This PR fixes some broken links that where found using
[lychee](https://github.com/lycheeverse/lychee/) as discussed today with
@JosephTLyons and @nathansobo at the RustNL hackathon. Using
[lychee-action](https://github.com/lycheeverse/lychee-action/) we can
scan for broken links daily to prevent issues in the future.
There are still 6 broken links that I didn't know how to fix myself.
See https://github.com/thomas-zahner/zed/actions/runs/15075808232 for
details.
## Missing images
```
Errors in ./docs/src/channels.md
[ERROR] file:///home/runner/work/zed/zed/docs/.gitbook/assets/channels-3.png | Cannot find file
[ERROR] file:///home/runner/work/zed/zed/docs/.gitbook/assets/channels-1.png | Cannot find file
[ERROR] file:///home/runner/work/zed/zed/docs/.gitbook/assets/channels-2.png | Cannot find file
```
These errors are showing up as missing images on
https://zed.dev/docs/channels
I tried to search the git history to see when or why they were deleted
but didn't find anything.
## ./crates/assistant_tools/src/edit_agent/evals/fixtures/zode/prompt.md
There are three errors in that file. I don't fully understand how these
issues were caused historically. Technically it would be possible to
ignore the files but of course if possible we should address the issues.
Release Notes:
- N/A
---------
Co-authored-by: Marshall Bowers <git@maxdeviant.com>
Co-authored-by: Ben Kunkle <ben@zed.dev>
Closes#31491
### Problem
Mistral API enforces strict conversation flow requirements that other
providers don't. Specifically, after a `tool` message, the next message
**must** be from the `assistant` role, not `user`. This causes the
error:
```
"Unexpected role 'user' after role 'tool'"
```
This can also occur in normal conversation flow where mistral doesn't
return the assistant message but that is something which can't be
reproduce reliably.
### Root Cause
When users interrupt an ongoing tool call sequence by sending a new
message, we insert a `user` message directly after a `tool` message,
violating Mistral's protocol.
**Expected Mistral flow:**
```
user → assistant (with tool_calls) → tool (results) → assistant (processes results) → user (next input)
```
**What we were doing:**
```
user → assistant (with tool_calls) → tool (results) → user (interruption) ❌
```
### Solution
Insert an empty `assistant` message between any `tool` → `user` sequence
in the Mistral provider's request construction. This satisfies Mistral's
API requirements without affecting other providers or requiring UX
changes.
### Testing
To reproduce the original error:
1. Start agent chat with `codestral-latest`
2. Send: "Describe this project using tool call only"
3. Once tool calls begin, send: "stop this"
4. Main branch: API error
5. This fix: Works correctly
Release Notes:
- Fixed Mistral tool calling in some cases
Ideally we would show multi-keystroke binding, but I'd say this improves
over the status quo.
A partial solution to #27334
Release Notes:
- Fixed spurious warning for lack of edit prediction on multi-keystroke
binding
Co-authored-by: Ben Kunkle <ben@zed.dev>
Now immediately initializes the zlog filter even when there isn't an env
config. Before this change the default filters were applied after
settings load - I was seeing some `zbus` logs on init.
Also defaults to allowing warnings and errors from the suppressed log
sources. If these turn out to be chatty (they don't seem to be so far),
can bring back more suppression.
Release Notes:
- N/A
Fixes two things:
* ~3 months ago [in PR
#26420](https://github.com/zed-industries/zed/pull/26420/files#diff-33b58aa2da03d791c2c4761af6012851b7400e348922d64babe5fd48ac2a8e60)
`bindings_for_action` was changed to return bindings even when they are
shadowed (when the keystrokes would actually do something else).
* For edit prediction keybindings there was some odd behavior where
bindings for `edit_prediction_conflict` were taking precedence over
bindings for `edit_prediction` even when the `edit_prediction_conflict`
predicate didn't match. The workaround for this was #24812. The way it
worked was:
- List all bindings for the action
- For each binding, get the highest precedence binding with the same
input sequence
- If the highest precedence binding has the same action, include this
binding. This was the bug - this meant that if a binding in the keymap
has the same keystrokes and action it can incorrectly take display
precedence even if its context predicate does not pass.
- Fix is to check that the highest precedence binding is a full match.
To do this efficiently, it's based on an index within the keymap
bindings.
Also adds `highest_precedence_binding_*` variants which avoid the
inefficiency of building lists of bindings just to use the last.
Release Notes:
- Fixed display of keybindings to skip bindings that are shadowed by a
binding that uses the same keystrokes.
- Fixed display of `editor::AcceptEditPrediction` bindings to use the
normal precedence that prioritizes user bindings.
Cleaner to have references to this be `toast::RunAction` matching how it
appears in the keymap, instead of `workspace::RunAction`.
Release Notes:
- N/A
Closes#27998
Also fixed an issue where jumping back from closing to opening tags
didn't work in javascript due to missing brackets in our tree-sitter
query.
Release Notes:
- N/A
---------
Co-authored-by: Conrad Irwin <conrad.irwin@gmail.com>
Closes#31648
This PR adds support for:
- Expanding multiple cursors above/below
- Expanding multiple selections above/below
- Adding new cursors/selections when expansion has already been done.
Existing expansions preserve their state and expand/shrink according to
the action, while new cursors/selections act like freshly created ones.
Tests for both cursor and selections:
- below/above cases
- undo/redo cases
- adding new cursors/selections with existing expansion
Before/After:
https://github.com/user-attachments/assets/d2fd556b-8972-4719-bd86-e633d42a1aa3
Release Notes:
- Improved `AddSelectionAbove` and `AddSelectionBelow` to extend
multiple cursors/selections.
* `state.last_options` was never being updated, so the caching wasn't
working.
* If node on the PATH was too old it was logging errors on every
invocation even though managed node is being used.
Release Notes:
- Fixed caching of Node.js runtime paths and improved error messages.
Replace hardcoded 0.10 threshold with configurable parameter and set
0.05 default for most tests, with 0.2 for from_pixels_constructor
eval that produces more mismatched tags.
Release Notes:
- N/A
Closes#27567
Release Notes:
- Fixed default keybindings for `editor::AcceptPartialEditPrediction` to
work with subtle mode.
Co-authored-by: Richard <richard@zed.dev>
Closes [#13107](https://github.com/zed-industries/zed/issues/13107)
Enabled pull diagnostics by default, for the language servers that
declare support in the corresponding capabilities.
```
"diagnostics": {
"lsp_pull_diagnostics_debounce_ms": null
}
```
settings can be used to disable the pulling.
Release Notes:
- Added support for the LSP `textDocument/diagnostic` command.
# Brief
This is draft PR that implements the LSP `textDocument/diagnostic`
command. The goal is to receive your feedback and establish further
steps towards fully implementing this command. I tried to re-use
existing method and structures to ensure:
1. The existing functionality works as before
2. There is no interference between the diagnostics sent by a server and
the diagnostics requested by a client.
The current implementation is done via a new LSP command
`GetDocumentDiagnostics` that is sent when a buffer is saved and when a
buffer is edited. There is a new method called `pull_diagnostic` that is
called for such events. It has debounce to ensure we don't spam a server
with commands every time the buffer is edited. Probably, we don't need
the debounce when the buffer is saved.
All in all, the goal is basically to get your feedback and ensure I am
on the right track. Thanks!
## References
1.
https://microsoft.github.io/language-server-protocol/specifications/lsp/3.17/specification/#textDocument_pullDiagnostics
## In action
You can clone any Ruby repo since the `ruby-lsp` supports the pull
diagnostics only.
Steps to reproduce:
1. Clone this repo https://github.com/vitallium/stimulus-lsp-error-zed
2. Install Ruby (via `asdf` or `mise).
4. Install Ruby gems via `bundle install`
5. Install Ruby LSP with `gem install ruby-lsp`
6. Check out this PR and build Zed
7. Open any file and start editing to see diagnostics in realtime.
https://github.com/user-attachments/assets/0ef6ec41-e4fa-4539-8f2c-6be0d8be4129
---------
Co-authored-by: Kirill Bulatov <mail4score@gmail.com>
Co-authored-by: Kirill Bulatov <kirill@zed.dev>
Problem:
Initial keymap has "j k" keymap, which if uncommented will add +-1s
delay to every "j" key press
This workaround was taken from
https://github.com/zed-industries/zed/discussions/6661
Release Notes:
- N/A *or* Added/Fixed/Improved ...
---------
Co-authored-by: Michael Sloan <michael@zed.dev>
- [x] pass LSP tasks into list_debug_scenarios
- [x] load LSP tasks only once for both modals
- [x] align ordering
- [x] improve appearance of LSP debug task icons
- [ ] reconsider how `add_current_language_tasks` works
- [ ] add a test
Release Notes:
- Debugger Beta: Added debuggable LSP tasks to the "Debug" tab of the
new process modal.
---------
Co-authored-by: Anthony Eid <hello@anthonyeid.me>
Here’s how it looks after the fix:
.
White is white and black is black, as intended.
In some cases, dimmed colors were poorly defined, so I took
`text.dimmed` values.
Note that white is defined exactly as the background color for light
themes. Similarly, black is the exact background color for dark themes.
I didn’t change this, but many themes intentionally make white and black
slightly different from the background color. This prevents issues where
programs assume, say, a dark background and set the foreground to white,
making text invisible. I'm not sure if we want to adjust these themes to
address this; just noting it here.
Closes#29379
Release Notes:
- Fixed ANSI black and ANSI white colors in built-in themes
Closes#31438
Release Notes:
- agent: Fixed an edge case were the request would fail when using
Claude and multiple images were attached
---------
Co-authored-by: Richard Feldman <oss@rtfeldman.com>
I think using `Debugger Program` and `~/bin/debugger` here makes it seem
like that field is for specifying the path to the debugger itself, and
not the program being debugged.
Release Notes:
- N/A
This PR improves key handling on Windows by moving the
`TranslateMessage` call from the message loop to after we handled
`WM_KEYDOWN`. This brings Windows behavior more in line with macOS and
gives us finer control over key events. As a result, Vim keybindings now
work properly even when an IME is active. The trade-off is that it might
introduce a slight delay in text input.
Release Notes:
- N/A
We run the unit evals once a day in the middle of the night, and trigger
a Slack post if it fails.
Release Notes:
- N/A
---------
Co-authored-by: Oleksiy Syvokon <oleksiy.syvokon@gmail.com>
The commit #31073 had introduced `zed-main.rs`, which replaced the
previous `main.rs` to be the "true" entry of the whole program. But as
the macro `#![cfg_attr(not(debug_assertions), windows_subsystem =
"windows")]` only works in the "true" entry, the release build will also
arise the console window if this macro doesn't move to the new entry
(the `zed-main.rs` here).
Release Notes:
- N/A
This problem seems to be specific to Opus 4. Eval shows improvement from
89% to 97%.
Closes: https://github.com/zed-industries/zed/issues/32060
Release Notes:
- N/A
Co-authored-by: Ben Brandt <benjamin.j.brandt@gmail.com>
Based on a Slack conversation with @notpeter - this prevents secrets in
private/excluded files from being sent by the agent to third parties for
tools that don't require confirmation.
Of course, the agent can still use the terminal tool or MCP to access
these, but those require confirmation before they run (unlike these
tools).
This change doesn't seem to cause any trouble for evals:
<img width="730" alt="Screenshot 2025-06-03 at 8 48 33 PM"
src="https://github.com/user-attachments/assets/d90221be-f946-4af2-b57b-4aa047e86853"
/>
Release Notes:
- N/A
This brings back [linux(x11): Add support for pasting images from
clipboard · Pull Request
#29387](https://github.com/zed-industries/zed/pull/29387) while fixing
#30523 (which caused it to be reverted).
Commit message from that PR:
> Closes:
https://github.com/zed-industries/zed/pull/29177#issuecomment-2823359242
>
> Removes dependency on
[quininer/x11-clipboard](https://github.com/quininer/x11-clipboard) as
it is in [maintenance
mode](https://github.com/quininer/x11-clipboard/issues/19).
>
> X11 clipboard functionality is now built-in to GPUI which was
accomplished by stripping the non-x11-related code/abstractions from
[1Password/arboard](https://github.com/1Password/arboard) and extending
it to support all image formats already supported by GPUI on wayland and
macos.
>
> A benefit of switching over to the `arboard` implementation, is that
we now make an attempt to have an X11 "clipboard manager" (if available
- something the user has to setup themselves) save the contents of
clipboard (if the last copy operation was within Zed) so that the copied
contents can still be pasted once Zed has completely stopped.
Before the fix for reapply, it was iterating through the formats and
requesting conversion to each. Some clipboard providers just respond
with a different format rather than saying the format is unsupported.
The fix is to use this response if it matches a supported format. It
also now typically avoids this iteration by requesting the `TARGETS` and
taking the highest precedence supported target.
Closes#30523
Release Notes:
- Linux (X11): Restored the ability to paste images.
---------
Co-authored-by: Ben <ben@zed.dev>
Fix a segfault in CVDisplayLink
We see 1-2 crashes a day on macOS on the `CVDisplayLink` thread.
```
Segmentation fault: 11 on thread 9325960 (CVDisplayLink)
CoreVideo CVHWTime::reset()
CoreVideo CVXTime::reset()
CoreVideo CVDisplayLink::runIOThread()
libsystem_pthread.dylib _pthread_start
libsystem_pthread.dylib thread_start
```
With the help of the Zed AI, I dove into the crash report, which looks
like this:
```
Crashed Thread: 49 CVDisplayLink
Exception Type: EXC_BAD_ACCESS (SIGSEGV)
Exception Codes: KERN_INVALID_ADDRESS at 0x00000000000001f6
Exception Codes: 0x0000000000000001, 0x00000000000001f6
Thread 49 Crashed:: CVDisplayLink
0 CoreVideo 0x18c1ed994 CVHWTime::reset() + 64
1 CoreVideo 0x18c1ee474 CVXTime::reset() + 52
2 CoreVideo 0x18c1ee198 CVDisplayLink::runIOThread() + 176
3 libsystem_pthread.dylib 0x18285ac0c _pthread_start + 136
4 libsystem_pthread.dylib 0x182855b80 thread_start + 8
Thread 49 crashed with ARM Thread State (64-bit):
x0: 0x0000000000000000 x1: 0x000000018c206e08 x2: 0x0000002c00001513 x3: 0x0001d4630002a433
x4: 0x00000e2100000000 x5: 0x0001d46300000000 x6: 0x000000000000002c x7: 0x0000000000000000
x8: 0x000000000000002e x9: 0x000000004d555458 x10: 0x0000000000000000 x11: 0x0000000000000000
x12: 0x0000000000000000 x13: 0x0000000000000000 x14: 0x0000000000000000 x15: 0x0000000000000000
x16: 0x0000000182856a9c x17: 0x00000001f19bc540 x18: 0x0000000000000000 x19: 0x0000600003c56ed8
x20: 0x000000000002a433 x21: 0x0000000000000000 x22: 0x0000000000000000 x23: 0x0000000000000000
x24: 0x0000000000000000 x25: 0x0000000000000000 x26: 0x0000000000000000 x27: 0x0000000000000000
x28: 0x0000000000000000 fp: 0x000000016b02ade0 lr: 0x000000018c1ed984
sp: 0x000000016b02adc0 pc: 0x000000018c1ed994 cpsr: 0x80001000
far: 0x00000000000001f6 esr: 0x92000006 (Data Abort) byte read Translation fault
Binary Images:
0x1828c9000 - 0x182e07fff com.apple.CoreFoundation (6.9) <df489a59-b4f6-32b8-9bb4-9b832960aa52> /System/Library/Frameworks/CoreFoundation.framework/Versions/A/CoreFoundation
```
Using lldb to disassemble `CVHWTime::reset()` (and the AI to interpret
it), the crash is caused by dereferencing the pointer at the start of
the CVHWTime struct + 0x1c8. In this case the pointer has (the clearly
nonsense) value 0x2e (and 0x2e + 0x1c8 = 0x1f6, the failing address).
As to how this could happen...
Looking at the implementation of `CVDisplayLinkRelease`, it calls
straight into `CFRelease` on the main thread; and so it is not safe to
call `CVDisplayLinkRelease` concurrently with other threads that access
the CVDisplayLink. While we already stopped the display link, it turns
out that `CVDisplayLinkStop` just sets a flag on the struct to instruct
the io-thread to exit "soon", and returns immediately. That means we
don't know when the other thread will actually exit, and so we can't
safely call `CVDisplayLinkRelease`.
So, for now, we just leak these objects. They should be created
relatively infrequently (when the app is foregrounded/backgrounded), so
I don't think this is a huge problem.
Release Notes:
- Fix a rare crash on macOS when putting the app in the background.
