Release Notes:
- Swapped the default value for `agent.single_file_review` to `false`.
Agent diffs will no longer override the git diff in your buffer. You can
still review the agent's changes via the action log review button, or by
flipping this setting back to `true`
Closes AI-10
The thinking tool is no longer needed. This removes it from the agent
codebase and updates all tests to use other tools instead.
Release Notes:
- Removed the thinking tool from the Agent, as models now have their own
first-class thinking APIs.
Closes#48857
- [x] Code Reviewed
- [x] Manual QA
Release Notes:
- Fixed an issue where some ACP agents would not be loading correctly
when unauthenticated
---------
Co-authored-by: cameron <cameron.studdstreet@gmail.com>
This [change](https://github.com/agentclientprotocol/registry/pull/34)
broke the ACP registry in Zed, since we did not have `#[serde(default)]`
on the `_extensions` field.
- [x] Tests or screenshots needed?
- [x] Code Reviewed
- [x] Manual QA
Release Notes:
- Fixed an issue where the ACP registry would fail to load
Closes#48274
Previously, the Copilot language server would continue running even when
`disable_ai: true` was set in settings. This change ensures Copilot
properly responds to the `disable_ai` setting:
- Add `disable_ai` check in `start_copilot()` to prevent starting when
AI is disabled
- Modify the `SettingsStore` observer to shut down the running language
server when `disable_ai` changes from false to true
- Add tests for all scenarios:
- Copilot doesn't start when `disable_ai` is true
- Copilot stops when `disable_ai` becomes true
- Copilot can start again when `disable_ai` becomes false
Release Notes:
- Fixed Copilot starting when disabled_ai: true
`ChunkSlice::point_utf16_to_offset` returned `line.len()` (a relative
line byte length) instead of `row_offset_range.end` (the absolute byte
offset within the chunk) when clipping an out-of-range UTF-16 column on
a non-first row.
This caused incorrect position calculations in LSP coordinate
conversions whenever a UTF-16 point had a column beyond the line's
extent on any row after the first in a chunk.
Added a regression test covering clipping on both row 0 and row 1. Not
sure if it's the best place for a test like this one though.
Release Notes:
- N/A
Fixes a regression that was introduced by the thread view refactor
(Nightly only).
Also fixed another minor focus issue: Dropping files into the agent
panel would not focus the message editor.
- [x] Code Reviewed
- [x] Manual QA
Release Notes:
- N/A
I have added the `code_actions_on_format` setting. I also temporarily
removed the mention of `"formatter": null`. Based on our findings in
#48600, its current behavior is inconsistent, so I've omitted it for now
to avoid confusing users. I will add it back once the underlying issue
is fixed (likely in a subsequent PR).
Closes#48600.
- [ ] Tests or screenshots needed?
- [x] Code Reviewed
- [x] Manual QA
Release Notes:
- N/A
---------
Co-authored-by: Kunall Banerjee <hey@kimchiii.space>
`read_fd` calls `File::from_raw_fd(fd.as_raw_fd())`, which creates a new
`File` that assumes ownership of the raw fd, while the original
`FileDescriptor` also retains ownership. when both drop at the end of
the function, the same fd is closed twice. between the two closes
another thread can receive that fd number from the kernel for an
unrelated file/socket, and the second close silently kills it. this can
cause hard to diagnose data corruption or i/o failures.
fix: use `fd.into_raw_fd()` instead of `fd.as_raw_fd()` to transfer
ownership to the `File` without running `FileDescriptor`'s drop.
Release Notes:
- N/A
We try to remove old excerpt IDs from the companion when mutating
excerpts for a path, but we were incorrectly doing this using the
excerpt IDs for the right-hand side multibuffer _after_ any excerpt
mutations had occurred, causing stale excerpt IDs to accumulate.
Release Notes:
- N/A
- Fix panics caused by reusing cached matches for the wrong side
- Highlight matches on the side that was searched only
- Clear matches in non-searched editor when initiating a new search
Release Notes:
- N/A
---------
Co-authored-by: Eric <eric@zed.dev>
Co-authored-by: Jakub <jakub@zed.dev>
Although this introduces a small possiblity that we will regress the nix
build,
the existing check is very slow and fails very rarely.
Release Notes:
- N/A
Closes#48683
Per the devcontainer spec, `.devcontainer.json` in the project root is
a valid config location. It is only used when no configurations are
found inside `.devcontainer/`.
Extract `find_configs_in_snapshot` for testability and add tests.
Release Notes:
- Added support for `.devcontainer.json` in project root
Currently, whenever Zed has a new version available for download, the
update button shows up in the status bar. Problem is: that status bar
slot can also display other buttons/information, such as problems with
your language server or general errors. In case the two things exist (a
problem and a new version), just one of them would be displayed, which
is not great; you should be able to see both. Additionally, given we
ship new versions pretty often, I've frequently saw feedback about
wanting to hide away the new version button... at least temporarily
while there's no immediate interest in upgrading.
So, this PR tackles all of that. The button to update a new version is
moved up to the title bar, nearby your avatar, and you have the ability
to dismiss, which effectively just moves the button from the title bar
to inside your user menu.
https://github.com/user-attachments/assets/e3f1d76d-9b85-4bee-a70f-e22dd5e7fdb3
Release Notes:
- Moved the update Zed button to the title bar and allowed it to be
dismissed.
---------
Co-authored-by: Conrad Irwin <conrad.irwin@gmail.com>
Various improvements to settings UI for granular tool permissions,
extracted from a larger change:
- Add **copy_path** tool to the permissions configuration list
- Replace local `RuleType` enum with `ToolPermissionMode` from the
settings crate
- Improve rule summary display: singular "1 rule" instead of always "1
rules", show invalid pattern count
- Add **terminal command parsing preview** — uses `extract_commands` to
split chained commands (`&&`, `||`, `;`) so the preview matches the real
permission engine's behavior
- Add **verdict label** showing the authoritative permission decision
(Allow/Deny/Confirm)
- Log disagreements between the pattern preview and the engine's
authoritative verdict
- Display **invalid regex patterns section** with compilation error
details
- Add regex validation error banner (dismissible) on tool config pages
- Add **compile-time validated** `tool_index` for macro-generated render
functions (fails to compile if a tool ID is misspelled)
- Update move_path/copy_path regex explanations for multi-input tools
- Add test validating all tools have `ToolInfo` entries or are in the
exclusion list
- Add `clear_on_confirm()` — clears the input after confirming (used for
"add pattern" inputs)
- Add editor reconciliation fix — when settings change externally (e.g.
editing settings.json), syncs the cached editor text when not focused
- Rename "Configure Tool Rules" → "Tool Permissions" in the settings
sidebar
- Update description text
- Add `regex_validation_error` field to `SettingsWindow` (cleared on
page navigation)
(No release notes because granular tool permissions are behind a feature
flag.)
Release Notes:
- N/A
---------
Co-authored-by: Danilo Leal <daniloleal09@gmail.com>
Helps #48601
<img width="1649" height="1071" alt="image"
src="https://github.com/user-attachments/assets/ff3dfee0-cc65-430f-a5fa-b4b4c36e8183"
/>
`syntax_layers` does some offset conversion that might require getting
some chunks from the rope, which is quite expensive. For detecting
autoclose, we only use the language from those syntax layers, so having
a short path that skips all the conversion should skip some sum_tree
traversals.
I'm pretty sure other places would benefit from this as well, but I
haven't searched them yet.
Release Notes:
- N/A
---------
Co-authored-by: Smit Barmase <heysmitbarmase@gmail.com>
Follow-up to #47420 and #44193.
#47420 fixed child-exit status handling via upstream alacritty change
(https://github.com/alacritty/alacritty/pull/8825). However,
stable/preview/nightly builds still reproduced hangs that dev builds did
not.
The root cause matches #44193: spawned children inherit crash-handler
Mach exception ports and hang. We already reset exception ports in
`util::command::new_smol_command` and
`util::set_pre_exec_to_start_new_session`, but PTY child spawning in
`alacritty_terminal` was not covered.
This PR updates `alacritty_terminal` to include PTY `pre_exec`
exception-port reset:
https://github.com/zed-industries/alacritty/commit/9d9640d4e56d67a09d049f9c0a300aae08d4f61e
Upstream:
- https://github.com/alacritty/alacritty/pull/8835
Release Notes:
- Fixed terminal tasks hanging on macOS when a spawned process is killed
by a signal.
Add `workspace::CloseItemInAllPanes` action that closes the active
item's buffer in every pane where it's open, matching Vim's `:bdelete`
semantics. Pane layout is preserved, only the buffer is removed.
`:bd` respects pinned tabs, `:bd!` overrides them and skips save.
Also refactors the tab switcher's close button to use the new
`close_items_with_project_path` method, removing duplicated logic.
Release Notes:
- Vim: `:bd` (`:bdelete`) now closes the file in all panes where it's
open
- Added `workspace::CloseItemInAllPanes` action to close a file across
all panes
---------
Co-authored-by: dino <dinojoaocosta@gmail.com>
Reverts zed-industries/zed#48607
The problem we've ran into is that CI started reusing already-cached
runs from newer branches (that did not have that change pulled in).
Yet this unveiled another potential issue, which is that relying on
mtime might make it so that an artifact for a different (newer) commit
would be reused by an unrelated run of CI.
The answer is: checksum-based-freshness.
Release Notes:
- N/A
Otherwise, it's not easily possible to set *any* of these variables when
running Zed.
Release Notes:
- Added forwarding of `ZED_*` environment variables when using the
Flatpak
Conrad Irwin