Related issue: https://github.com/zed-industries/zed/issues/20167
Release Notes:
- Changed the default keybinding to accept partial inline completions
from `ctrl-right` to `ctrl-cmd-right` on macOS, because `ctrl-right` is
already bound to jump to the end of the line.
Co-authored-by: Antonio <antonio@zed.dev>
Co-authored-by: Kirill <kirill@zed.dev>
Co-authored-by: Bennet <bennet@zed.dev>
This fixes an error message that has shown up for me when joining collab
projects: "Unable to load shell environment in /<path on another
machine/"
Release Notes:
- Fixed error message about shell environment failing to load when
joining projects in collaboration.
Various fixes for Zeta and one fix that's visible to non-Zeta-using
users of inline completions.
Release Notes:
- Changed inline completions (Copilot, Supermaven, ...) to not show up
in empty buffers.
---------
Co-authored-by: Antonio <antonio@zed.dev>
Co-authored-by: Antonio Scandurra <me@as-cii.com>
Co-authored-by: Bennet <bennet@zed.dev>
As @hferreiro points out in [this
comment](https://github.com/zed-industries/zed/pull/18752#issuecomment-2589340565):
C++ and prettier don't work well together, so let's make the default
formatter for C++ the primary language server. We get that by disabling
prettier.
Release Notes:
- Changed default formatter for C++ to be the primary language server,
not Prettier. Format-on-save is still disabled by default for C++, but
if one uses the `editor: format` command now, it will default to the
language server. `clangd` can format C++ files, whereas prettier cannot.
This changes the behaviour of `<tab>` when inline completion is visible.
When the cursor is before the suggested indentation level, accepting a
completion should just indent.
cc @nathansobo @maxdeviant
Release Notes:
- Changed the behavior of `<tab>` at start of line when an inline
completion (Copilot, Supermaven, ...) is visible. If the cursor is
before the suggested indentation, `<tab>` now indents the line instead
of accepting the visible completion.
Co-authored-by: Antonio <antonio@zed.dev>
This fixes the issue described in this comment:
https://github.com/zed-industries/zed/pull/22439#issuecomment-2563896422
Essentially, we'd clip in the wrong direction when there were multi-line
inlay hints.
It also fixes inline completions for non-Zeta-providers showing up in
normal mode.
Release Notes:
- N/A
This is a follow-up to #22615 and fixes the issue of `alacritty`
resulting in broken shell/CLI apps if `alacritty` is not in the terminfo
database.
Closes #ISSUE
Release Notes:
- Set `TERM` to `xterm-256color` in Zed's built-in terminal
This message has confused me many times too: we printed the status as
"Downloading" when we were only checking whether we need to install a
given version of a language server.
This fixes the issue for Node-based language servers where we had the
same check in all implementations.
Closes #22241
Release Notes:
- Fixed some language servers reporting status as "Downloading..." when
only a version check was being done.
Fixes#22151.
Turns out Vim also has some weird behavior with sentence deletion in
case it's on the first character of a line.
Release Notes:
- vim: Fixed deleting sentence-wise (i.e. `d(` and `d)`), which would
previously delete the whole line instead of just a sentence.
This doesn't include the outline in the prompt yet, but it does send it
up via telemetry so we can use it to see whether it would have improved
generated output.
Release Notes:
- N/A
Co-authored-by: Agus <agus@zed.dev>
Fixes https://github.com/zed-industries/zed/issues/21972
This fixes two bugs:
**Bug 1**: this bug caused us to only ever load a single environment in
a multi-worktree project, thanks to this line:
```rust
if let Some(task) = self.get_environment_task.as_ref()
```
We'd only ever run a single task per project, which is wrong.
What does code does is to cache the tasks per `worktree_id`, which means
we don't even need to cache the environments again, since we can just
cache the `Shared<Task<...>>`.
**Bug 2**: we assumed that every `worktree_abs_path` is a directory,
which lead to `Failed to run direnv` log messages when opening a project
that had a worktree with a single file open (easy to reproduce: open a
normal project, open your settings, close Zed, reopen it — the settings
faile caused environments to not load)
It's fixed by checking whether the `worktree_abs_path` is an absolute
directory. Since this is always running locally, it's fine to use
`smol::fs` here instead of using our `Fs`.
Release Notes:
- Fixed shell environments not being loaded properly to be used by
language servers and terminals in case a project had multiple worktrees.
- Fixed `Failed to run direnv` messages showing up in case Zed restored
a window that contained a worktree with a single file.
https://github.com/zed-industries/zed/issues/21972
If an inline completion isn't shown in a menu, we highlight text in the
editor as deleted.
But if it's shown in the menu, we didn't even show deleted text, which
makes it hard to understand what's going on.
This fixes it.
Release Notes:
- N/A
This changes #22093 by making the change only have an effect for the
supported provider: Zeta.
Made the change because the UX is still experimental and I don't want to
break existing workflows for Copilot/Supermaven users.
Even Zeta users can opt-out of it by setting
`"show_inline_completions_in_menu": false` in their settings, in case
they want the old show-inline-completion-or-show-lsp-completion behavior
back.
Release Notes:
- N/A
Possible panic here in case we can't find the excerpt in the
multibuffer.
I thought this could happen when the inline completion references an
excerpt that disappeared from the multibuffer, but looking at the code
I'm not sure anymore - we use the same multibuffer snapshot in this
whole function and the `anchor_in_excerpt` method clips the anchors.
Still, let's be safe here.
Release Notes:
- N/A
This makes inline completions show up in the completion menu even if the
user has set `"show_completion_documentation": false` in their settings,
because there is no other way to show the Zeta completion.
Follow-up to #22093
Release Notes:
- N/A
Co-authored-by: Danilo <danilo@zed.dev>
This is related to #22069 and #21858: before both of these PRs, we would
only ever show inline completions OR completions, never both at the same
time.
Now we show both at the same, but we still had this piece of logic here,
that prevented non-inline completions from showing up if there was
already an inline completion.
With this change, it's possible to get LSP completions without having to
dismiss inline completions before.
Release Notes:
- Inline completions (Copilot, Supermaven, ...) don't stop other
completions from showing up anymore. Both can now be visible at the same
time.
---------
Co-authored-by: Bennet <bennet@zed.dev>
I'm not sure about this yet.
On one hand: it's nice that the completion doesn't just disappear when I
hit escape because I was typing and in the flow.
On the other hand: no other inline completion provider keeps the
suggestion when leaving insert mode.
I'm going to merge this so we can get it into nightly and try it out for
the next couple of days. cc @ConradIrwin
Release Notes:
- vim: Do not dismiss inline completions when leaving insert/replace
mode with `<esc>`.
This reverts part of #21858 by changing how `tab` works again:
- If both, completions and inline completions, are visible, then `tab`
accepts the completion and `shif-tab` the inline completion.
- If only one of them is shown, then `tab` accepts it.
I'm not a fan of this solution, but I think it's a short-term fix that
avoids breaking people's `tab` muscle memory.
Release Notes:
- (These release notes invalidate the release notes contained in:
https://github.com/zed-industries/zed/pull/21858)
- Changed how inline completions (Copilot, Supermaven, ...) and normal
completions (from language servers) interact. Zed will now also show
inline completions when the completion menu is visible. The user can
accept the inline completion with `<shift-tab>` and the active entry in
the completion menu with `<tab>`.
This changes the behavior of how we display inline completions and
non-inline completions (i.e. completion menu).
Previously we would never show inline completions if a completion menu
was visible, meaning that we'd never show Copilot/Supermaven/...
suggestions if the language server had a suggestion.
With this change, we now display the inline completions even if there is
a completion menu visible.
In that case `<tab>` then accepts the inline completion and `<enter>`
accepts the selected entry in the completion menu.
Release Notes:
- Changed how inline completions (Copilot, Supermaven, ...) and normal
completions (from language servers) interact. Zed will now also show
inline completions when the completion menu is visible. The user can
accept the inline completion with `<tab>` and the active entry in the
completion menu with `<enter>`. Previously, `<tab>` would also select
the active entry in the completion menu.
---------
Co-authored-by: Antonio <antonio@zed.dev>
We also need to check whether the selection is empty, not just whether
its head is on an empty line.
Release Notes:
- N/A
Co-authored-by: Antonio <antonio@zed.dev>
This removes the `low_speed_timeout` setting from all providers as a
response to issue #19509.
Reason being that the original `low_speed_timeout` was only as part of
#9913 because users wanted to _get rid of timeouts_. They wanted to bump
the default timeout from 5sec to a lot more.
Then, in the meantime, the meaning of `low_speed_timeout` changed in
#19055 and was changed to a normal `timeout`, which is a different thing
and breaks slower LLMs that don't reply with a complete response in the
configured timeout.
So we figured: let's remove the whole thing and replace it with a
default _connect_ timeout to make sure that we can connect to a server
in 10s, but then give the server as long as it wants to complete its
response.
Closes#19509
Release Notes:
- Removed the `low_speed_timeout` setting from LLM provider settings,
since it was only used to _increase_ the timeout to give LLMs more time,
but since we don't have any other use for it, we simply remove the
setting to give LLMs as long as they need.
---------
Co-authored-by: Antonio <antonio@zed.dev>
Co-authored-by: Peter Tripp <peter@zed.dev>
This is a follow-up (or related to) #20412.
It fixes entries being marked when navigating between tabs with `cmd-[`
and `cmd-]`.
Turns out that deep in the bowels of the project panel, we check whether
a `shift` modifier was pressed - which is the case with `cmd-[` on a US
ANSI layout - and if so mark an entry.
I think that's a left-over, because all the other code paths that
select/reveal an entry mark it explicitly too.
Release Notes:
- Fixed entries in project panel being marked when navigating between
tabs with keybinding that uses `shift` modifier.
This adds a setting to disable inline completions in language scopes to,
for example, disable them in comments or strings.
This setting can be made per language.
Examples:
```json
{
"languages": {
"Go": {
"inline_completions_disabled_in": ["comment", "string"]
}
}
}
```
```json
{
"inline_completions_disabled_in": ["comment"]
}
```
Closes#9133
Release Notes:
- Added language setting to disable inline comments in certain scopes.
Example: `{"languages": {"Go": {"inline_completions_disabled_in":
["comment", "string"]}}}`
Co-authored-by: Bennet <bennet@zed.dev>
This addresses this comment:
https://github.com/zed-industries/zed/pull/13923#issuecomment-2467213210
With the change in here it's now possible to use the following settings:
```json
{
"lsp": {
"tailwindcss-language-server": {
"settings": {
"rootFontSize": 50
}
}
}
}
```
Closes https://github.com/zed-industries/zed/issues/10840
Release Notes:
- Added ability to configure `rootFontSize` for the
`tailwindcss-language-server`. Example settings: `{"lsp":
{"tailwindcss-language-server": {"settings": { "rootFontSize": 50}}}}`
Co-authored-by: Bennet <bennet@zed.dev>
This fixes the language server log menu only showing a single entry when
using SSH remoting.
Culprit was the `return menu;` statement that should've been a
`continue;`
Rest of the change is just refactoring.
Release Notes:
- Fixed `language server logs` menu only showing a single entry when
using SSH remoting.
Co-authored-by: Bennet <bennet@zed.dev>
This fixes the issue of multiple language servers showing up as `node`
in the language server logs dropdown.
It does this by changing `language_server.name()` to return the
adapter's name, not the binary name, and changing types to make sure
that we always use this.
Release Notes:
- Fixed language server names showing up only as `"node"`
---------
Co-authored-by: Sam Rose <hello@samwho.dev>
Co-authored-by: Bennet <bennet@zed.dev>
PR #20154 changed the project panel to focus the editor on click in case
preview tabs are disabled.
That lead to a flickering behavior: on mouse-down the border of the
still-selected entry in the project panel would flash, only to disappear
as soon as the entry was opened and editor focused.
This change fixes it by manually keeping track of the mouse-down state,
because we couldn't find a better solution that allows us to simply not
show the border while a "click" is going on.
Release Notes:
- Fixed project panel entries flickering the border when user clicks on
another entry to open it (when preview tabs are disabled.)
Co-authored-by: Piotr <piotr@zed.dev>
This addresses #17746 by marking entries when they're opened in the
project panel.
I think that was the original intention behind the code too, because it
explicitly marks entries before opening them. An event that is emitted
by the workspace reset the mark though.
So what I did was try to emulate the logic I saw in VS Code: when
opening the file, mark it, when the active entry changes, unmark it,
except if you explicitly marked a group of files.
Closes#17746
Release Notes:
- Changed project panel to mark files when opening them, which should
make it more intuitive to mark multiple files after opening a single
one.
PR #20154 introduced a regression and essentially disabled preview tabs
in code.
This fixes it and restores the old preview tabs behavior.
Release Notes:
- Fixed preview tabs being disabled in code, even if they were enabled
in the settings.
Co-authored-by: Piotr <piotr@zed.dev>
This fixes#15995 by adding a `project panel: remove from project`
action that can be used in a keybinding.
Release Notes:
- Added a `project panel: remove from project` action so that users can
now add a keybinding to trigger it: `project_panel::RemoveFromProject`.
This avoids inline completions being shown (and overriding `<tab>`
behavior) when a snippet is active and the user wants to go through
snippet placeholders with `<tab>`.
Easy to reproduce:
Open a Rust file and use the `tfn` snippet to produce a test function.
Delete the placeholder. Without the change here, the inline provider
would suggest a function name. If you `<tab>`, you accept it, but then
you can't `<tab>` into the function body.
With this change the inline completions are deactivated as long as a
snippet is active.
Closes#19484
Release Notes:
- Fixed inline completions (Copilot, Supermaven, ...) taking over when a
snippet completion was active. That resulted in `tab` not working to
jump to the next placeholder in the snippet.
This is a follow-up to #19913 and adds another "index" to the `Chunk`,
this time indexing the location of tabs.
Release Notes:
- N/A
---------
Co-authored-by: Antonio <antonio@zed.dev>
This adds support for [git
worktrees](https://matklad.github.io/2024/07/25/git-worktrees.html). It
fixes the errors that show up (git blame not working) and actually adds
support for detecting git changes in a `.git` folder that's outside of
our path (and not even in the ancestor chain of our root path).
(While working on this we discovered that our `.gitignore` handling is
not 100% correct. For example: we do stop processing `.gitignore` files
once we found a `.git` repository and don't go further up the ancestors,
which is correct, but then we also don't take into account the
`excludesFile` that a user might have configured, see:
https://git-scm.com/docs/gitignore)
Closes https://github.com/zed-industries/zed/issues/19842
Closes https://github.com/zed-industries/zed/issues/4670
Release Notes:
- Added support for git worktrees. Zed can now open git worktrees and
the git status in them is correctly handled.
---------
Co-authored-by: Antonio <antonio@zed.dev>
Co-authored-by: Bennet <bennet@zed.dev>
This previously didn't work: `d G` would delete to the end of the "first
of the soft-wrapped lines" of the last line.
To fix it, we special case the delete behavior for `shift-g`, which is
what Neovim also seems to do.
Release Notes:
- Fixed `d G` in Vim mode not deleting until the actual end of the
document if soft-wrap is turned on.
This spawns the runnable task that that's closest to the cursor.
One thing missing right now is that it doesn't find tasks that are
attached to non-outline symbols, such as subtests in Go.
Release Notes:
- Added a new reveal option for tasks: `"no_focus"`. If used, the tasks
terminal panel will be opened and shown, but not focused.
- Added a new `editor: spawn nearest task` action that spawns the task
with a run indicator icon nearest to the cursor. It can be configured to
also use a `reveal` strategy. Example:
```json
{
"context": "EmptyPane || SharedScreen || vim_mode == normal",
"bindings": {
", r t": ["editor::SpawnNearestTask", { "reveal": "no_focus" }],
}
}
```
Demo:
https://github.com/user-attachments/assets/0d1818f0-7ae4-4200-8c3e-0ed47550c298
---------
Co-authored-by: Bennet <bennet@zed.dev>
When the language server gave us a prompt and we'd close the window, we
wouldn't release the `project` until the next `flush_effects` call that
came in when opening a window.
With this change, we no longer hold a strong reference to the project in
the future. Fixes the leak and makes sure we clean up the SSH connection
when closing a window.
Release Notes:
- N/A
Co-authored-by: Bennet <bennet@zed.dev>
Turns out that we used client-side `fs` to check whether something is a
directory or not, which obviously doesn't work with SSH projects.
Release Notes:
- N/A
---------
Co-authored-by: Bennet <bennet@zed.dev>
We display the errors in another window anyway and if the connection
takes a while it looks like a bug that the modal stays open.
Release Notes:
- N/A
Co-authored-by: Bennet <bennet@zed.dev>
Before this change, we'd see a ton of requests from the Ollama provider
trying to fetch models:
```
[2024-10-28T15:00:52+01:00 DEBUG reqwest::connect] starting new connection: http://localhost:11434/
[2024-10-28T15:00:52+01:00 DEBUG reqwest::connect] starting new connection: http://localhost:11434/
[2024-10-28T15:00:52+01:00 DEBUG reqwest::connect] starting new connection: http://localhost:11434/
[2024-10-28T15:00:52+01:00 DEBUG reqwest::connect] starting new connection: http://localhost:11434/
[2024-10-28T15:00:52+01:00 DEBUG reqwest::connect] starting new connection: http://localhost:11434/
[2024-10-28T15:00:52+01:00 DEBUG reqwest::connect] starting new connection: http://localhost:11434/
[2024-10-28T15:00:52+01:00 DEBUG reqwest::connect] starting new connection: http://localhost:11434/
[2024-10-28T15:00:52+01:00 DEBUG reqwest::connect] starting new connection: http://localhost:11434/
[2024-10-28T15:00:52+01:00 DEBUG reqwest::connect] starting new connection: http://localhost:11434/
[2024-10-28T15:00:52+01:00 DEBUG reqwest::connect] starting new connection: https://api.zed.dev/
[2024-10-28T15:00:52+01:00 DEBUG reqwest::connect] starting new connection: http://localhost:11434/
[2024-10-28T15:00:52+01:00 DEBUG reqwest::connect] starting new connection: http://localhost:11434/
[2024-10-28T15:00:52+01:00 DEBUG reqwest::connect] starting new connection: http://localhost:11434/
[2024-10-28T15:00:52+01:00 DEBUG reqwest::connect] starting new connection: http://localhost:11434/
[2024-10-28T15:00:52+01:00 DEBUG reqwest::connect] starting new connection: http://localhost:11434/
[2024-10-28T15:00:52+01:00 DEBUG reqwest::connect] starting new connection: http://localhost:11434/
[2024-10-28T15:00:52+01:00 DEBUG reqwest::connect] starting new connection: http://localhost:11434/
```
Turns out we'd send a request on *every* change to settings.
Now, with this change, we only send a single request.
Release Notes:
- N/A
Co-authored-by: Bennet <bennet@zed.dev>
PR #19653 change the code in this diff, which lead to the remote_server
binary trying to load language grammars, which in turn failed, and
stopped languages from being loaded correctly.
That then lead to language servers not starting up.
This change reintroduces what #19653 removed, so that we don't load the
grammar on the remote_server, by ignoring the grammar name from the
config. The tests still all work.
Release Notes:
- N/A
Co-authored-by: Bennet <bennet@zed.dev>
This removes the old `remote_server { "download_binary_on_host": bool }`
field and replaces it with a `upload_binary: bool` on every
`ssh_connection`.
@ConradIrwin it compiles, it connects, but I haven't tested it really
yet
Release Notes:
- N/A
---------
Co-authored-by: Conrad <conrad@zed.dev>
Co-authored-by: Conrad Irwin <conrad.irwin@gmail.com>
This changes the download logic to not fetch the latest version, but to
fetch the version matching the current version of Zed.
Release Notes:
- Changed the update logic of the SSH remote server to not fetch the
latest version for a current channel, but to fetch the version matching
the current Zed version. If Zed is updated, the server is updated too.
If the server is newer than the Zed version an error will be displayed.
Release Notes:
- Changed the update process of the remote server binary to not attempt
an update if we can detect that the current binary is used by another
process.
- Changed the update process of the remote server binary to mark the
lock file as stale in case the SSH connection of the process that
created the lock file isn't open anymore.
Before this change, we would save the working directory *on the client*
of each shell that was running in a terminal.
While it's technically right, it's wrong in all of these cases where
`working_directory` was used:
- in inline assistant
- when resolving file paths in the terminal output
- when serializing the current working dir and deserializing it on
restart
Release Notes:
- Fixed terminals opened on remote hosts failing to deserialize with an
error message after restarting Zed.
This snuck in when Bennet and I were debugging why our connection to the
SSH host would break. We suspected that somewhere something was logging
to STDOUT and, I guess, we changed all `println!` to `eprintln!`.
Now, two weeks later, I'm sitting here, wondering why the version check
doesn't work anymore. The server always reports a version of `""`.
Turns out we take the command's STDOUT and not STDERR, which is correct.
But it also turns out we started to print the version to STDERR, which
breaks the version check.
One-character bug & one-character fix.
Release Notes:
- N/A
This adds the following optional setting:
```json
{
"remote_server": {
"download_on_host": false
}
}
```
Right now, it's **off by default** because I haven't tested it enough.
Release Notes:
- N/A
This creates a `<binary_path>.lock` file and checks for it whenever
uploading a binary.
Parameters:
- Wait for other instance to finish: 10m
- Mark lockfile as stale and ignore it after: 10m
- When waiting on another process, check every 5seconds
We can tweak all of them.
Ideally we'd have a value in the lockfile, that lets us detect whether
the other process is still there, but I haven't found something stable
yet:
- We don't have a stable PID on the server side when we run multiple
commands
- The `ControlPath` is on the client side
Release Notes:
- N/A
This contains a bunch of smallish but nasty fixes:
- Heartbeat timer was never reset after first heartbeat
- Use same return value when stderr is closed as when stdout is closed
- Always check proxy process status since it should also be done when we
get to this point (either it died and our task stopped, or our task
stopped and we dropped the process handle and it was killed on drop)
- make error messages less wrongly-specific
Release Notes:
- N/A
Co-authored-by: Bennet <bennet@zed.dev>
This fixes the `ssh` proxy process not being notified when the proxy
process dies. Turns out that the server would have stdout/stderr/stdin
connected to the grand-parent ssh process connected to it and as long as
the server kept running (even once it was daemonized into the
background) the grand-parent ssh process wouldn't exit.
That in turn meant that the Zed client wasn't notified when the proxy
process died.
Release Notes:
- N/A
---------
Co-authored-by: Bennet <bennet@zed.dev>
I suspect that this might have something to do with saving files not
being possible after a while.
I want to merge this and bump nightly.
Release Notes:
- N/A
Ran into this this morning. At least I suspect I ran into it. In any
case: we need to reset the missed hearbeats to 0 in case we got any
connection activity.
Release Notes:
- N/A
Before this change we had a race condition bug: if stderr was closed
before the other two sockets, we wouldn't properly detect when the
server died, and not report or retry anything.
That's because we treated a closed stderr as a non-error.
Technically, it isn't an error (closing a connection is okay!), but
until we have a proper shutdown ceremony between all three processes, we
can treat it as an error, because that lets us to detect when the server
is gone.
On the client-side, we also always react to these errors by
reconnecting. Except when we shutdown: there we do a proper shutdown and
won't error on the proxy exit code.
So, this works, even if I wish there was a better way for the server to
communicate to the proxy that it shutdown properly. But I don't want a
fourth socket.
Release Notes:
- N/A
This restores the change from #19193 that I erroneously reverted in
#19234.
I think the bug in #19275 got in my way when testing.
With that bug fixed, the changes in here also work fine.
Release Notes:
- N/A
When the SSH command dies or the server, the channel gets dropped and
the heartbeat method went into an infinite loop causing a hang.
Oversight from yesterday. Fixed now.
Release Notes:
- N/A
This does two things.
Important one: it reverts #19193, which lead to our whole process
handling breaking. When the `proxy` process was killed, it apparently
didn't close the stdout/stderr anymore, which meant we would not detect
when it died. (Watching its `status()` in the io loop also didn't work!)
We should figure out how to keep our process handling working before we
make this change in #19193, which sounds reasonable.
Second, less important thing: I think we should treat the process being
killed from a signal as non-zero, as an error.
Release Notes:
- N/A
Thorsten Ball
