands:
- `:r[ead] [name]`
- `:{range}r[ead] [name]`
The most important parts of this feature are outlined
[here](https://vimhelp.org/insert.txt.html#%3Ar).
The only intentional difference between this and Vim is that Vim only
allows `:read` (no filename) for buffers with a file attached. I am
allowing it for all buffers because I think that could be useful.
Release Notes:
- vim: Added the [`:r[ead] [name]` Vim
command](https://vimhelp.org/insert.txt.html#:read)
---------
Co-authored-by: Ben Kunkle <ben@zed.dev>
Fixes#29073
This PR reduces unnecessary GPU usage by being more selective about when
we present frames to prevent display underclocking (VRR optimization).
## Problem
Previously, we would keep presenting frames for 1 second after *any*
input event, regardless of whether it triggered a re-render. This caused
unnecessary GPU work when the user was idle or during low-frequency
interactions.
## Solution
1. **Only track input that triggers re-renders**: We now only record
input timestamps when the input actually causes the window to become
dirty, rather than on every input event.
2. **Rate-based activation**: The VRR optimization now only activates
when input arrives at a high rate (≥ 60fps over the last 100ms). This
means casual mouse movements or occasional keystrokes won't trigger
continuous frame presentation.
3. **Sustained optimization**: Once high-rate input is detected (e.g.,
during scrolling or dragging), we sustain frame presentation for 1
second to prevent display underclocking, even if input briefly pauses.
## Implementation
Added `InputRateTracker` which:
- Tracks input timestamps in a 100ms sliding window
- Activates when the window contains ≥ 6 events (60fps × 0.1s)
- Extends a `sustain_until` timestamp by 1 second each time high rate is
detected
Release Notes:
- Reduced GPU usage when idle by only presenting frames during bursts of
high-frequency input.
This fixes a number of issues where zed depends on the order of polling which changed when switching scheduler. We have adjusted the algorithm so it matches the previous order while keeping the prioritization feature.
Release Notes:
- N/A
This PR is an additive change introducing the `truncate_start` method to
labels, which gives us the ability to add an ellipsis at the beginning
of the text as opposed to the regular `truncate`. This will be generally
used for truncating file paths, where the end is typically more relevant
than the beginning, but given it's a general method, there's the
possibility to be used anywhere else, too.
<img width="500" height="690" alt="Screenshot 2025-12-17 at 12 35@2x"
src="https://github.com/user-attachments/assets/f853f5a3-60b3-4380-a11c-bb47868a4470"
/>
Release Notes:
- N/A
---------
Co-authored-by: Lukas Wirth <lukas@zed.dev>
Closes#42268
We've migrated user selections when a given workspace has a single
worktree (as then we could determine what the target worktree is).
Release Notes:
- python: Fixed selected virtual environments not being
persisted/deserialized correctly within long-running Zed sessions (where
multiple different projects might've been opened). This is a breaking
change for users of multi-worktree projects - your selected toolchain
for those projects will be reset.
Co-authored-by: Dino <dino@zed.dev>
Joseph T. Lyons