- On macOS, pasting now inserts the actual file path when the clipboard
contains a file URL (public.file-url/public.url)
- Terminal paste remains text-only; no temp files or data URLs are
created. If only raw image bytes exist on the clipboard, paste is a
no-op.
- Scope: macOS only; no dependency changes.
- Added a test (test_file_url_converts_to_path) that verifies URL→path
conversion using a unique pasteboard.
Release Notes:
- Improved pasting on macOS: now inserts the actual file path when the
clipboard contains a file URL (enables image paste support for Claude
Code)
On macOS, the Function key is reserved for system use and should not be
used in application code.
This commit updated keystroke matching and key event handling to ignore
the Function key modifier while users are typing or pressing
keybindings.
For some keyboards with compact layout (like my 65% keyboard), there is
no separated backtick key. Esc and it shares the same physical key. To
input backtick, users may press `Fn-Esc`. However, macOS will still
deliver events with Fn key modifier to applications. Cocoa framework can
handle this correctly, which typically ignore the Fn directly. GPUI
should also follow the same rule, otherwise, the backtick key on those
keyboards won't work.
Release Notes:
- Fixed a bug where typing fn-\` on macOS would not insert a `.
This makes it possible to do login via things like `cmd: "node", args:
["my-node-file.js", "login"]`
Also, that command will now use Zed's managed `node` instance.
Release Notes:
- ACP extensions can now run terminal login commands using relative
paths
Closes#41471
We were killing the crash handler when it received a second copy of any
of the messages, but this GPU specs one is sent on each new window
rather than once at startup. We could gate the sending to only happen
once, but it's simpler to just allow multiple gpu specs messages.
Release Notes:
- N/A
## Summary
Fixes#36754
This PR fixes an issue where LSPs fail to spawn after the crash handler
is initialized.
## Problem
After PR #35263 added minidump crash reporting, some users experienced
LSP spawn failures. The issue manifests as:
- LSPs fail to spawn with no clear error messages
- The problem only occurs after crash handler initialization
- LSPs work when a debugger is attached, revealing a timing issue
### Root Cause
The crash handler installs Mach exception ports for minidump generation.
Due to a timing issue, child processes inherit these exception ports
before they're fully stabilized, which can block child process spawning.
## Solution
Reset exception ports in child processes using the `pre_exec()` hook,
which runs after `fork()` but before `exec()`. This prevents children
from inheriting the parent's crash handler exception ports.
### Implementation
- Adds macOS-specific implementation of `new_smol_command()` that resets
exception ports before exec
- Calls `task_set_exception_ports` to reset all exception ports to
`MACH_PORT_NULL`
- Graceful error handling: logs warnings but doesn't fail process
spawning if port reset fails
Release Notes:
- Fixed LSPs failing to spawn on some macOS systems
---------
Co-authored-by: Julia Ryan <juliaryan3.14@gmail.com>
The Alt+Numpad buffer that's maintained by the input stack is getting
corrupted, leading to garbage characters being inserted on keystrokes
like Alt+Up. Disable the automatic handling of Alt+Numpad for now until
the cause of this corruption is understood. The Alt+Numpad input did not
work anyway, so this does not regress anything.
Release Notes:
- windows: Fixed corrupted characters being inserted when Alt is pressed
(preview only)
Release Notes:
Updated how clicking on multi-buffer headers works to provide better
control and prevent unexpected navigation:
Clicking the header now collapses/expands the file section instead of
opening the file.
Opening files can be done by clicking the filename or the "Open file"
button on the right side of the header.
Existing shortcuts continue to work: use the left chevron to collapse or
your keyboard shortcut to jump to the file
**Demo:**
https://github.com/user-attachments/assets/dca9ccc5-bd98-416c-97af-43b4e4b2f903
Reverts zed-industries/zed#41466
This PR would add "in_snippet" context when there wasn't a completion
menu visible, causing some actions to not be hit.
Release Note:
- N/A
This PR changes the loading/generating indicator when in the "waiting
for tool call confirmation" state so that's a bit more visible and
discernible as needing your attention, as opposed to a regular
generating state.
<img width="400" alt="Screenshot 2025-11-05 at 10 46@2x"
src="https://github.com/user-attachments/assets/88adbf97-20fb-49c4-9c77-b0a3a22aa14e"
/>
Release Notes:
- agent: Improved the "waiting for confirmation" state visibility so
that you more rapidly know the agent is waiting for you to act.
Mostly just removing "thread" from all external agent menu items; I
think we can do without it and it already becomes much better/cleaner.
Release Notes:
- N/A
I am using an Azure OpenAI instance since that is what is provided at
work and with how they have it setup not all responses contain a delta,
which lead to errors and truncated responses. This is related to how
they are filtering potentially offensive requests and responses. I don't
believe this filter was made in-house, instead I believe it is provided
by Microsoft/Azure, so I suspect this fix may help other users.
Release Notes:
- N/A
Co-authored-by: Bennet Bo Fenner <bennetbo@gmx.de>
This brings the documentation on Java in line with the much changed
reality of the Java extension.
Note that the correctness of this is contingent on
https://github.com/zed-industries/extensions/pull/3745 being merged.
Release Notes:
- N/A
Closes#40632
**TL;DR:** The `wrap selections in tag` action was unavailable in ERB
files, even when the cursor was positioned in HTML content (outside of
Ruby code blocks). This happened because `syntax_layer_at()` incorrectly
returned the Ruby language for positions that were actually in HTML.
**NOTE:** I am not familiar with that part of Zed so it could be that
the fix here is completely incorrect.
Previously, `syntax_layer_at` incorrectly reported injected languages
(e.g., Ruby in ERB files) even when the cursor was in the base language
content (HTML). This broke actions like `wrap selections in tag` that
depend on language-specific configuration.
The issue had two parts:
1. Missing start boundary check: The filter only checked if a layer's
end was after the cursor (`end_byte() > offset`), not if it started
before, causing layers outside the cursor position to be included. See
the `BEFORE` video: when I click on the HTML part it reports `Ruby`
language instead of `HTML`.
2. Wrong boundary reference for injections: For injected layers with
`included_sub_ranges` (like Ruby code blocks in ERB), checking the root
node boundaries returned the entire file range instead of the actual
injection ranges.
This fix:
- Adds the containment check using half-open range semantics [start,
end) for root node boundaries. That ensures proper reporting of the
detected language when a cursor (`|`) is located right after the
injection:
```
<body>
<%= yield %>|
</body>
```
- Checks `included_sub_ranges` for injected layers to determine if the
cursor is actually within an injection
- Falls back to root node boundaries for base layers without sub-ranges.
This is the original behavior.
Fixes ERB language support where actions should be available based on
the cursor's actual language context. I think that also applies to some
other template languages like HEEX (Phoenix) and `*.pug`. On short
videos below you can see how I navigate through the ERB template and the
terminal on the right outputs the detected language if you apply the
following patch:
```diff
diff --git i/crates/editor/src/editor.rs w/crates/editor/src/editor.rs
index 15af61f5d2..54a8e0ae37 100644
--- i/crates/editor/src/editor.rs
+++ w/crates/editor/src/editor.rs
@@ -10671,6 +10671,7 @@ impl Editor {
for selection in self.selections.disjoint_anchors_arc().iter() {
if snapshot
.language_at(selection.start)
+ .inspect(|language| println!("Detected language: {:?}", language))
.and_then(|lang| lang.config().wrap_characters.as_ref())
.is_some()
{
```
**Before:**
https://github.com/user-attachments/assets/3f8358f4-d343-462e-b6b1-3f1f2e8c533d
**After:**
https://github.com/user-attachments/assets/c1b9f065-1b44-45a2-8a24-76b7d812130d
Here is the ERB template:
```
<!DOCTYPE html>
<html>
<head>
<meta http-equiv="Content-Type" content="text/html; charset=utf-8">
<style>
/* Email styles need to be inline */
</style>
</head>
<body>
<%= yield %>
</body>
</html>
```
Release Notes:
- N/A
Hi, this is a follow-up change for
https://github.com/zed-industries/zed/pull/41754 I think it important to
keep existing things working. So add notes to the Ruby extension doc
about enabling Ruby LSP for ERB files as well. Thanks!
Release Notes:
- N/A
Works around an issue, which can be reproduced in the following program:
```rs
use windows::Win32::UI::Input::KeyboardAndMouse::{GetKeyboardState, VK_CONTROL};
fn main() {
let mut keyboard_state = [0u8; 256];
unsafe {
GetKeyboardState(&mut keyboard_state).unwrap();
}
let ctrl_down = (keyboard_state[VK_CONTROL.0 as usize] & 0x80) != 0;
println!("Is Ctrl down: {ctrl_down}");
}
```
In debug mode, this program prints the correct answer. In release mode,
it always prints false. The optimizer appears to think that
`keyboard_state` isn't mutated and remains zeroed, and folds the
`modifier_down` comparisons to `false`.
Release Notes:
- N/A
As discussed with @benbrandt and @mikayla-maki:
* We now tell ACP clients we support the nonstandard `terminal-auth`
`_meta` field for terminal-based authentication
* In the future, we anticipate ACP itself supporting *some* form of
terminal-based authentication, but that hasn't been designed yet or gone
through the RFD process
* For now, this unblocks terminal-based auth
Release Notes:
- Added experimental terminal-based authentication to ACP support
This has been possible in the inline assistant for ages now and maybe
you didn't know because we didn't say anything about it! This PR fixes
that by including that you can @-mention context on it the same you can
in the agent panel.
Release Notes:
- N/A
Closes #ISSUE
More accurately map localhost to `127.0.0.1`. Previously we would
lowercase and simply replace all instances of localhost inside of the
URL string, meaning query parameters, username, password etc. could not
contain the string `localhost` or contain uppercase letters without
getting modified. Added a test ensuring the mapping logic works. The
previous implementation would fail this new test.
Release Notes:
- Improved the behavior of mapping `localhost` to `127.0.0.1` when
passing configured proxy urls to `node`
### Summary
This PR changes `gpui/build.rs` to look up the Windows SDK directory in
the registry instead of falling back to a hard-coded path.
---
### Problem
Currently, building `gpui` on Windows requires `fxc.exe` to be in `PATH`
or at a predefined location (unless `GPUI_FXC_PATH` is set). This
requires to maintain a certain build environment with proper paths/vars
or to install the specific SDK version.
It is possible to find the SDK automatically using the registry keys it
creates upon installation. Specifically in
`SOFTWARE\\WOW6432Node\\Microsoft\\Microsoft SDKs\\Windows\\v10.0`
branch there are:
* `InstallationFolder` telling the SDK installation location;
* `ProductVersion` telling the SDK version in use.
These keys provide enough information to locate the SDK binaries, with
added robustness:
* handles non-standard SDK installation path;
* deterministically selects the latest SDK when multiple versions are
present.
---
### Changes Made
* **Updated `crates/gpui/build.rs`**:
* added dependency on `winreg`
* introduced `find_latest_windows_sdk_binary()` helper
* updated fallback logic to use registry lookup
This PR only changes the fallback location, and does not touch the
established environment-based workflow.
Release Notes:
- N/A
---
### Impact
Reduces manual configuration needed to build GPUI on Windows.
---------
Co-authored-by: John Tur <john-tur@outlook.com>
Closes#41457#41806#41801
Copilot started using `node:sqlite` module which is an experimental
feature between node v22-v23 (stable in v24). The fix was passing in the
experimental flag when Zed starts the copilot LSP.
I tested this with v20.19.5 and v24.11.0. The fix got v20.19 working and
didn't affect v24.11 which was already working.
Release Notes:
- AI: Fix Github Copilot edit predictions failing to start
Closes#41774
Release Notes:
- settings_ui: Fixed an integer underflow panic when attempting to hit
the `-` sign on settings item that take delays in milliseconds
`ICustomDestinationList::AppendCategory` rejects an empty `IObjectArray`
and returns an `E_INVALIDARG` error. Error propagated and caused an
early-return from `update_jump_list()`.
<img width="1628" height="540" alt="image"
src="https://github.com/user-attachments/assets/f8143297-c71e-42a1-a505-66cd77dfa599"
/>
Release Notes:
- N/A
- lsp: Fix dynamic registration of diagnostic capabilities not taking
effect when an initial capability is not specified
Gist of the issue lies within use of .get_mut instead of .entry. If we
had not created any dynamic capability beforehand, we'd miss a
registration, essentially
- **Determine whether to update remote caps in a smarter manner**
Release Notes:
- Fixed document diagnostics with Ty language server.
Closes#41533
Both of the issues in the release notes that are fixed in this PR, were
caused by incorrect usage of the `window.use_state` API.
The first issue was caused by calling `window.use_state` in a render
helper, resulting in the element ID used to share state being the same
across different pages, resulting in the state being re-used when it
should have been re-created. The fix for this was to move the
`window.state` (and rendering logic) into a `impl RenderOnce` component,
so that the IDs are resolved during the render, avoiding the state
conflicts.
The second issue is caused by using a `move` closure in the
`window.use_state` call, resulting in stale closure values when the
window state is re-used.
Release Notes:
- settings_ui: Fixed an issue where some dropdown menus would show
options from a different dropdown when clicked
- settings_ui: Fixed an issue where attempting to change a setting in a
dropdown back to it's original value after changing it would do nothing
This PR implements the `zeta-cli eval` command. It will:
- Run the edit prediction model if there are no cached results
- Compute precision/recall/F1 for context retrieval at the line level:
every retrieved line of context is counted as a true positive (correct
retrieval), false positive (retrieved something that was not expected),
or false negative (didn't retrieve an expected line)
- Compute similar metrics for edit predictions
- Pretty-print results, highlighting the difference between actual and
expected when printing to tty
Other changes:
- `zeta-cli predict` accepts a `--format` argument with options `md`,
`json`, `diff`
- Code restructure
Release Notes:
- N/A
---------
Co-authored-by: Piotr Osiewicz <24362066+osiewicz@users.noreply.github.com>
Co-authored-by: Agus Zubiaga <agus@zed.dev>
