Three motivations for this:
* Changing provider URL could cause credentials for the prior URL to be
sent to the new URL.
* The UI is in a misleading state after URL change - it shows a
configured API key, but on restart it will show no API key.
* #34110 will add support for both URL and key configuration for Ollama.
This is the first provider to have UI for setting the URL, and this
makes these issues show up more directly as odd UI interactions.
#37610 implemented something similar for the OpenAI and OpenAI
compatible providers. This extracts out some shared code, uses it in all
relevant providers, and adds more safety around key use.
I haven't tested all providers, but the per-provider changes were pretty
mechanical, so hopefully work properly.
Release Notes:
- Fixed handling of changes to LLM provider URL in settings to also load
the associated API key.
This has noticeable misbehavior when framerates are low (in my case this
sometimes happens when CPUs are throttled and compilation is happening),
as now a batch of x11 events can contain events over the span of 100s of
millis. So in that case, key press repetitions with quite normal typing
are skipped.
Under normal operating conditions it can be reproduced by running this
and quickly switching to Zed:
> sleep 1; for i in {1..5}; do xdotool type --delay 5 "aaaaaa "; xdotool
key Return; done
Output before looks like:
```
aaa
aaaaa
aaa
aaa
aaaa
```
Output after looks like:
```
aaaaaa
aaaaaa
aaaaaa
aaaaaa
aaaaaa
```
This behavior was added in #13955.
Release Notes:
- N/A
This allows you to write `buffer_snapshot.debug(ranges, value)` and it
will be displayed in the buffer (or multibuffer!) until that callsite
runs again. `ranges` can be any position (`usize`, `Anchor`, etc), any
range, or a slice or vec of those. `value` just needs a `Debug` impl.
These are stored in a mutable global for convenience, and this is only
available in debug builds.
For example, using this to visualize the captures of the brackets
Tree-sitter query:
<img width="1215" height="480" alt="image"
src="https://github.com/user-attachments/assets/c1878fc7-f6b3-4e27-949e-ecf67a7906b9"
/>
Release Notes:
- N/A
Now logs warnings for unrecognized capture names and logs errors for
missing required captures.
Release Notes:
- N/A
---------
Co-authored-by: Max <max@zed.dev>
#35053 split out these utility functions. I found the names / doc
comments a bit confusing so this improves that. Before that PR there was
also a mild inefficiency - it would walk the cursor all the way down to
a leaf and then back up to an ancestor.
Release Notes:
- N/A
Potentially fixes#37367. Just going ahead with the change even though
it's unclear whether this is the fix, since it is quite low risk.
Release Notes:
- N/A
Also:
* Adds tests for can_collect_data.
* Temporarily removes collection of diagnostics.
Release Notes:
- Edit Prediction: Fixed a bug where requests were marked eligible for
data collection despite the recent edit history in the request involving
files that may not be open source. The requests affected by this bug
will not be used in training data.
This change also causes Zeta to not do anything for editors that are not
associated with a project. In practice, this shouldn't affect any
behavior - those editors shouldn't have edit predictions anyway.
Release Notes:
- Edit Prediction: Requests no longer include recent edits from other
projects (other Zed windows).
See discussion on #36564
Adds a simple ad-hoc substring matching pattern language which allows
skipping a bounded number of chars between matched substrings. Before
this change compiling the regex was taking ~120ms on a fast machine and
~8mb of memory. This new version is way faster and uses minimal memory.
Checked the behavior of this vs by running it against 10k licenses that
happened to be in my home dir. There were only 4 differences of behavior
with the regex implementation, and these were false negatives for the
regex implementation that are true positives with the new one.
Of the ~10k licenses in my home dir, ~1k do not match one of these
licenses, usually because it's GPL/MPL/etc.
Release Notes:
- N/A
See discussion on #36564. Makes the license regexes a less fragile by
not matching on symbols, while also excluding cases where a long file
ends with a valid license. Also adds Zlib license, a commented out test
to check all license-like files discovered in the homedir, and more
testcases.
Not too happy with the efficiency here, on my quite good computer it
takes ~120ms to compile the regex and allocates ~8mb for it. This is
just not a great use of regexes, I think something using eager substring
matching would be much more efficient - hoping to followup with that.
Release Notes:
- Edit Prediction: Added Zlib license to open-source licenses eligible
for data collection.
To show these notifications, Zeta was being initialized with the initial
workspace it's used on - which may not even still exist! This removes a
confusing/misleading workspace field from Zeta.
Release Notes:
- N/A
The scanner is restarted after loading initial settings, and there was
an optimization to not re-discover and re-watch git repositories if they
already exist in the snapshot. #35865 added cleanup of watches that
occurred when the scanner restarts, and so in some cases repos were no
longer watched.
Release Notes:
- Linux: Fixed a case where Git repositories might not be watched for
changes, causing branch switching to not update the UI.
Co-authored-by: Julia <julia@zed.dev>
Accidentally regressed this in #35512, causing this migration to not
work and an error log to appear when one of these actions is in the user
keymap
Release Notes:
- N/A
An alternative might be to adjust the logic to not nest items when their
ranges are the same, but then clicking them doesn't work properly /
moving the cursor does not change which is selected. This could probably
be made to work with some extra logic there, but it seems overkill.
The downside of fixing it at the query level is that other parts of the
declaration are not inside the item range. This seems to be fine for
single line declarations - the nearest outline item is highlighted.
However, if a part of the declaration is not included in an item range
and is on its own line, then no outline item is highlighted.
Release Notes:
- Outline Panel: Fixed nesting of var and field declarations with
multiple identifiers in Go and C++
C++ before:
<img width="743" height="227" alt="image"
src="https://github.com/user-attachments/assets/af1a1d76-ecdc-4999-ae9c-95591726ccca"
/>
C++ after:
<img width="795" height="250" alt="image"
src="https://github.com/user-attachments/assets/49667ed3-e088-48b3-a9f0-6a119b5e7648"
/>
Go before:
<img width="859" height="306" alt="image"
src="https://github.com/user-attachments/assets/ecc7530a-ca16-4f37-b8d1-60687f178b12"
/>
Go after:
<img width="900" height="334" alt="image"
src="https://github.com/user-attachments/assets/d741cfb0-59e5-4d27-bd6a-f422204dc972"
/>
Closes#35070
Release Notes:
- Edit Prediction: Made license detection update eligibility for data
collection when license files change.
- Edit Prediction: Added Apache 2.0 license to opensource licenses
eligible for data collection.
- Edit Prediction: Made license detection less sensitive to whitespace
differences and check more files.
This was causing text thread summarization to be counted as a usage of 1
prompt
Release Notes:
- Fixed bug with agent text threads (not chat threads) counting
summarization as a usage of 1 prompt.
Co-authored-by: Oleksiy <oleksiy@zed.dev>
Release Notes:
- Edit Prediction: Added Git info to edit predictions requests (only
sent for opensource projects when data collection is enabled). The sent
Git info is the SHA of the current commit and the URLs for the `origin`
and `upstream` remotes.
Even after #35327 edit predictions were still being queried and shown
after setting `"disable_ai": true`
Also moves `DisableAiSettings` to the `project` crate so that it gets
included in tests via existing use of `Project::init_settings(cx)`.
Release Notes:
- Fixed `"disable_ai": true` setting disabling edit predictions.
This is consistent with there being no copilot expecific variant of
`editor::AcceptEditPrediction`. It also fixes a case where the
`disable_ai: true` has effects at init time that aren't undone when
changed, added in #35327.
Release Notes:
- N/A
* Experimentally `scroll_manager.anchor()` appears to be the same before
and after this 2nd call of `autoscroll_horizontally`
* Nothing these depend on seem to be mutated between the calls (and
since this is prepaint, stuff within editor also shouldn't be mutated)
Release Notes:
- N/A
Co-authored-by: Finn <finn@zed.dev>
Followup to #33678, doing the same thing for all JSON Schema files
provided to json-language-server
Release Notes:
* Added warnings for unknown fields when editing `tasks.json` /
`snippets.json`.
* Updates to `zed_llm_client-0.8.5` which adds support for `retry_after`
when anthropic provides it.
* Distinguishes upstream provider errors and rate limits from errors
that originate from zed's servers
* Moves `LanguageModelCompletionError::BadInputJson` to
`LanguageModelCompletionEvent::ToolUseJsonParseError`. While arguably
this is an error case, the logic in thread is cleaner with this move.
There is also precedent for inclusion of errors in the event type -
`CompletionRequestStatus::Failed` is how cloud errors arrive.
* Updates `PROVIDER_ID` / `PROVIDER_NAME` constants to use proper types
instead of `&str`, since they can be constructed in a const fashion.
* Removes use of `CLIENT_SUPPORTS_EXA_WEB_SEARCH_PROVIDER_HEADER_NAME`
as the server no longer reads this header and just defaults to that
behavior.
Release notes for this is covered by #33275
Release Notes:
- N/A
---------
Co-authored-by: Richard Feldman <oss@rtfeldman.com>
Co-authored-by: Richard <richard@zed.dev>
Closes#30017
* While generating the settings JSON schema, defaults all schema
definitions to reject unknown fields via `additionalProperties: false`.
* Uses `unevaluatedProperties: false` at the top level to check fields
that remain after the settings field names + release stage override
field names.
* Changes json schema version from `draft07` to `draft_2019_09` to have
support for `unevaluatedProperties`.
Release Notes:
- Added warnings for unknown fields when editing `settings.json`.
The major change in schemars 1.0 is that now schemas are represented as
plain json values instead of specialized datatypes. This allows for more
concise construction and manipulation.
This change also improves how settings schemas are generated. Each top
level settings type was being generated as a full root schema including
the definitions it references, and then these were merged. This meant
generating all shared definitions multiple times, and might have bugs in
cases where there are two types with the same names.
Now instead the schemar generator's `definitions` are built up as they
normally are and the `Settings` trait no longer has a special
`json_schema` method. To handle types that have schema that vary at
runtime (`FontFamilyName`, `ThemeName`, etc), values of
`ParameterizedJsonSchema` are collected by `inventory`, and the schema
definitions for these types are replaced.
To help check that this doesn't break anything, I tried to minimize the
overall [schema
diff](https://gist.github.com/mgsloan/1de549def20399d6f37943a3c1583ee7)
with some patches to make the order more consistent + schemas also
sorted with `jq -S .`. A skim of the diff shows that the diffs come
from:
* `enum: ["value"]` turning into `const: "value"`
* Differences in handling of newlines for "description"
* Schemas for generic types no longer including the parameter name, now
all disambiguation is with numeric suffixes
* Enums now using `oneOf` instead of `anyOf`.
Release Notes:
- N/A
Closes#32219#29666
Release Notes:
- Linux: Now skips insertion of characters when modifiers are held. Before, characters were inserted if there's no match in the keymap.
`git_panel::GenerateCommitMessage` has no handler,
`git::GenerateCommitMessage` should be preferred. Could add a
`#[action(deprecated_aliases = ["git_panel::GenerateCommitMessage"])]`,
but decided not to because that action didn't work. So instead uses of
it will show up as keymap errors.
Closes#32667
Release Notes:
- N/A
Flushes should happen after sending messages to X11 when effects should
be applied quickly. This is not needed for requests that return replies
since it automatically flushes in that case.
Release Notes:
- N/A
This reduces code complexity and avoids unnecessary roundtripping
through `DisplayPoint`. Hopefully this doesn't cause behavior changes,
but has one known behavior improvement:
`clip_at_line_ends` logic caused `is_inside_word` to return false when
on a word at the end of the line. In vim mode, this caused
`select_all_matches` to not select words at the end of lines, and in
some cases crashes due to not finding any selections.
Closes#29823
Release Notes:
- N/A
This is an easy way to shave some microseconds off the critical path for
frame rendering. On my machine this reduces typical frame rendering
latency by ~100 microseconds, probably quite a bit more on slower
machines.
Here is how long it typically takes to drop elements from the arena,
from a fairly brief run:
Release Notes:
- N/A
Instead of a menagerie of macros for implementing `Action`, now there
are just two:
* `actions!(editor, [MoveLeft, MoveRight])`
* `#[derive(..., Action)]` with `#[action(namespace = editor)]`
In both contexts, `///` doc comments can be provided and will be used in
`JsonSchema`.
In both contexts, parameters can provided in `#[action(...)]`:
- `namespace = some_namespace` sets the namespace. In Zed this is
required.
- `name = "ActionName"` overrides the action's name. This must not
contain "::".
- `no_json` causes the `build` method to always error and
`action_json_schema` to return `None`
and allows actions not implement `serde::Serialize` and
`schemars::JsonSchema`.
- `no_register` skips registering the action. This is useful for
implementing the `Action` trait
while not supporting invocation by name or JSON deserialization.
- `deprecated_aliases = ["editor::SomeAction"]` specifies deprecated old
names for the action.
These action names should *not* correspond to any actions that are
registered. These old names
can then still be used to refer to invoke this action. In Zed, the
keymap JSON schema will
accept these old names and provide warnings.
- `deprecated = "Message about why this action is deprecation"`
specifies a deprecation message.
In Zed, the keymap JSON schema will cause this to be displayed as a
warning. This is a new feature.
Also makes the following changes since this seems like a good time to
make breaking changes:
* In `zed.rs` tests adds a test with an explicit list of namespaces. The
rationale for this is that there is otherwise no checking of `namespace
= ...` attributes.
* `Action::debug_name` renamed to `name_for_type`, since its only
difference with `name` was that it
* `Action::name` now returns `&'static str` instead of `&str` to match
the return of `name_for_type`. This makes the action trait more limited,
but the code was already assuming that `name_for_type` is the same as
`name`, and it requires `&'static`. So really this just makes the trait
harder to misuse.
* Various action reflection methods now use `&'static str` instead of
`SharedString`.
Release Notes:
- N/A
I don't have a concrete misbehavior from it, but this update of
`insert_range` doesn't make sense for two reasons:
* If the resolved completion doesn't have a new `text_edit` it would
clear out the `insert_range`.
* It doesn't update the completion if it has already been resolved,
except this update of `insert_range` happened before that.
Guessing it was written this way because this field needed to only be
mutated within the `CompletionSource::Lsp` case and this was a
convenient match below.
Release Notes:
- N/A
Co-authored-by: Smit <smit@zed.dev>
Having `Thread::last_usage` as an override of the initially fetched
usage could cause the initial usage to be displayed when the current
thread is empty or in text threads. Fix is to just store last usage info
in `UserStore` and not have these overrides
Release Notes:
- Agent: Fixed request usage display to always include the most recently
known usage - there were some cases where it would show the initially
requested usage.
Bug in #31872Closes#32774
Release Notes:
- Fixed a bug in LSP completions caching where prior completions may be
used when they should not, after typing a trigger char like `.`
For me this is a panic that started occurring today in my use of Zed.
The repro is to type `ctrl-x` to start a pending key sequence and then
close the collab side panel with the mouse. The issue is that
dispatching the action based on pending keystrokes uses the same
`DispatchNodeId` as when the 1 second timer was started.
`DispatchNodeId` is not stable across frames. This also means that the
wrong `DispatchNodeId` can be used in the non-panicing case, potentially
causing the action to not occur.
The mystery here is why did this only start happening now in my use of
Zed, and why isn't it showing up in the panics dashboard / issue
reports.
Panic looks like
```
{
"thread": "main",
"payload": "index out of bounds: the len is 467 but the index is 1861",
"location_data": {
"file": "crates/gpui/src/key_dispatch.rs",
"line": 519
},
"backtrace": [
"zed::reliability::init_panic_hook::{{closure}}::he1d8257b19b16eec+155265758",
"std::panicking::rust_panic_with_hook::h33b18b24045abff4+128544307",
"std::panicking::begin_panic_handler::{{closure}}::hf8313cc2fd0126bc+128543530",
"std::sys::backtrace::__rust_end_short_backtrace::h57fe07c8aea5c98a+128537145",
"__rustc[95feac21a9532783]::rust_begin_unwind+128542669",
"core::panicking::panic_fmt::hd54fb667be51beea+9456688",
"core::panicking::panic_bounds_check::h1a9bf3d94de0fc80+9457170",
"gpui::key_dispatch::DispatchTree::dispatch_path::hce77d277881569bf+73992023",
"gpui::app::App::spawn::{{closure}}::hb1e79bbbdead3012+73687056",
"async_task::raw::RawTask<F,T,S,M>::run::hd13f66f99bb24bbd+70694231",
"<gpui::platform::linux::x11::client::X11Client as gpui::platform::linux::platform::LinuxClient>::run::h5a92ddaaf9a06dd1+74465138",
"gpui::platform::linux::platform::<impl gpui::platform::Platform for P>::run::hd19ac52b2d94268e+74064525",
"gpui::app::Application::run::hee83110c717a5af0+151862692",
"zed::main::hca7e2265584c4139+153307630",
"std::sys::backtrace::__rust_begin_short_backtrace::h2e04f4034c2d82c5+153146899",
"std::rt::lang_start::{{closure}}::h91cf1ca0eeae23ae+154454121",
"std::rt::lang_start_internal::h418648f91f5be3a1+128467809",
"main+153326748",
"__libc_start_call_main+25056432783818",
"__libc_start_main_impl+25056432784011",
"_start+12389486"
],
"app_version": "0.190.6",
"app_commit_sha": "9a2dcbbe244407fed51d61f38e4a4a59ec1cccc6",
"release_channel": "stable",
"target": "x86_64-unknown-linux-gnu",
"os_name": "Linux X11",
"os_version": "ubuntu 24.04",
"architecture": "x86_64",
"panicked_on": 1750185799233,
"system_id": "abae7201-61fb-442b-922b-202071ae81c0",
"installation_id": "69a0fb9a-11a2-4065-ad8c-b281e68525ad",
"session_id": "bc5b5f2f-e4c3-44a8-948e-c0550a2e2ef2"
}
```
Release Notes:
- Fixed a rare panic / potential incorrect action dispatch when a
pending keysequence is applied after the 1 second timer elapsing.
Continuing this work from a while back in #21079, now greatly aided by
agent + sonnet 4. With this change, there are now only a few spots that
explicitly panic, though errors during initialization will panic.
Motivation was this recent user panic in `handle_event`, figured fixing
all this use of unwrap was a great use of the agent.
> called `Result::unwrap()` on an `Err` value: X11 GetProperty for
_NET_WM_STATE failed.
Release Notes:
- N/A
On the panics dashboard, saw this panic of `There must be at least one
selection` in `open_locations_in_multibuffer`. Only seems to have
happened once in the past month.
Fix is to include the pending selection. Since `selections.all()` cannot
provide anchor selections, added `selections.all_anchors()` which only
really does any work if there is a pending selection.
Also fixes a corner case in jump-to-definitions where if the definition
is `HoverLink::InlayHint` and the `compute_target_location` fails for
all definitions it could potentially also trigger this case (and return
`Navigated::Yes` instead of `Navigated::No`
Release Notes:
- N/A
Crashes look like:
```
Panic `offset 632 is greater than the snapshot.len() 631` on thread 0 (com.apple.main-thread)
<multi_buffer::MultiBufferSnapshot>::innermost_enclosing_bracket_ranges::<usize>
editor::highlight_matching_bracket::refresh_matching_bracket_highlights
<gpui::app::App>::update_window_id::<bool, <gpui::app::context::Context<editor::Editor>>::subscribe_in<multi_buffer::MultiBuffer, multi_buffer::Event, <editor::Editor>::on_buffer_event>::{closure#0}::{closure#0}>::{closure#0}
<gpui::app::context::Context<editor::Editor>>::subscribe_in::<multi_buffer::MultiBuffer, multi_buffer::Event, <editor::Editor>::on_buffer_event>::{closure#0}
<gpui::app::App>::flush_effects
<project::lsp_store::LocalLspStore>::format_buffer_locally::{closure#0}
<project::lsp_store::LspStore>::format::{closure#1}::{closure#0}::<i32>
```
Though `format_buffer_locally` is not always present. Both issue reports
mention usage of the agent. I suspect this is somehow a result of agent
format-on-save combined with the user's cursor being at the end of the
buffer as it's getting edited by the agent.
The offsets are always off-by-one in the error, so at first I thought
the issue was the condition `head < snapshot.buffer_snapshot.len()`
before setting `tail` to be `head + 1`, but an offset equal to len is
valid. Seems like to get a `to_offset` crash, `head` must be greater
than `len`. Which is quite weird, a selection's offset should never be
out of bounds.
Since this code is just about highlighting brackets, this PR logs an
error instead of crashing in the `head > len` case.
Closes#32732, #32171
Release Notes:
- N/A
Most of the default icon sets on Ubuntu do not use the names that were
there. To fix, using the icon synonyms from the chromium source. This
will probably fix some of the linux mouse cursor issues tracked in
#26141
Also adds a note in the load failure logs mentioning that misconfigured
`XCURSOR_PATH` may be the issue. I ran into this because [the alacritty
snap incorrectly sets
XCURSOR_PATH](https://github.com/snapcrafters/alacritty/issues/21).
On X11 also adds:
* Caching of load errors to log once for missing cursor icons.
* Fallback on default cursor icon. This way if there was a transition
from a non-default icon to a missing icon it doesn't get stuck showing
the non-default icon.
Leaving release notes blank as I have other mouse cursor fixes and would
prefer to just have one entry in the release notes.
Release Notes:
- N/A
Closes#27384
I wrote #32771 before seeing #27384, and hoped that change would fix it.
It didn't because `XkbSelectNotify` wants a mask of which types of
`XkbMapNotify` to deliver, and otherwise won't send them.
I noticed quite a few events are sent just for remapping a single
keycode, so I updated the event loop to deduplicate these events (since
the handler does not attempt to apply them and instead just re-queries
keyboard info).
Also adds a missing call of `keyboard_layout_change` on `XkbMapNotify`
and `XkbNewKeyboardNotify`.
Release Notes:
- x11: Fixed handling of key remapping occurring while Zed is running
(e.g. xmodmap)
This adds handling of UnmapNotify / MapNotify / VisibilityNotify to
track whether windows are visible. When hidden, the refresh loop is
halted until visible again. Often these refreshes were just checking if
the window is dirty, but I believe it sometimes did a full re-render for
things that change without user interaction (cursor blink, animations).
This also changes handling of Expose events to set a flag indicating the
next refresh should have `require_presentation: true`.
Release Notes:
- x11: No longer refreshes windows that aren't visible.
Keyboard hot reloading was added in #15059, but also reloaded this on
MapNotify instead of XkbMapNotify, so it wasn't handling keymap change
events and was instead reloading when windows are mapped (typically when
they go from a minimized / other workspace state to being visible).
Release Notes:
- N/A
The condition for displaying the first char of the placeholder text in
the block cursor was `cursor_column == 0`. This meant that it was
displayed on the first column even when the placeholder text is not
being displayed. Instead this now shows it only when
`snapshot.is_empty()` - the same condition used to determine whether to
show placeholder text.
In the case of vim mode + agent panel message editor, this meant that if
you did `shift-enter` to make a newline and then `escape` to enter
normal mode, the block cursor would show `M` inside it as that's the
first character of the placeholder text "Message the agent - @ to
include context"
Release Notes:
- N/A
Michael Sloan
