Co-authored-by: Antonio <antonio@zed.dev>
Resurrected this from some assistant work I did in Spring of 2023.
- [x] Resurrect streaming responses
- [x] Use streaming responses to enable AI via Zed's servers by default
(but preserve API key option for now)
- [x] Simplify protobuf
- [x] Proxy to OpenAI on zed.dev
- [x] Proxy to Gemini on zed.dev
- [x] Improve UX for switching between openAI and google models
- We current disallow cycling when setting a custom model, but we need a
better solution to keep OpenAI models available while testing the google
ones
- [x] Show remaining tokens correctly for Google models
- [x] Remove semantic index
- [x] Delete `ai` crate
- [x] Cloud front so we can ban abuse
- [x] Rate-limiting
- [x] Fix panic when using inline assistant
- [x] Double check the upgraded `AssistantSettings` are
backwards-compatible
- [x] Add hosted LLM interaction behind a `language-models` feature
flag.
Release Notes:
- We are temporarily removing the semantic index in order to redesign it
from scratch.
---------
Co-authored-by: Antonio <antonio@zed.dev>
Co-authored-by: Antonio Scandurra <me@as-cii.com>
Co-authored-by: Thorsten <thorsten@zed.dev>
Co-authored-by: Max <max@zed.dev>
This practice makes it difficult to locate todo!s in my code when I'm
working. Let's take out the bang if we want to keep doing this.
Release Notes:
- N/A
I don't want to clutter this post with the asset source, so I made it
optional. I'm also checking in an example which helps me ensure my code
examples are valid.
Release Notes:
- N/A
Ensure we `panic()` instead of crash on graphics memory buffer overflow
Also bump the buffer size to 32Mb from 8Mb to make this rarer (but still
possible)
Release Notes: Fixes some crahes due to lack of graphics buffer spacae
Since the message headers are buttons, we need to shift them relatively to
compensate for the fact that the background is only visible when hovered.
I'm ok with the background not being aligned so long as the unhovered text is.
- Returning `IntoElement` because it's more universal than `Element` and
allows us to easily return components.
- Using `impl IntoElement` in `RenderOnce` for consistency, which
requires `Component` to allocate an `AnyElement`. We use a bump
allocated arena anyway for these, and my benchmark doesn't show an
impact.
In this histogram, frames-0 is this PR, frames-1 is main.
Release Notes:
- N/A
Previously, we changed borders to be drawn after content, so they are no
longer part of the same quads as the background. In our change, we gave
the background quad a transparent black border and the border quads
transparent black backgrounds. However, this caused the other channels
to blend toward that black color before becoming fully transparent,
causing them to become darker.
In this PR, I source the "placeholder" color by duplicating the values
for the other channels and only adjust the alpha down to zero.
Release Notes:
- N/A
This makes it a bit more approachable to implement.
This required converting the state of Component to AnyElement, which
costs an allocation. I'm hoping this will be ok performance-wise now
that AnyElements use bump allocation, but I need to benchmark still.
Getting some scaffold docs in place and making some naming adjustments
to improve consistency with Rust APIs and clarity.
- Render::Element -> Render::Output, and Render::Output is now only
required to be IntoElement instead of Element so we can return
components.
- RenderOnce::Element -> RenderOnce::Output
- Element::layout -> Render::request_layout - Clarify that this doesn't
actually *perform* layout, but simply requests
Release Notes:
- N/A
Previously, we changed borders to be drawn after content, so they are no longer
part of the same quads as the background. In our change, we gave the background
quad a transparent black border and the border quads transparent black
backgrounds. However, this caused the other channels to blend toward that black
color before becoming fully transparent, causing them to become darker.
In this PR, I source the "placeholder" color by duplicating the values for
the other channels and only adjust the alpha down to zero.
This prevents the `cmd-k` keystroke, which clears the terminal, from
staying around as a pending keystroke on the parent workspace.
Release Notes:
- N/A
Rather than relying on the focused element, instead explicitly pass the
focus handle for the query editor when determining the prev/next
bindings. Only compute these values once.
Release Notes:
- N/A
Once a drag starts, we won't fire click listeners or style any elements
as active.
- Don't fire click listeners or show active state once a drag is in
progress
- Don't show hover style when a drag is in progress
- Draw borders above content
- If borders are opaque, apply them to the content mask. This prevents
hovers from firing on content underneath the border, which was creating
issues where the drag handle was inside the border, so we'd flicker the
hover when the mouse moved out of the drag handle and into the 1px
border on the left dock.
- Add a `block_mouse` helper which causes transparent elements to paint
an "opaque" layer to prevent mouse events from falling through. We use
this for the drag handle as well to disable hover, click, etc on items
in the panel.
Release Notes:
- N/A
This allows the content mask to correctly apply to bounds used in event handlers,
which prevents content under opaque borders from being hovered in overflow hidden
containers.
Co-Authored-By: Antonio <antonio@zed.dev>
Co-Authored-By: Max Brunsfeld <max@zed.dev>
This PR binds the `workspace::NewFile` action only in the `Workspace`
context. It previously was bound globally, which caused us to hijack the
`cmd-n` binding when the context menu was visible in the project panel.
In the process, we also added some other global workspace actions and
move some other global bindings to be bound only on `Pane` and
`Workspace`.
Release Notes:
- N/A
Without this, hitting cmd-n on the context menu in the project browser
invokes the workspace::NewFile action instead of the project::NewFile
action. We're considering changing the behavior so that bindings with no
context can only invoke global actions.
Co-Authored-By: Max <max@zed.dev>
ShapedLines are never wrapped, whereas WrappedLines are optionally
wrapped if they are associated with a wrap width. Originally, when
rewriting GPUI, I tried to combine everything because wrapping is
inherently optional for the Text element, but we have a bunch of APIs
that don't make sense on a line that may wrap, so we need a distinct
type for that case.
This is a precursor to implementing clickable links in markdown. I
noticed multiple places where we were confused about whether or not the
line was wrapped so this felt important.
Release Notes:
- N/A
ShapedLines are never wrapped, whereas WrappedLines are optionally wrapped if
they are associated with a wrap width. I tried to combine everything because
wrapping is inherently optional for the Text element, but we have a bunch of
APIs that don't make sense on a line that may wrap, so we need a distinct type
for that case.
Initially, we imagined registering keyboard handlers in the initialize
phase so we would understand the relationships between focus handles
during the layout pass, which would allow us to assign assign `focus_in`
styles that impact layout.
However, we soon realized that many elements aren't created until paint
time anyway, such as within the uniform list. Since it's impossible to
know prior to paint whether an element contains the focused element, it
makes more sense to eliminate the `focus_in` styling helper.
Release Notes:
- N/A
We need to notify when we set the pending mouse down so we attach the
mouse up event listener before the mouse button is released.
Release Notes:
- N/A
Nathan Sobo
