Refs #29733
This pull request introduces a new field to the `StreamingEditFileTool`
that lets the model create or overwrite a file in a streaming way. When
one of the `assistant.stream_edits` setting / `agent-stream-edits`
feature flag is enabled, we are going to disable the `CreateFileTool` so
that the agent model can only use `StreamingEditFileTool` for file
creation.
Release Notes:
- N/A
---------
Co-authored-by: Ben Brandt <benjamin.j.brandt@gmail.com>
Co-authored-by: Oleksiy Syvokon <oleksiy.syvokon@gmail.com>
This pull request introduces a new tool for streaming edits. The
short-term goal is for this tool to replace the existing `EditFileTool`,
but we want to get this out the door as soon as possible so that we can
start testing it.
`StreamingEditFileTool` is mutually exclusive with `EditFileTool`. It
will be enabled by default for anyone who has the `agent-stream-edits`
feature flag, as well as people that set `assistant.stream_edits` to
`true` in their settings.
### Implementation
Streaming is achieved by requesting a completion while the `edit_file`
tool gets called. We invoke the model by taking the existing
conversation with the agent and appending a prompt specifically tailored
for editing. In that prompt, we ask the model to produce a stream of
`<old_text>`/`<new_text>` tags. As the model streams text in, we
incrementally parse it and start editing as soon as we can.
### Evals
Note that, as part of this pull request, I also defined some new evals
that I used to drive the behavior of the recursive LLM call. To run
them, use this command:
```bash
cargo test --package=assistant_tools --features eval -- eval_extract_handle_command_output
```
Or comment out the `#[cfg_attr(not(feature = "eval"), ignore)]` macro.
I recommend running them one at a time, because right now we don't
really have a way of orchestrating of all these evals. I think we should
invest into that effort once the new agent panel goes live.
Release Notes:
- N/A
---------
Co-authored-by: Nathan Sobo <nathan@zed.dev>
Co-authored-by: Bennet Bo Fenner <bennetbo@gmx.de>
Co-authored-by: Oleksiy Syvokon <oleksiy.syvokon@gmail.com>
- Stop merging same row diagnostics
- (for Rust) show code fragments surrounded by `'s in monospace
Co-authored-by: Serge Radinovich <sergeradinovich@gmail.com>
Closes#29362
Release Notes:
- diagnostics: Diagnostics are no longer merged when they're on the same
line
- rust: Diagnostics now show code snippets in monospace font:
<img width="551" alt="Screenshot 2025-04-29 at 16 13 45"
src="https://github.com/user-attachments/assets/d289be31-717d-404f-a76a-a0cda3e96fbe"
/>
Co-authored-by: Serge Radinovich <sergeradinovich@gmail.com>
Closes#28135Closes#4388Closes#28136
Release Notes:
- diagnostics: Show the diagnostic code if available
---------
Co-authored-by: Neo Nie <nihgwu@live.com>
Co-authored-by: Zed AI <ai+claude-3.7@zed.dev>
This PR uses Tree Sitter to show inline values while a user is in a
debug session.
We went with Tree Sitter over the LSP Inline Values request because the
LSP request isn't widely supported. Tree Sitter is easy for
languages/extensions to add support to. Tree Sitter can compute the
inline values locally, so there's no need to add extra RPC messages for
Collab. Tree Sitter also gives Zed more control over how we want to show
variables.
There's still more work to be done after this PR, namely differentiating
between global/local scoped variables, but it's a great starting point
to start iteratively improving it.
Release Notes:
- N/A
---------
Co-authored-by: Piotr Osiewicz <peterosiewicz@gmail.com>
Co-authored-by: Anthony Eid <hello@anthonyeid.me>
Co-authored-by: Cole Miller <m@cole-miller.net>
Co-authored-by: Anthony <anthony@zed.dev>
Co-authored-by: Kirill <kirill@zed.dev>
Closes#29176
This PR fix an issue where uncommenting a code block in Markdown would
add Markdown comments instead of removing the language-specific
comments.
Why?
`language_scope_at` for comments in a code block in Markdown would
result in the language being detected as Markdown. This happens because
the smallest range, such as `//` or `#` on the Markdown layer, is
preferred over `// whole comment line` for any other language. This
results in language detection as Markdown for that point.
To fix this, we also use a depth factor and try to prefer the layer with
greater depth over one with lesser depth. In this case, the code block's
language depth would be preferred over Markdown. The smallest range is
now used as a tiebreaker.
Added test for this case.
Release Notes:
- Fixed issue where uncommenting a code block in Markdown would add
Markdown comments instead of removing the language comments.
Release Notes:
- "Block" diagnostics (that show up in the diagnostics view, or when
using `f8`/`shift-f8`) are rendered more clearly
- `f8`/`shift-f8` now always go to the "next" or "prev" diagnostic,
regardless of the state of the editor
---------
Co-authored-by: Kirill Bulatov <mail4score@gmail.com>
Co-authored-by: Julia Ryan <juliaryan3.14@gmail.com>
This PR fixes a minor regression introduced in #27718, where snippets
stopped working when the language was set to plaintext because
`languages_at` doesn't include plaintext, while `language_at` does.
Release Notes:
- Fixed plaintext snippets not working
Closes#21541Closes#22726
This should fix snippets in languages, like PHP, that are based on the
HTML syntax layer. To be honest, I don't totally get where HTML comes
into it, but the issues outlined in #21541 and #22726 both boil down to
"Zed only shows me HTML snippets in PHP/ERB files; I expected to see
PHP/ERB snippets". This solution is based on the comments between
@mrnugget and @osiewicz in #22726: resolve/combine snippets for all
language layers at the given position, whereas current behavior is to
resolve snippets only for the `.last()` language layer at the given
position.
- add `Buffer:languages_at()` (note the plural)
- update `snippet_completions()` in `editor.rs` to loop over each
language, gathering snippets as it goes
- the primary logic for resolving snippets within a single language has
not changed
### Verifying this change
I couldn't find tests related to snippet and currently active languages
(CI may show them to me 😆 ) but I can add some if desired and w/ perhaps
a little coaching or prompting about another test to look to for
inspiration. I have confirmed that this works for PHP, but I have not
checked ERB because I'm not familiar with it or set up for it.
To check this manually:
1. install the PHP extension
2. install at least 1 snippet for each of html, php and phpdoc. If you
don't have any, these should work:
```sh
# BEWARE these will clobber existing snippets!
echo '{"dddd":{"body":"hello from phpdoc"}}' > ~/.config/zed/snippets/phpdoc.json
echo '{"pppp":{"body":"hello from PHP"}}' > ~/.config/zed/snippets/php.json
echo '{"hhhh":{"body":"hello from HTML"}}' > ~/.config/zed/snippets/html.json
```
3. open any PHP file. If you don't have one, here's one that should
work:
```php
<?php
/**
*
*/
function function_name()
{
}
```
4. Place your cursor in a PHPdoc comment (eg after the `/**` on line 3)
- you should be able to use the `dddd`, `pppp` and `hhhh` snippets; on
`main`, only the `dddd` snippet works here
5. Move your cursor to a non-comment PHP area (eg after the `{` on line
7)
- you should be able to use the `pppp` and `hhhh` snippets, but not
`dddd`; on `main`, only `hhhh` works here
### Performance
This adds 2 separate (not nested) loops to `snippet_completions()`, each
of which will iterate over the active language scopes at the given
location. I have not looked into the specifics of how many layers most
languages have, but I suspect that *most* users will see identical
performance as before because there will only be 1 scope active most of
the time.
In some cases, though (eg PHP, ERB, maybe template strings in JS), the
editor will be looping over more layers, possibly many in some deeply
injected/embedded cases (I'm thinking of a regex template string in a JS
heredoc string in a PHP script in an HTML file). I don't expect this to
be an issue – nor has it been in my usage and testing – but performance
of snippets could be affected in pathological cases.
### Alternate solutions
Instead of resolving snippets for *all* layers, we could just change how
we pick which language to resolve. Instead of always using `.last()`,
perhaps we could do something more clever. This feels like it could be
tricky and potentially error prone, though.
Release Notes:
- Snippets are now resolved for all languages active at the cursor
location.
- Fixed snippets in PHP, ERB and other languages whose syntax layers are
based on HTML
Follow-up of https://github.com/zed-industries/zed/pull/26410
The action does not sort the items the way Vim does, but still better
than the previous state.
Release Notes:
- N/A
Co-authored-by: Conrad Irwin <conrad.irwin@gmail.com>
Release Notes:
- Fixed a rendering bug that caused context in the agent to not wrap
properly.
---------
Co-authored-by: Conrad Irwin <conrad.irwin@gmail.com>
Co-authored-by: Zed AI <ai+claude-3.7@zed.dev>
When a file is deleted outside of Zed, but it doesn't have any unsaved
changes, it shouldn't be considered "dirty" (prompting you before you
close it).
Release Notes:
- Fixed an bug where unchanged buffers were marked as conflicting if
their files were deleted outside of Zed.
---------
Co-authored-by: Antonio <antonio@zed.dev>
Closes#18581
Now characters for completing query and word characters, which are
responsible for selecting words by double clicking or navigating, are
different. This fixes a bunch of things:
For settings.json, this improves completions to treat the whole string
as a completion query, instead of just the last word. We now added
"space" as a completion query character without it being a word
character.
For keymap.json, this improves selecting part of an action as the ":"
character is only a completion character and not a word character. So,
completions would still trigger on ":" and query capture will treat ":"
as a word, but for actions like selections and navigation, ":" will be
treated as punctuation.
Before:
Unnecessary related suggestions as query is only the last word which is
"d".
<img width="300" alt="image"
src="https://github.com/user-attachments/assets/8199a715-7521-49dd-948b-e6aaed04c488"
/>
Double clicking `ToggleFold` selects the whole action:
<img width="300" alt="image"
src="https://github.com/user-attachments/assets/c7f91a6b-06d5-45b6-9d59-61a1b2deda71"
/>
After:
Now query is "one d" and it shows only matched ones.
<img width="300" alt="image"
src="https://github.com/user-attachments/assets/1455dfbc-9906-42e8-b8aa-b3f551194ca2"
/>
Double clicking `ToggleFold` only selects part of the action, which is
more refined behavior.
<img width="300" alt="image"
src="https://github.com/user-attachments/assets/34b1c3c2-184f-402f-9dc8-73030a8c370f"
/>
Release Notes:
- Improved autocomplete suggestions in `settings.json`, now whole string
is queried instead of just last word of string, which filters out lot of
false positives.
- Improved selection of action in `keymap.json`, where now you can
double click to only select certain part of action, instead of selecting
whole action.
---------
Co-authored-by: Max Brunsfeld <maxbrunsfeld@gmail.com>
Co-authored-by: Ben Kunkle <ben@zed.dev>
This is the core change:
https://github.com/zed-industries/zed/pull/26758/files#diff-044302c0d57147af17e68a0009fee3e8dcdfb4f32c27a915e70cfa80e987f765R1052
TODO:
- [x] Use AsyncFn instead of Fn() -> Future in GPUI spawn methods
- [x] Implement it in the whole app
- [x] Implement it in the debugger
- [x] Glance at the RPC crate, and see if those box future methods can
be switched over. Answer: It can't directly, as you can't make an
AsyncFn* into a trait object. There's ways around that, but they're all
more complex than just keeping the code as is.
- [ ] Fix platform specific code
Release Notes:
- N/A
Follow-up of https://github.com/zed-industries/zed/pull/26410
* Extract word completions into their own, `editor::ShowWordCompletions`
action so those could be triggered independently of completions
* Assign `ctrl-shift-space` binding to this new action
* Still keep words returned along the completions as in the original PR,
but:
* Tone down regular completions' fallback logic, skip words when the
language server responds with empty list of completions, but keep on
adding words if nothing or an error were returned instead
* Adjust the defaults to wait for LSP completions infinitely
* Skip "words" with digits such as `0_usize` or `2.f32` from completion
items, unless a completion query has digits in it
Release Notes:
- N/A
Instead of getting the whole text from the buffer, replacing with
`String::replace`, and getting a whole diff, we'll now use `SearchQuery`
to get a range, diff only that range, and apply it (all in the
background).
When we match zero strings, we'll record a "bad search", keep going and
report it to the model at the end.
Release Notes:
- N/A
---------
Co-authored-by: Max <max@zed.dev>
Closes https://github.com/zed-industries/zed/issues/4957https://github.com/user-attachments/assets/ff491378-376d-48ec-b552-6cc80f74200b
Adds `"completions"` language settings section, to configure LSP and
word completions per language.
Word-based completions may be turned on never, always (returned along
with the LSP ones), and as a fallback if no LSP completion items were
returned.
Future work:
* words are matched with the same fuzzy matching code that the rest of
the completions are
This might worsen the completion menu's usability even more, and will
require work on better completion sorting.
* completion entries currently have no icons or other ways to indicate
those are coming from LSP or from word search, or from something else
* we may work with language scopes more intelligently, group words by
them and distinguish during completions
Release Notes:
- Supported word-based completions
---------
Co-authored-by: Max Brunsfeld <max@zed.dev>
This is motivated by trying to make the Project Diff view usable with
huge Git change sets.
Release Notes:
- Improved performance of rendering multibuffers with very large numbers
of buffers
Closes https://github.com/zed-industries/zed/issues/24914 (again)
Release Notes:
- Fixed an issue where multi-line pasted content was auto-indented
incorrectly if copied from the middle of an existing line.
Closes#4271
Implemented by kicking of a task on the main thread at the end of
`Editor::handle_input` which waits for the buffer to be re-parsed before
checking if JSX tag completion possible based on the recent edits, and
if it is then it spawns a task on the background thread to generate the
edits to be auto-applied to the buffer
Release Notes:
- Added support for auto-closing of JSX tags
---------
Co-authored-by: Cole Miller <cole@zed.dev>
Co-authored-by: Max Brunsfeld <max@zed.dev>
Co-authored-by: Marshall Bowers <git@maxdeviant.com>
Co-authored-by: Mikayla <mikayla@zed.dev>
Co-authored-by: Peter Tripp <peter@zed.dev>
This PR adds toasts for reporting success and errors from remote git
operations. This PR also adds a focus handle to notifications, in
anticipation of making them keyboard accessible.
Release Notes:
- N/A
---------
Co-authored-by: julia <julia@zed.dev>
Closes: #25556
We were always comparing `disabled_globs` against the relative file
path, we'll now use the absolute path if the glob is also absolute.
Release Notes:
- Support absolute globs in `edit_predictions.disabled_globs`
Closes#25471
In languages like Swift, names can be concatinated in form like `class
Example: UI`, notice here `Example` and `:` are two different words.
Before, `name_ranges`translation of above text would look like:
```
"class" -> [0..5]
" Example" -> [5..13] (Spaces are intentional)
"e:" -> [12..14] (This is incorrect, and should be ":" -> [13..14])
" UI" -> [14..16]
```
Because this translation does not account for concatinated words, this
might affect queries, but most importantly this panics when multi-byte
character (`ф`) is used in place of `e`, as it then tries to access
index which lies inside that multi-byte. For example, it panics on
`class Examplф: UI`.
---
This PR fixes this by handing concatinated words when calculating
`name_ranges`.
Now, the corrected ranges will look like:
```
"class" -> [0..5]
" Example" -> [5..13]
":" -> [13..14] (Now it's correct)
" UI" -> [14..16]
```
and for multi-byte character
```
"class" -> [0..5]
" Examplф" -> [5..14] (Notice ф takes two bytes)
":" -> [14..15]
" UI" -> [15..17]
```
This way, it no longer tries to access a previous index, preventing a
panic when that index contains a multi-byte character.
Release Notes:
- Fixed a panic when Cyrillic characters are used in languages like
Swift.
* When staging in a buffer whose file has been deleted, do not save the
file
* Fix logic for writing to index when file is deleted
Release Notes:
- N/A
Closes#12064
It feels a bit strange to use `brackets` for this but it seems to work
without unintended consequences from my testing so far.
Release Notes:
- N/A
---------
Co-authored-by: Marshall Bowers <git@maxdeviant.com>
Closes https://github.com/zed-industries/zed/issues/10122
Closes https://github.com/zed-industries/zed/issues/25034
When formatting buffers or reloading them after they change on disk, we
performed a diff between the buffer's current contents and the new
content. We need this diff in order preserve the positions of cursors
and other decorations when updating the buffer's text.
In order to handle changes within lines, we would previously compute a
*character-wise* diff. This was extremely expensive for large files.
This PR gets rid of the character-wise diff, and instead performs a
normal line-wise diff. Then, for certain replace hunks, we compute a
secondary word-based diff. Also, I've switched to the
[`imara-diff`](https://github.com/pascalkuthe/imara-diff) crate, instead
of `similar`.
Release Notes:
- Fixed a hang that could occur when large files were changed on disk or
formatted.
The language::markdown crate had been superceded by markdown::Mardown.
After #25117, the only two remaining use-cases were rendering git commit
messages (which are arguably not really markdown) and the signature help
(which is definitely not markdown).
Updated the former to use the new markdown component, and the latter to
do syntax highlighting manually.
Release Notes:
- Allow selecting the commit message in git commits
Closes#18641
Contributes: #13194
Release Notes:
- Open LSP documentation file links in Zed not the system opener
- Render completion documentation markdown consistently with
documentation markdown
Done automatically with
> ast-grep -p '$A.background_executor().spawn($B)' -r
'$A.background_spawn($B)' --update-all --globs "\!crates/gpui"
Followed by:
* `cargo fmt`
* Unexpected need to remove some trailing whitespace.
* Manually adding imports of `gpui::{AppContext as _}` which provides
`background_spawn`
* Added `AppContext as _` to existing use of `AppContext`
Release Notes:
- N/A
Release Notes:
- Fixed a panic when displaying a whitespace-only line in the edit
prediction preview
---------
Co-authored-by: Antonio <antonio@zed.dev>
We've seen a few crashes in `SyntaxSnapshot::reparse_with_ranges` during
`Buffer::preview_edits`, where an offset conversion fails because it is
out of range.
We are not sure how exactly this is happening.
Our theory is that the syntax snapshot is using an outdated state when
edits happen in the meantime (while interpolating). This is an attempt
to see if it helps with the panics, hopefully we can revisit this when
we have a better understanding of the issue.
Co-Authored-by: Antonio <antonio@zed.dev>
Release Notes:
- N/A
Co-authored-by: Antonio <antonio@zed.dev>
This PR makes it so we send up the diagnostic groups as additional data
with the edit prediction request.
We're not yet making use of them, but we are recording them so we can
use them later (e.g., to train the model).
Release Notes:
- N/A
---------
Co-authored-by: Nathan <nathan@zed.dev>
- **store `buffer::Diagnostic`as NumberOrString instead of assuming
String**
- **update zed-industries/lsp-types rev**
Closes#24081
Release Notes:
- Fixed an issue where language server diagnostic codes would be converted to strings leading to errors with some language servers
* fix syntax highlighting of deleted text when buffer language changes
* do not highlight entire untracked files as created, except in the
project diff view
Release Notes:
- N/A
Co-authored-by: ConradIrwin <conrad.irwin@gmail.com>
Co-authored-by: cole-miller <m@cole-miller.net>
