After the user confirmation, only the current instance of the
completions provider had the answer stored.
In this PR, the changes are propagated by having each provider have an
`Entity<choice>`, and having a lookup map with one `Entity<choice>` for
each worktree that `Zeta` has seen.
Release Notes:
- N/A
This PR changes the edit predictions URL for Zed Staff back to
`llm.zed.dev/predict_edits`.
This endpoint is now being routed to the Cloudflare Workers instead of
the LLM service.
Release Notes:
- N/A
Realized that the logic in #23814 was more than needed, and harder to
maintain. Something like that could make sense if using the tokenizer
and wanting to precisely hit a token limit. However in the case of edit
predictions it's more of a latency+expense vs capability tradeoff, and
so such precision is unnecessary.
Happily this change didn't require much extra work, just copy-modifying
parts of that change was sufficient.
Release Notes:
- N/A
This ensures that we do not fetch a new completion when the edits of the
user can be interpolated.
E.g. (suggestions in `[]`):
```rust
s[truct Person {}]
```
Then if i type out `truct` we will not fetch a new completion
Release Notes:
- N/A
Limits the size of the buffer excerpt and the size of change history.
Release Notes:
- N/A
---------
Co-authored-by: Richard <richard@zed.dev>
Co-authored-by: Joao <joao@zed.dev>
This PR makes it so staff edit predictions now go through Cloudflare
Workers instead of going to the LLM service.
This will allow us to dogfood the new LLM worker to make sure it is
working as expected.
Release Notes:
- N/A
Previously we returned an error when the interpolation failed in
`process_completion_response`.
However, it is not an error when interpolation returns `None`. That just
means that the predicted edits can be discarded, because the user typed
something that is not a subset of what the model predicted OR if the
model responds with a no-op.
```
2025-01-29T09:44:30.221135+01:00 [ERROR] zeta prediction failed
Caused by:
Interpolated edits are empty
```
Release Notes:
- N/A
This fixes the panics we we're seeing in `EditPreview::highlight_edits`.
The reason for this was that we were interpolating edits incorrectly.
Here's an example:
```rust
let a = 0; // existing code
let c = 2; // suggested by edit prediction
```
The edits would look like this: `[(Point(1, 0)..Point(1, 0), "let c =
2;"]`
Now i type:
```rust
let a = 0; // existing code
let b = 1; // added this line
let c = 2; // suggested by edit prediction
```
Before this change, the `interpolate` function would allow insertions
before the edit prediction edits, the anchors will move to the next
line.
The edits would look now like this: `[(Point(2, 0)..Point(2, 0), "let c
= 2;"]`
However, now we end up with a call to `EditPreview::highlight_edits`,
with the following parameters:
- current_snapshot:
```rust
let a = 0;
let b = 1;
```
- edits: `[(Point(2, 0)..Point(2, 0), "let c = 2;"]`
- applied_edits_snapshot:
```rust
let a = 0;
let c = 2;
```
And here you can see the issue, applying the `edits` to the
`current_snapshot` should always end up re-creating the text that is
present in the `applied_edits_snapshot`. That is not the case here
though, meaning that the offsets in the new buffer are not correct,
which can either lead to a confusing popup or a crash if the suggestion
is at the end of the file.
Here's a real world example (edit prediction is ONLY suggesting to
delete a new line):
<img width="487" alt="Screenshot 2025-01-27 at 13 05 26"
src="https://github.com/user-attachments/assets/a0a8064e-8cfa-48b2-9f1c-efc2d0d9d7d4"
/>
We fixed this by only allowing interpolation if the user is editing
after all the edit predictions OR if the user edit is a subset of the
model suggestion.
Co-Authored-by: Antonio <antonio@zed.dev>
Release Notes:
- N/A
Co-authored-by: Antonio <antonio@zed.dev>
Fix a bug where a GPUI macro still used `ModelContext`
Rename `AsyncAppContext` -> `AsyncApp`
Rename update_model, read_model, insert_model, and reserve_model to
update_entity, read_entity, insert_entity, and reserve_entity
Release Notes:
- N/A
There's still a bit more work to do on this, but this PR is compiling
(with warnings) after eliminating the key types. When the tasks below
are complete, this will be the new narrative for GPUI:
- `Entity<T>` - This replaces `View<T>`/`Model<T>`. It represents a unit
of state, and if `T` implements `Render`, then `Entity<T>` implements
`Element`.
- `&mut App` This replaces `AppContext` and represents the app.
- `&mut Context<T>` This replaces `ModelContext` and derefs to `App`. It
is provided by the framework when updating an entity.
- `&mut Window` Broken out of `&mut WindowContext` which no longer
exists. Every method that once took `&mut WindowContext` now takes `&mut
Window, &mut App` and every method that took `&mut ViewContext<T>` now
takes `&mut Window, &mut Context<T>`
Not pictured here are the two other failed attempts. It's been quite a
month!
Tasks:
- [x] Remove `View`, `ViewContext`, `WindowContext` and thread through
`Window`
- [x] [@cole-miller @mikayla-maki] Redraw window when entities change
- [x] [@cole-miller @mikayla-maki] Get examples and Zed running
- [x] [@cole-miller @mikayla-maki] Fix Zed rendering
- [x] [@mikayla-maki] Fix todo! macros and comments
- [x] Fix a bug where the editor would not be redrawn because of view
caching
- [x] remove publicness window.notify() and replace with
`AppContext::notify`
- [x] remove `observe_new_window_models`, replace with
`observe_new_models` with an optional window
- [x] Fix a bug where the project panel would not be redrawn because of
the wrong refresh() call being used
- [x] Fix the tests
- [x] Fix warnings by eliminating `Window` params or using `_`
- [x] Fix conflicts
- [x] Simplify generic code where possible
- [x] Rename types
- [ ] Update docs
### issues post merge
- [x] Issues switching between normal and insert mode
- [x] Assistant re-rendering failure
- [x] Vim test failures
- [x] Mac build issue
Release Notes:
- N/A
---------
Co-authored-by: Antonio Scandurra <me@as-cii.com>
Co-authored-by: Cole Miller <cole@zed.dev>
Co-authored-by: Mikayla <mikayla@zed.dev>
Co-authored-by: Joseph <joseph@zed.dev>
Co-authored-by: max <max@zed.dev>
Co-authored-by: Michael Sloan <michael@zed.dev>
Co-authored-by: Mikayla Maki <mikaylamaki@Mikaylas-MacBook-Pro.local>
Co-authored-by: Mikayla <mikayla.c.maki@gmail.com>
Co-authored-by: joão <joao@zed.dev>
#23460 brought up we are showing the new "tab Accept" marker for single
line suggestions for non-zeta providers. We think this might be valid
for any provider, but we only want to enable it for zeta initially so it
doesn't affect an existing user base.
Release Notes:
- N/A
Two issues i ran into while looking at the completion rating modal
- Single-file worktrees file names are not displayed at all
- Hard to see the filename when the path is long (lots of directories)
This PR fixes this by displaying the filename on the left, followed by
the full path (including the worktree name), similar to how we do it in
the file finder/assistant panel /file command
| Before | After |
|--------|--------|
| <img width="1067" alt="Screenshot 2025-01-14 at 16 09 05"
src="https://github.com/user-attachments/assets/628fde18-da9a-4d98-8ddf-ed0ab0cd8d35"
/> | <img width="1161" alt="Screenshot 2025-01-14 at 16 17 52"
src="https://github.com/user-attachments/assets/80c6a4e1-065d-4b0a-b9c0-5f3391af4557"
/> |
Release Notes:
- N/A
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>
This PR also removes the `ThumbsUp` action that wasn't being triggered
correctly. We didn't have it's counterpart `ThumbsDown`, too, so I
mostly assumed it would be harmless to remove `ThumbsUp` as well.
<img width="800" alt="Screenshot 2025-01-10 at 6 18 44 PM"
src="https://github.com/user-attachments/assets/9fd5da9f-9dff-454d-9f31-c02f1370b937"
/>
Release Notes:
- N/A
---------
Co-authored-by: Marshall Bowers <elliott.codes@gmail.com>
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
@iamnbutler noticed slowness in assistant panel which I've pinned down
to the fact that we're calculating buffer diff on foreground thread.
This PR moves this computation into the background; I don't know much
about Zeta but it seems fine to do, as the call-site is asynchronous
anyways.
Closes #ISSUE
Release Notes:
- N/A
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>
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
- Adds the Switch component
- Updates `Selected`, `Selectable` -> `ToggleState`, `Toggleable`
- Adds `checkbox` and `switch` functions to align better with other
elements in our layout system.
We decided not to merge Switch and Checkbox. However, in a followup I'll
introduce a Toggle or AnyToggle enum so we can update
`CheckboxWithLabel` -> `ToggleWithLabel` as this component will work
exactly the same with either a Checkbox or a Switch.
Release Notes:
- N/A
Most notably, adding a title bar-ish in the left column as so to add the
"from most recent to oldest" info, which is supposed to make scanning
the list of completions easier to do (at least it would've helped me
figure out that was sorted that way when I was wondering about it!).
<img width="800" alt="Screenshot 2024-12-12 at 16 24 36"
src="https://github.com/user-attachments/assets/1acc9951-3df0-4cd2-96ff-94ed555ecae5"
/>
Release Notes:
- N/A
When a user types, chances are the model will anticipate what they are
about to do. Previously, we would continuously cancel the pending
completion until the user stopped typing. With this commit, we allow at
most two completions to race with each other (the first and the last
one):
- If the completion that was requested first completes first, we will
show it (assuming we can interpolate it) but avoid canceling the last
one.
- When the completion that was requested last completes, we will cancel
the first one if it's pending.
In both cases, if a completion is already on-screen we have a special
case for when the completions are just insertions and the new completion
is a superset of the existing one. In this case, we will replace the
existing completion with the new one. Otherwise we will keep showing the
old one to avoid thrashing the UI.
This should make latency a lot better. Note that I also reduced the
debounce timeout to 8ms.
Release Notes:
- N/A
Most notably, adding a current iteration of a possible logo to feel it
out! :) Also, I'm hiding the input and instructions container after the
review has been sent. In the future, if we allow changing an already
sent review, we can change this behavior.
<img width="800" alt="Screenshot 2024-12-12 at 10 42 44"
src="https://github.com/user-attachments/assets/37e63d0d-d847-445e-bdf8-bf5c97d0fe4c"
/>
Release Notes:
- N/A
João Marcos
Marshall Bowers
Michael Sloan
Bennet Bo Fenner
Mikayla Maki
Piotr Osiewicz
Joseph T. Lyons
Danilo Leal
Kirill Bulatov
Conrad Irwin
Nate Butler