I ran `scripts/linux` on Debian Trixie 13. It suggested manually
installing Mold, but [mold](http://packages.debian.org/mold) is packaged
on Debian and so we could install it automatically.
The version packaged there seems to work well enough for `cargo t` to
pass, at least.
## Tested
```
; sudo apt remove mold libstdc++-14-dev
... uninstalls them
; ./script/linux
The following NEW packages will be installed:
build-essential clang clang-19 clang-tools-19 g++ g++-14 g++-14-x86-64-linux-gnu g++-x86-64-linux-gnu libstdc++-14-dev mold
; cargo t
(passes)
```
Release Notes:
- N/A
Previously, `cargo test --package component` failed due to missing
imports for a doctest:
```
Doc-tests component
running 1 test
test crates/component/src/component.rs - Component::description (line 229) ... FAILED
failures:
---- crates/component/src/component.rs - Component::description (line 229) stdout ----
error: cannot find derive macro `Documented` in this scope
--> crates/component/src/component.rs:231:10
|
4 | #[derive(Documented)]
| ^^^^^^^^^^
error[E0599]: no associated item named `DOCS` found for struct `MyComponent` in the current scope
--> crates/component/src/component.rs:236:20
|
5 | struct MyComponent;
| ------------------ associated item `DOCS` not found for this struct
...
9 | Some(Self::DOCS)
| ^^^^ associated item not found in `MyComponent`
error: aborting due to 2 previous errors
For more information about this error, try `rustc --explain E0599`.
Couldn't compile the test.
failures:
crates/component/src/component.rs - Component::description (line 229)
test result: FAILED. 0 passed; 1 failed; 0 ignored; 0 measured; 0 filtered out; finished in 0.29s
error: doctest failed, to rerun pass `-p component --doc`
bobcat ~/src/zed (doctests) 18:33
```
This might be unnoticed if you mostly run nextest, as it does not run
doctests.
Release Notes:
- N/A
Follow-up: #36700
This PR adds basic support for showing images inline inside a text.
As you can see inside the before screenshot, the image was displayed
right below the `Some inline text` text. This was because we didn't
consider the image to be inline with the text (paragraph). Now we do :)
All the test changes are making sure it is not more than 1 element
parsed, instead of only checking for the first parsed element. This
could work out bad when we return more than 1 result.
**Before**
<img width="1717" height="1344" alt="Screenshot 2025-08-31 at 13 49 45"
src="https://github.com/user-attachments/assets/13c5f9dd-0e0a-4e08-b2a6-28e9a4e0cab8"
/>
**After**
<img width="1719" height="1343" alt="Screenshot 2025-08-31 at 13 42 14"
src="https://github.com/user-attachments/assets/bf7aa82f-3743-4fb3-87aa-4a97a550c4d1"
/>
**Code example**:
```markdown
<p>some inline text <img src="https://picsum.photos/200/300" alt="Description of image" style="height: 100px" /> asdjkflsadjfl</p>
# Html Tag
<img src="https://picsum.photos/200/300" alt="Description of image" />
# Html Tag with width and height
<img src="https://picsum.photos/200/300" alt="Description of image" width="100" height="200" />
# Html Tag with style attribute with width and height
<img src="https://picsum.photos/200/300" alt="Description of image" style="width: 100px; height: 200px" />
# Normal Tag
```
Release Notes:
- Markdown: Added support for inline HTML `img` tags inside paragraphs
Closes#37093
Also check this: #37099.
So currently in zed for both OpenAI and OpenAI Compatible provider when
the url is changed from settings the api_key stored in the provider
state is not cleared and it is still used. But if you restart zed the
api_key is cleared. Currently zed uses the api_url to store and fetch
the api key from credential provider. The behaviour is not changed
overall, it's just that we have made it consistent it with the zed
restart logic where it re-authenticates and fetches the api_key again. I
have attached the video below to show case before and after of this.
So all in all the problem was we were not re-authenticating the in case
api_url change while zed is still running. Now we trigger a
re-authentication and clear the state in case authentication fails.
OpenAI Compatible Provider:
| Before | After |
|--------|--------|
| <video
src="https://github.com/user-attachments/assets/324d2707-ea72-4119-8981-6b596a9f40a3"
/> | <video
src="https://github.com/user-attachments/assets/cc7fdb73-8975-4aaf-a642-809bb03ce319"
/> |
OpenAI Provider:
| Before | After |
|--------|--------|
| <video
src="https://github.com/user-attachments/assets/a1c07d1b-1909-4b49-b33c-fc05123e92e7"
/> | <video
src="https://github.com/user-attachments/assets/d78aeccd-5cd3-4d0c-8b9f-6f98e499d7c8"
/> |
Release Notes:
- Fixed OpenAI and OpenAI Compatible provide API keys being persisted
when changing the API URL setting. Authentication is now properly
revalidated when settings change.
---------
Signed-off-by: Umesh Yadav <git@umesh.dev>
Fix an issue that resulted in Ollama models not being able to not being
able to access the input of the commands they executed (only being able
to access the result).
This properly return the function history as shown in
https://github.com/ollama/ollama/blob/main/docs/api.md#chat-request-with-history-with-tools
Previously, function input where not returned and result where returned
as a "user" role.
Release Notes:
- ollama: Improved format when returning tool results to the models
Closes#37597
Release Notes:
- N/A
---
## Problem
When using "Tab Switcher: Toggle All", temporary files (untitled buffers
without associated file paths) cannot be searched by their displayed
content. This creates an inconsistent user experience where:
- **UI Display**: Shows dynamic titles based on the first line of
content (up to 40 characters)
- **Search Text**: Only searches for the static text "untitled"
### Example
- A temporary file containing `Hello World` is displayed as "Hello
World" in the tab
- However, searching for "Hello" in Tab Switcher returns no results
- Only searching for "untitled" will find this temporary file
## Root Cause
The issue stems from inconsistent title generation logic between display
and search:
1. **Display Title** (`items.rs:724`): Uses `self.title(cx)` →
`MultiBuffer::title()` → `buffer_content_title()`
- Returns the first line of content (max 40 chars) for temporary files
2. **Search Text** (`items.rs:650-656`): Uses `tab_content_text()`
method
- Returns hardcoded "untitled" for files without paths
## Solution
Modified the `tab_content_text()` method in `crates/editor/src/items.rs`
to use the same logic as the displayed title for consistency:
```rust
fn tab_content_text(&self, detail: usize, cx: &App) -> SharedString {
if let Some(path) = path_for_buffer(&self.buffer, detail, true, cx) {
path.to_string_lossy().to_string().into()
} else {
// Use the same logic as the displayed title for consistency
self.buffer.read(cx).title(cx).to_string().into()
}
}
```
Also:
* Adds tests for can_collect_data.
* Temporarily removes collection of diagnostics.
Release Notes:
- Edit Prediction: Fixed a bug where requests were marked eligible for
data collection despite the recent edit history in the request involving
files that may not be open source. The requests affected by this bug
will not be used in training data.
Minor docs fix.
Seems like 0a4ff2f475 accidentally added
"hidden" to the docs of both – `close_position` and `show_close_button`.
Release Notes:
- N/A
Co-authored-by: tastenbier <>
Follow-up of https://github.com/zed-industries/zed/pull/37352
Closes https://github.com/zed-industries/zed/issues/37132
* disabled word completions in the agent panel's editor
* if not disabled, allow to trigger word completions with an action even
if the completions threshold is not reached
Release Notes:
- Fixed word completions appearing in the agent panel's editor and not
appearing when triggered with the action before the completion threshold
is reached
This PR switches the OpenRouter integration from fetching all models to
fetching only the models specified in the user's account preferences.
This will help improve the experience
**The Problem**
The previous implementation used the `/models` endpoint, which returned
an exhaustive list of all models supported by OpenRouter. This resulted
in a long and cluttered model selection dropdown in Zed, making it
difficult for users to find the models they actually use.
**The Solution**
We now use the `/models/user` endpoint. This API call returns a curated
list based on the models and providers the user has selected in their
[OpenRouter dashboard](https://openrouter.ai/models).
Ref: [OpenRouter API Docs for User-Filtered
Models](https://openrouter.ai/docs/api-reference/list-models-filtered-by-user-provider-preferences)
Release Notes:
- language_models: Support OpenRouter user preferences for available
models
Closes#37302
Related: #37614
In case of open_ai_compatible providers like Zhipu AI and z.ai they
return empty content along with usage data. below is the example json
captured from z.ai. We now ignore empty content returned by providers
now to avoid this issue where we would return the same empty content
back to provider which would error out.
```
OpenAI Stream Response JSON:
{
"id": "2025090518465610d80dc21e66426d",
"created": 1757069216,
"model": "glm-4.5",
"choices": [
{
"index": 0,
"finish_reason": "tool_calls",
"delta": {
"role": "assistant",
"content": ""
}
}
],
"usage": {
"prompt_tokens": 7882,
"completion_tokens": 150,
"total_tokens": 8032,
"prompt_tokens_details": {
"cached_tokens": 7881
}
}
}
```
Release Notes:
- Skip empty delta text content in OpenAI and OpenAI compatible provider
Signed-off-by: Umesh Yadav <git@umesh.dev>
Closes#37473
### Background
Previously, we enabled foreign keys at all times for our sqlite database
that we use for client-side state.
The problem with this is that In sqlite, `alter table` is somewhat
limited, so for many migrations, you must *recreate* the table: create a
new table called e.g. `workspace__2`, then copy all of the data from
`workspaces` into `workspace__2`, then delete the old `workspaces` table
and rename `workspaces__2` to `workspaces`. The way foreign keys work in
sqlite, when we delete the old table, all of its associated records in
other tables will be deleted due to `on delete cascade` clauses.
Unfortunately, one of the types of associated records that can be
deleted are `editors`, which sometimes store unsaved text. It is very
bad to delete these records, as they are the *only* place that this
unsaved text is stored.
This has already happened multiple times as we have migrated tables as
we develop Zed, but I caused it to happened again in
https://github.com/zed-industries/zed/pull/36714.
### The Fix
The Sqlite docs recommend a multi-step approach to migrations where you:
* disable foreign keys
* start a transaction
* create a new table
* populate the new table with data from the old table
* delete the old table
* rename the new table to the old name
* run a foreign key check
* if it passes, commit the transaction
* enable foreign keys
In this PR, I've adjusted our sqlite migration code path to follow this
pattern more closely. Specifically, we disable foreign key checks before
running migrations, run a foreign key check before committing, and then
enable foreign key checks after the migrations are done.
In addition, I've added a generic query that we run *before* running the
foreign key check that explicitly deletes any rows that have dangling
foreign keys. This way, we avoid failing the migration (and breaking the
app) if a migration deletes data that *does* cause associated records to
need to be deleted.
But now, in the common case where we migrate old data in the new table
and keep the ids, all of the associated data will be preserved.
Release Notes:
- Fixed a bug where workspace state would be lost when upgrading from
Zed 0.201.x. or below.
- **toolchains: Add new state to toolchain selector**
- **Use toolchain term for Add Toolchain button**
- **Hoist out a meta function for toolchain listers**
Closes#27332
Release Notes:
- python: Users can now specify a custom path to their virtual
environment from within the picker.
---------
Co-authored-by: Danilo Leal <daniloleal09@gmail.com>
Closes#36978
This PR fixes an issue that we would only show the first `root -> child`
session in compact mode, but the session that came after it, we would
only show the child session label instead of also adding the parent
label due to compact mode.
## Before
<img width="348" height="173" alt="Screenshot 2025-08-27 at 22 18 39"
src="https://github.com/user-attachments/assets/ad6afd3a-196d-497f-812a-00698676ee90"
/>
## After
<img width="563" height="211" alt="Screenshot 2025-08-27 at 21 57 16"
src="https://github.com/user-attachments/assets/a953ef2a-a796-4160-b868-96e96f81c858"
/>
With 3 parent + child sessions and one parent session only.
<img width="484" height="223" alt="Screenshot 2025-08-27 at 22 22 13"
src="https://github.com/user-attachments/assets/a26f79a4-63a5-43d0-a714-d62cb1995e6e"
/>
cc @cole-miller I know we hacked on this some while ago, so figured you
might be the best guy to ask for a review.
Release Notes:
- Debugger: Fix to allow showing more than 1 compact session item
---------
Co-authored-by: Anthony <anthony@zed.dev>
Closes#37520
This change makes the attach modal load processes from the remote server
when connecting via SSH, rather than showing local processes from the
client machine.
This works by using the new GetProcessesRequest RPC message to allow
downstream clients to get the correct processes to display. It also only
works with downstream ssh clients because the message handler is only
registered on headless projects.
Release Notes:
- debugger: Fix bug where SSH attach modal showed local processes
instead of processes from the server
Bug:
<img width="196" height="95" alt="Screenshot 2025-09-06 at 1 21 39 AM"
src="https://github.com/user-attachments/assets/66ec0fc9-961e-4289-bd75-68b24dad485e"
/>
The fold marker we use, `⋯`, isn’t rendered at the same size as the
editor’s font. Notice how the fold marker appears larger than the same
character typed directly in the editor buffer.
<img width="146" height="82" alt="image"
src="https://github.com/user-attachments/assets/a059d221-6b55-4cf9-bc1e-898ff5444006"
/>
When we shape the line, we use the editor’s font size, and it ends up
determining the element’s width. To fix this, we should treat the
ellipsis as a UI element rather than a buffer character, since current
visual size looks good to me.
<img width="196" height="95" alt="Screenshot 2025-09-06 at 1 29 28 AM"
src="https://github.com/user-attachments/assets/1b766d46-00ab-40c7-b98a-95ea2d4b29bf"
/>
Release Notes:
- Fixed an issue where the fold placeholder’s hover area was smaller
than the marker.
Follow-up to https://github.com/zed-industries/zed/pull/37489
Notification panel settings were always missing the content, hence this
PR adds it. After #37489, the use of the same content twice broke
things, which currently makes the notification panel non-configurable on
Nightly. This PR fixes this.
There once was an issue about the documentation for the panel being
wrong as well. However, I was just unable to find that sadly.
Release Notes:
- N/A
Co-Authored-By: Brandan <b5@n0.computer>
Release Notes:
- Added a new action `terminal::Toggle` that is by default bound to
'ctrl-\`'. This copies the default behaviour from VSCode and Jetbrains
where the terminal opens and closes correctly. If you'd like the old
behaviour you can rebind 'ctrl-\`' to `terminal::ToggleFocus`
Co-authored-by: Brandan <b5@n0.computer>
This commit updates the implementation for
`editor::Editor.manipulate_text` to use
`editor::selections_collection::SelectionsCollection.all_adjusted`
instead of `editor::selections_collection::SelectionsCollection.all`, as
the former takes into account the selection's `line_mode`, fixing the
issue where, if an user was in vim's visual line mode, running the
`editor: convert to upper case` command would not work as expected.
Closes#36953
Release Notes:
- Fixed bug where using the editor's convert case commands while in
vim's Visual Line mode would not work as expected
macOS versions are currently reported as `macOS 26.0.0`.
But this makes it impossible to differentiate amongst macOS Beta
releases which have the same version number (`X.0.0`) but are different
builds.
This PR adds build number info to `os_version` for macOS Betas and
[Rapid Security Response](https://support.apple.com/en-us/102657)
release that have identical version numbers to stable release, but have
different builds numbers. We can differentiate them because the build
numbers end with a letter.
| Version | Before | After |
| - | - | - |
| macOS Sonoma 14.7.8 | 14.7.8 | 14.7.8 |
| macOS Sequoia 15.6.1 | 15.6.1 | 15.6.1 |
| mcOS Ventura 13.3.1 | 13.3.1 | 13.3.1 |
| macOS Ventura 13.3.1 (a) | 13.3.1 | 13.3.1 (Build 22E772610a) |
| macOS Tahoe 26.0.0 (Beta1) | 26.0.0 | 26.0.0 (Build 25A5316a) |
| macOS Tahoe 26.0.0 (Beta5) | 26.0.0 | 26.0.0 (Build 25A5349a) |
This should cause minimal telemetry changes and only impacting a macOS
betas and a couple specific older macOS versions, but will allow
differentiation between macOS beta releases in GitHub issues.
Alternatives:
1. Leave as-is (can't differentiate between macOS beta builds)
2. Always include build number info (impacts telemetry; more consistent
going forward; differentiates non-final Release Candidates which don't
include a trailing letter)
I couldn't find a cocoa method to retrieve macOS build number, so I
switched dependencies from `cocoa` to `objc2-foundation` in the client
crate. We already depend upon this crate as a dependency of
`blade-graphics` so I matched the features of that and so workspace-hack
doesn't change.
1ebc69a447/tooling/workspace-hack/Cargo.toml (L355)
Release Notes:
- N/A
This is a small fix for default values in task variables. The
[documentation](https://zed.dev/docs/tasks) states
> You can also use verbose syntax that allows specifying a default if a
given variable is not available: ${ZED_FILE:default_value}
I found, however, that this doesn't actually work. Instead, the Zed
variable and the default value are just appended in the output. For
example, if I run a task `echo ${ZED_ROW:100}` the result I get is
`447:100` (in this case it should just be `447`).
This PR fixes that. I also added a new test case for handling default
values.
I also tested the fix in a dev build and it seems to work.
There are no UI adjustments.
AI disclosure: I used Claude Code to write the code, including the fix
and the tests.
This is actually my first open-source PR ever, so if I did something
wrong, I'd appreciate any tips and I'll make it right!
Release Notes:
- Fixed task variable substitution always appending the default
In an effort to improve the experience while developing extensions and
improving themes, this PR updates the syntax tree views behavior
slightly.
Before, the view would always update to the current active editor whilst
being used. This was quite painful for improving extension scheme files,
as you would always have to change back and forth between editors to
have a view at the relevant syntax tree.
With this PR, the syntax tree view will now stay attached to the editor
it was opened in, similar to preview views. Once the view is shown, the
`UseActiveEditor` will become available in the command palette and
enable the user to update the view to the last focused editor. On file
close, the view will also be updated accordingly.
https://github.com/user-attachments/assets/922075e5-9da0-4c1d-9e1a-51e024bf41ea
A button is also shown whenever switching is possible.
Futhermore, improved the empty state of the view.
Lastly, a drive-by cleanup of the `show_action_types` method so there is
no need to call `iter()` when calling the method.
Release Notes:
- The syntax tree view will now stay attached to the buffer it was
opened in, similar to the Markdown preview. Use the `UseActiveEditor`
action when the view is shown to change it to the last focused editor.
Doesn't fix anything, but it seems that we do not need to assert and
convert into an error until after the closure run to completion,
especially since this is the only error we throw.
Release Notes:
- N/A
This isn't quite right, but using the env manager path causes conda to
scream and I am not yet sure why, either way this is an improvement over
the status quo
Release Notes:
- N/A\
