Reverts #38175, which is not correct, since in fact we do need to
pre-quote the command and arguments for the shell when using
`SpawnInTerminal` (although we should probably change the API so that
this isn't necessary). Then, applies the same fix as #38565 to fix the
root cause of being unable to spawn the login task on macOS, or in any
case where the command/args contain spaces.
Release Notes:
- Fixed being unable to login with Claude Code or Gemini using the
terminal.
- Reflect that basedpyright is the new primary language server
- Discuss Ruff
- Deemphasize manual venv configuration for language servers
Release Notes:
- N/A
---------
Co-authored-by: Katie Geer <katie@zed.dev>
Co-authored-by: Piotr <piotr@zed.dev>
The built-up command for the WSL remote connection looks like
```
wsl.exe --distribution Ubuntu --user cole --cd /home/cole -- bash -c SCRIPT
```
Where `SCRIPT` is a command itself. We don't need extra quotes around
`SCRIPT` because we already pass it whole as a separate argument to
`wsl.exe`.
This isn't yet enough to get ACP servers working in WSL projects
(#38332), but it removes one roadblock.
Release Notes:
- windows: Fixed an issue that could prevent running binaries in WSL
remote projects.
Tests that the downstream project can see custom agents configured in
the remote server's settings, and that it constructs an appropriate
`AgentServerCommand`.
Release Notes:
- N/A
In local projects, initialize the list of agents in the agent server
store immediately. Previously we were initializing the list only after a
delay, in an attempt to avoid sending the `ExternalAgentsUpdated`
message to the downstream client (if any) before its handlers were
initialized. But we already have a separate codepath for that situation,
in the `AgentServerStore::shared`, and we can insert the delay in that
place instead.
Release Notes:
- acp: Fixed a bug where starting an external agent thread soon after
Zed starts up would show a "not registered" error.
---------
Co-authored-by: Michael <michael@zed.dev>
Co-authored-by: Agus <agus@zed.dev>
This can lead to an infinite regress when using a language server that
supports pull diagnostics, since the excerpts for the diagnostics editor
are set based on the project's diagnostics.
Closes#36772
Release Notes:
- Fixed a bug that could cause duplicated diagnostics with some language
servers.
Release Notes:
- python: The Ruff native language server is now available without
installing an extension.
---------
Co-authored-by: Piotr Osiewicz <24362066+osiewicz@users.noreply.github.com>
Follow-up to #35250. Let's experiment with having this by default on
nightly.
Release Notes:
- Added built-in support for the basedpyright language server for Python
code. basedpyright is now enabled by default, and pyright (previously
the primary Python language server) remains available but is disabled by
default. This supersedes the basedpyright extension, which can be
uninstalled. Advantages of basedpyright over pyright include support for
inlay hints, semantic highlighting, auto-import code actions, and
stricter type checking. To switch back to pyright, add the following
configuration to settings.json:
```json
{
"languages": {
"Python": {
"language_servers": ["pyright", "pylsp", "!basedpyright"]
}
}
}
```
---------
Co-authored-by: Piotr Osiewicz <24362066+osiewicz@users.noreply.github.com>
Co-authored-by: Lukas Wirth <lukas@zed.dev>
This PR adds support for using external agents in SSH projects via ACP,
including automatic installation of Gemini CLI and Claude Code,
authentication with API keys (for Gemini) and CLI login, and custom
agents from user configuration.
Co-authored-by: maan2003 <manmeetmann2003@gmail.com>
Release Notes:
- agent: Gemini CLI, Claude Code, and custom external agents can now be
used in SSH projects.
---------
Co-authored-by: maan2003 <manmeetmann2003@gmail.com>
For read_file and edit_file, show the worktree-relative path if there's
only one visible worktree, and the "full path" otherwise. Also restores
the display of line numbers for read_file calls.
Release Notes:
- N/A
Closes#37469
Release Notes:
- agent: The project shell environment is now passed to external agent
processes.
Co-authored-by: Richard Feldman <oss@rtfeldman.com>
Co-authored-by: Nia Espera <nia-e@haecceity.cc>
Related to #37213, #37150
When listing previously-downloaded versions of an external agent, don't
try to use any downloads that are missing the agent entrypoint
(indicating that they're corrupt/unusable), and delete those versions,
so that we can attempt to download the latest version again.
Also report clearer errors when failing to start a session due to an
agent server entrypoint or root directory not existing.
Release Notes:
- N/A
When the first visible worktree is a single-file worktree, we would
previously try to use the absolute path of that file as the root
directory for external agents, causing an error. This PR changes how we
handle this situation: we'll use the root of the first non-single-file
visible worktree if there are any, and if there are none, the parent
directory of the first single-file visible worktree.
Related to #37213
Release Notes:
- acp: Fixed being unable to run external agents when a single file (not
part of a project) was opened in Zed.
We'll just SSH into the Windows runners and look for crashes there.
Reverts #35926
Release Notes:
- N/A
---------
Co-authored-by: Peter Tripp <petertripp@gmail.com>
We render terminals as inline if their content is below a certain line
count, and scrollable past that point. In the scrollable case we weren't
setting a height for the terminal's container, causing it to be rendered
at height 0, which means no lines would be displayed. This PR fixes that
by setting an explicit height for the scrollable case, like we do in the
agent1 UI code.
Release Notes:
- agent: Fixed a bug that caused terminals in the panel to be empty
after their content reached a certain size.
Closes#37171
Release Notes:
- agent: Fixed a bug that caused the agent information in the panel
header to be incorrect when opening a thread from history.
In the case where we fail to create an ACP connection to Gemini, only
report the "unsupported version" error if the version for the found
binary is at least our minimum version. That means we'll surface the
real error in this situation.
This also fixes incorrect sorting of downloaded Gemini versions--as @kpe
pointed out we were effectively using the version string as a key. Now
we'll correctly use the parsed semver::Version instead.
Release Notes:
- N/A
Closes: https://github.com/zed-industries/zed/issues/37089
Instead of looking for the gemini command on `$PATH`, by default we'll
install our own copy on demand under our data dir, as we already do for
language servers and debug adapters. This also means we can handle
keeping the binary up to date instead of prompting the user to upgrade.
Notes:
- The download is only triggered if you open a new Gemini thread
- Custom commands from `agent_servers.gemini` in settings are respected
as before
- A new `agent_servers.gemini.ignore_system_version` setting is added,
similar to the existing settings for language servers. It's `true` by
default, and setting it to `false` disables the automatic download and
makes Zed search `$PATH` as before.
- If `agent_servers.gemini.ignore_system_version` is `false` and no
binary is found on `$PATH`, we'll fall back to automatic installation.
If it's `false` and a binary is found, but the version is older than
v0.2.1, we'll show an error.
Release Notes:
- acp: By default, Zed will now download and use a private copy of the
Gemini CLI binary, instead of searching your `$PATH`. To make Zed search
your `$PATH` for Gemini CLI before attempting to download it, use the
following setting:
```
{
"agent_servers": {
"gemini": {
"ignore_system_version": false
}
}
}
```
Release Notes:
- acp: Fixed a bug that caused external agent server subprocesses to be
leaked.
---------
Co-authored-by: Agus Zubiaga <agus@zed.dev>
Co-authored-by: Bennet Bo Fenner <bennetbo@gmx.de>
Co-authored-by: Antonio Scandurra <me@as-cii.com>
Adds support for per-session prompt capabilities and capability changes
on the Zed side (ACP itself still only has per-connection static
capabilities for now), and uses it to reflect image support accurately
in 1PA threads based on the currently-selected model.
Release Notes:
- N/A
We were rendering a Markdown link like `[Read file x.rs (lines
Y-Z)](@selection)` while the tool ran, but then switching to just `x.rs`
as soon as we got the file location from the tool call (due to an
if/else in the UI code that applies to all tools). This caused a
flicker, which is fixed by having `initial_title` return just the
filename from the input as it arrives instead of a link that we're going
to stop rendering almost immediately anyway.
Release Notes:
- N/A
Merge conflict resolution for #36741 accidentally reverted the changes
in #36670 to allow expanding terminals individually and in #36675 to
allow collapsing edit cards. This PR re-applies those changes, fixing
the regression.
Release Notes:
- N/A
- Add pulsating animation for context creases while they're loading
- Add spinner in message editors (replacing send button) during the
window where sending has been requested, but we haven't finished loading
the message contents to send to the model
- During the same window, ignore further send requests, so we don't end
up sending the same message twice if you mash enter while loading is in
progress
- Wait for context to load before rewinding the thread when sending an
edited past message, avoiding an empty-looking state during the same
window
Release Notes:
- N/A
This PR makes it so that all kinds of @-mentions start loading their
context as soon as they are confirmed. Previously, we were waiting to
load the context for file, symbol, selection, and rule mentions until
the user's message was sent. By kicking off loading immediately for
these kinds of context, we can support adding selections from unsaved
buffers, and we make the semantics of @-mentions more consistent.
Loading all kinds of context eagerly also makes it possible to simplify
the structure of the MentionSet and the code around it. Now MentionSet
is just a single hash map, all the management of creases happens in a
uniform way in `MessageEditor::confirm_completion`, and the helper
methods for loading different kinds of context are much more focused and
orthogonal.
Release Notes:
- N/A
---------
Co-authored-by: Conrad <conrad@zed.dev>
We had a frequent panic when the agent was using our edit file tool. The
root cause was that we were constructing a `BufferDiff` with
`BufferDiff::new`, then calling `set_base_text`, but not waiting for
that asynchronous operation to finish. This means there was a window of
time where the diff's base text was set to the initial value of
`""`--that's not a problem in itself, but it was possible for us to call
`PendingDiff::update` during that window, which calls
`BufferDiff::update_diff`, which calls
`BufferDiffSnapshot::new_with_base_buffer`, which takes two arguments
`base_text` and `base_text_snapshot` that are supposed to represent the
same text. We were getting the first of those arguments from the
`base_text` field of `PendingDiff`, which is set immediately to the
target base text without waiting for `BufferDiff::set_base_text` to run
to completion; and the second from the `BufferDiff` itself, which still
has the empty base text during that window.
As a result of that mismatch, we could end up adding `DeletedHunk` diff
transforms to the multibuffer for the diff card even though the
multibuffer's base text was empty, ultimately leading to a panic very
far away in rendering code.
I've fixed this by adding a new `BufferDiff` constructor for the case
where the buffer contents and the base text are (initially) the same,
like for the diff cards, and so we don't need an async diff calculation.
I also added a debug assertion to catch the basic issue here earlier,
when `BufferDiffSnapshot::new_with_base_buffer` is called with two base
texts that don't match.
Release Notes:
- N/A
---------
Co-authored-by: Conrad <conrad@zed.dev>
This may be responsible for a panic that we've been seeing with
increased frequency in agent2 threads.
Release Notes:
- N/A
Co-authored-by: Conrad Irwin <conrad.irwin@gmail.com>
- Render a helpful message when the installed CC version is too old
- Show the full path for agent binaries when the version is not recent
enough (helps in cases where multiple binaries are installed in
different places)
- Add UI for the case where a server binary is not installed at all
- Refresh thread view after installing/updating server binary
Release Notes:
- N/A
- Clean up failed load tasks for mentions that require async processing
- When dragging and dropping files, hold onto added worktrees until any
async processing has completed; this fixes a bug when dragging items
from outside the project
Release Notes:
- N/A
Highlight them as errors in the editor, and add a leading space when
sending them so users don't hit the odd behavior when sending these
commands to the SDK.
Release Notes:
- N/A
Note that (at least for now) this only works for creases in the "new
message" editor, not when editing past messages. That's because we don't
have the original image available when putting together the creases for
past messages, only the base64-encoded language model content.
Release Notes:
- N/A
- Support adding ImageContent to messages through copy/paste and through
path completions
- Ensure images are fully converted to LanguageModelImageContent before
sending them to the model
- Update ACP crate to v0.0.24 to enable passing image paths through the
protocol
Release Notes:
- N/A
---------
Co-authored-by: Conrad Irwin <conrad.irwin@gmail.com>
This is a bit simpler than for the original agent thread view, since we
don't have to deal with opening buffers or a context store.
Release Notes:
- N/A
Currently, when the agent updates its plan, there are a few frames where
the text after `Current:` in the plan summary is blank, causing a
flicker. This is because we treat that field as markdown, and the
`MarkdownElement` renders as blank until the raw text has finished
parsing in the background.
This PR fixes the flicker by changing `Markdown::new_text` to
optimistically render the source as a single `MarkdownEvent::Text` span
until background parsing has finished.
Release Notes:
- N/A
Co-authored-by: Agus Zubiaga <agus@zed.dev>
This PR adds handling for the case where an agent binary exits
unexpectedly after successfully establishing a connection.
Release Notes:
- N/A
---------
Co-authored-by: Agus <agus@zed.dev>
This PR moves the new agent thread controls so they're attached to the
last message and scroll with the thread history, instead of always being
shown above the message editor.
Release Notes:
- N/A
This PR adds handling for the case where the user's gemini-cli binary
fails to start up because it's too old to support the
`--experimental-acp` flag. We previously had such handling, but it got
lost as part of #35578.
This doesn't yet handle the case where the server binary exits
unexpectedly after the connection is established; that'll be dealt with
in a follow-up PR since it needs different handling and isn't specific
to gemini-cli.
Release Notes:
- N/A
Co-authored-by: Agus <agus@zed.dev>
The spec says:
> ⬅️ Initialized Event
> This event indicates that the debug adapter is ready to accept
configuration requests (e.g. setBreakpoints, setExceptionBreakpoints).
>
> A debug adapter is expected to send this event when it is ready to
accept configuration requests (but not before the initialize request has
finished).
Previously in tests, `intercept_debug_sessions` was just spawning off a
background task to send the event after setting up the client, so the
event wasn't actually synchronized with the flow of messages in the way
the spec says it should be. This PR makes it so that the `FakeTransport`
injects the event right after a successful response to the initialize
request, and doesn't send it otherwise.
Release Notes:
- N/A
This PR adds a built-in adapter for the basedpyright language server.
For now, it's behind the `basedpyright` feature flag, and needs to be
requested explicitly like this for staff:
```
"languages": {
"Python": {
"language_servers": ["basedpyright", "!pylsp", "!pyright"]
}
}
```
(After uninstalling the basedpyright extension.)
Release Notes:
- N/A
Follow-up to #26114
- Ensure that the previous commit message is filled in when toggling on
amend mode from the context menu
- Fix keybinding flicker in context menu
Release Notes:
- N/A
Should catch cases like #33667 where a dependency was bumped in
Cargo.toml without a corresponding lockfile update. (This can happen
when committing from outside of Zed or if rust-analyzer isn't running.)
Passing `--frozen --workspace` should ensure this doesn't fail just
because there's a newer patch version of some third-party dependency,
and prevent it from taking long.
We only need to run this on macOS because Cargo.lock is
platform-independent.
Release Notes:
- N/A
Previously we would append `js-debug/src/dapDebugServer.js` to the value
of the `dap.JavaScript.binary` setting and `src/debugpy/adapter` to the
value of the `dap.Debugpy.binary` setting, which isn't particularly
intuitive. This PR fixes that.
Release Notes:
- debugger: Made the semantics of the `dap.$ADAPTER.binary` setting more
intuitive for the `JavaScript` and `Debugpy` adapters. In the new
semantics, this should be the path to `dapDebugServer.js` for
`JavaScript` and the path to the `src/debugpy/adapter` directory for
`Debugpy`.
---------
Co-authored-by: Remco Smits <djsmits12@gmail.com>
Right now it doesn't work at all (the PID doesn't get set in the
generated scenario), and it's sort of redundant with the picker
functionality.
Release Notes:
- N/A
Closes#33945. Here's my attempt to describe what's going on in that
issue and what this fix is doing:
We always render the terminal inline assistant starting on the line
after the cursor, with a height of 4 lines. When deploying it, we scroll
the viewport to the bottom of the terminal so that the assistant will be
in view.
When scrolling while the assistant is deployed (including in that case),
we need to make an adjustment that "pushes up" the terminal content by
the height of the assistant, so that we can scroll to see all the normal
content plus the assistant itself. That quantity is `scroll_top`, which
represents _how much height in the current viewport is occupied by the
assistant that would otherwise be occupied by terminal content_. So when
you scroll up and a line of the assistant's height goes out of view,
`scroll_top` decreases by 1, etc.
When we scroll to the bottom after deploying the assistant, we set
`scroll_top` to the result of calling `max_scroll_top`, which computes
it this way:
```
block.height.saturating_sub(viewport_lines.saturating_sub(terminal_lines))
```
Which, being interpreted, is "the height of the assistant, minus any
viewport lines that are not occupied by terminal content", i.e. the
assistant is allowed to eat up vertical space below the last line of
terminal content without increasing `scroll_top`.
The problem comes when we clear the screen---this adds a full screen to
`terminal_lines`, but the cursor is positioned at the top of the
viewport with blank lines below, just like at the beginning of a session
when `terminal_lines == 1`. Those blank lines should be available to the
assistant, but the `scroll_top` calculation doesn't reflect that.
I've tried to fix this by basing the `max_scroll_top` calculation on the
position of the cursor instead of the raw `terminal_lines` value. There
was also a special case for `viewport_lines == terminal_lines` that I
think can now be removed.
Release Notes:
- Fixed the positioning of the terminal inline assistant when it's
deployed after clearing the terminal.
Closes#34044
`EditorMode::SingleLine { auto_width: true }` was only used for the
title editor in the rules library, and following
https://github.com/zed-industries/zed/pull/31994 we can replace that
with a normal single-line editor without problems. The auto-width editor
was interacting badly with the recently-added newline visualization
code, causing a panic during layout---by switching it to
`Editor::single_line` the newline visualization works there too.
Release Notes:
- Fixed a panic that could occur when opening the rules library.
---------
Co-authored-by: Finn <finn@zed.dev>
This PR updates the debugger panel's session list to be more useful in
some cases that are commonly hit when using the JavaScript adapter. We
make two adjustments, which only apply to JavaScript sessions:
- For a child session that's the only child of a root session, we
collapse it with its parent. This imitates what VS Code does in the
"call stack" view for JavaScript sessions.
- When a session has exactly one thread, we label the session with that
thread's name, instead of the session label provided by the DAP. VS Code
also makes this adjustment, which surfaces more useful information when
working with browser sessions.
Closes#33072
Release Notes:
- debugger: Improved the appearance of JavaScript sessions in the debug
panel's session list.
---------
Co-authored-by: Julia <julia@zed.dev>
Co-authored-by: Remco Smits <djsmits12@gmail.com>
Closes#34323Closes#34313
The previous PR #33932 introduced a way to "close" the
`pending_requests` buffer of the `TransportDelegate`, preventing any
more requests from being added. This prevents pending requests from
accumulating without ever being drained during the shutdown sequence;
without it, some of our tests hang at this point (due to using a
single-threaded executor).
The bug occurred because we were closing `pending_requests` whenever we
detected the server side of the transport shut down, and this closed
state stuck around and interfered with the retry logic for SSH+TCP
adapter connections.
This PR fixes the bug by only closing `pending_requests` on session
shutdown, and adds a regression test covering the SSH retry logic.
Release Notes:
- debugger: Fixed a bug causing SSH connections to some adapters
(Python, Go, JavaScript) to fail and restart endlessly.
Closes#34029
The crash is due to a stack overflow in our `html_to_markdown`
conversion; I've added a maximum depth of 200 for the recursion in that
crate to guard against this kind of thing.
Separately, we were treating all content-types other than `text/plain`
and `application/json` as HTML; I've changed this to only treat
`text/html` and `application/xhtml+xml` as HTML, and fall back to
plaintext. (In the original crash, the content-type was
`application/octet-stream`.)
Release Notes:
- agent: Fixed a potential crash when fetching large non-HTML files.
Restarting sessions was broken in #33273 when we moved away from calling
`kill` in the shutdown sequence. This PR re-adds that `kill` call so
that old debug adapter processes will be cleaned up when sessions are
restarted within Zed. This doesn't re-introduce the issue that motivated
the original changes to the shutdown sequence, because we still send
Disconnect/Terminate to debug adapters when quitting Zed without killing
the process directly.
We also now remove manually-restarted sessions eagerly from the session
list.
Closes#33916
Release Notes:
- debugger: Fixed not being able to restart sessions for Debugpy and
other adapters that communicate over TCP.
- debugger: Fixed debug adapter processes not being cleaned up.
---------
Co-authored-by: Remco Smits <djsmits12@gmail.com>
Closes#33951
There's an adjustment that kicks in to extend `name_ranges` when we
capture more than one `@name` for an outline `@item`. That was happening
here because we captured both the parameter name for the method receiver
and the name of the method as `@name`. It seems like only the second one
should have that annotation.
Release Notes:
- Fixed extraneous leading space in `$ZED_SYMBOL` when used with Go
methods.
`program` isn't required, and in fact our built-in `JavaScript debug
terminal` configuration doesn't have it.
Also add `node-terminal` to the list of allowed types.
Co-authored-by: Michael <michael@zed.dev>
Release Notes:
- N/A
Plain `f11` is a system keybinding. We already use `ctrl-f11` for this
on Linux.
Release Notes:
- debugger: Switched the macOS keybinding for `debugger::StepInto` from
`f11` to `ctrl-f11`.
Related to #33072
We use the JS adapter's suggested names for child sessions, but
sometimes it sends us `""`, so don't use that one.
Release Notes:
- debugger: Fixed nameless child sessions appearing with the JavaScript
adapter.
`M-q` is `fill-paragraph` which is like `editor::Rewrap`.
Release Notes:
- emacs: Bound `alt-q` to `editor::Rewrap` (like `M-q` or `M-x
fill-paragraph`)
Cole Miller
