Yehowshua/ZIm
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity Actions

Reuse conversation cache when streaming edits (#30245)

Browse source 
Release Notes:

- Improved latency when the agent applies edits.
This commit is contained in:
Antonio Scandurra 2025-05-08 14:36:34 +02:00 • committed by GitHub
parent 032022e37b
commit 9f6809a28d
No known key found for this signature in database
GPG key ID: B5690EEEBB952194
50 changed files with 847 additions and 21557 deletions
Download patch file Download diff file Expand all files Collapse all files

154
crates/assistant_tools/src/edit_agent/evals.rs
View file

@ -2,14 +2,16 @@ use super::*;
use crate::{ReadFileToolInput, edit_file_tool::EditFileToolInput, grep_tool::GrepToolInput};
use Role::*;
use anyhow::anyhow;
use assistant_tool::ToolRegistry;
use client::{Client, UserStore};
use collections::HashMap;
use fs::FakeFs;
use futures::{FutureExt, future::LocalBoxFuture};
use gpui::{AppContext, TestAppContext};
use indoc::indoc;
use indoc::{formatdoc, indoc};
use language_model::{
LanguageModelRegistry, LanguageModelToolResult, LanguageModelToolUse, LanguageModelToolUseId,
LanguageModelRegistry, LanguageModelRequestTool, LanguageModelToolResult, LanguageModelToolUse,
LanguageModelToolUseId,
};
use project::Project;
use rand::prelude::*;
@ -37,7 +39,7 @@ fn eval_extract_handle_command_output() {
conversation: vec![
message(
User,
[text(indoc! {"
[text(formatdoc! {"
Read the `{input_file_path}` file and extract a method in
the final stanza of `run_git_blame` to deal with command failures,
call it `handle_command_output` and take the std::process::Output as the only parameter.
@ -96,7 +98,7 @@ fn eval_delete_run_git_blame() {
conversation: vec![
message(
User,
[text(indoc! {"
[text(formatdoc! {"
Read the `{input_file_path}` file and delete `run_git_blame`. Just that
one function, not its usages.
"})],
@ -138,6 +140,61 @@ fn eval_delete_run_git_blame() {
);
}
#[test]
#[cfg_attr(not(feature = "eval"), ignore)]
fn eval_translate_doc_comments() {
let input_file_path = "root/canvas.rs";
let input_file_content = include_str!("evals/fixtures/translate_doc_comments/before.rs");
let edit_description = "Translate all doc comments to Italian";
eval(
200,
1.,
EvalInput {
conversation: vec![
message(
User,
[text(formatdoc! {"
Read the {input_file_path} file and edit it (without overwriting it),
translating all the doc comments to italian.
"})],
),
message(
Assistant,
[tool_use(
"tool_1",
"read_file",
ReadFileToolInput {
path: input_file_path.into(),
start_line: None,
end_line: None,
},
)],
),
message(
User,
[tool_result("tool_1", "read_file", input_file_content)],
),
message(
Assistant,
[tool_use(
"tool_2",
"edit_file",
EditFileToolInput {
display_description: edit_description.into(),
path: input_file_path.into(),
create_or_overwrite: false,
},
)],
),
],
input_path: input_file_path.into(),
input_content: Some(input_file_content.into()),
edit_description: edit_description.into(),
assertion: EvalAssertion::judge_diff("Doc comments were translated to Italian"),
},
);
}
#[test]
#[cfg_attr(not(feature = "eval"), ignore)]
fn eval_use_wasi_sdk_in_compile_parser_to_wasm() {
@ -152,7 +209,7 @@ fn eval_use_wasi_sdk_in_compile_parser_to_wasm() {
conversation: vec![
message(
User,
[text(indoc! {"
[text(formatdoc! {"
Read the `{input_file_path}` file and change `compile_parser_to_wasm` to use `wasi-sdk` instead of emscripten.
Use `ureq` to download the SDK for the current platform and architecture.
Extract the archive into a sibling of `lib` inside the `tree-sitter` directory in the cache_dir.
@ -160,7 +217,7 @@ fn eval_use_wasi_sdk_in_compile_parser_to_wasm() {
that's inside of the archive.
Don't re-download the SDK if that executable already exists.
Use these clang flags: -fPIC -shared -Os -Wl,--export=tree_sitter_{language_name}
Use these clang flags: -fPIC -shared -Os -Wl,--export=tree_sitter_{{language_name}}
Here are the available wasi-sdk assets:
- wasi-sdk-25.0-x86_64-macos.tar.gz
@ -261,11 +318,10 @@ fn eval_use_wasi_sdk_in_compile_parser_to_wasm() {
fn eval_disable_cursor_blinking() {
let input_file_path = "root/editor.rs";
let input_file_content = include_str!("evals/fixtures/disable_cursor_blinking/before.rs");
let output_file_content = include_str!("evals/fixtures/disable_cursor_blinking/after.rs");
let edit_description = "Comment out the call to `BlinkManager::enable`";
eval(
200,
0.6, // TODO: make this eval better
0.95,
EvalInput {
conversation: vec![
message(User, [text("Let's research how to cursor blinking works.")]),
@ -324,7 +380,11 @@ fn eval_disable_cursor_blinking() {
input_path: input_file_path.into(),
input_content: Some(input_file_content.into()),
edit_description: edit_description.into(),
assertion: EvalAssertion::assert_eq(output_file_content),
assertion: EvalAssertion::judge_diff(indoc! {"
- Calls to BlinkManager in `observe_window_activation` were commented out
- The call to `blink_manager.enable` above the call to show_cursor_names was commented out
- All the edits have valid indentation
"}),
},
);
}
@ -1031,7 +1091,8 @@ impl EvalAssertion {
fn eval(iterations: usize, expected_pass_ratio: f32, mut eval: EvalInput) {
let mut evaluated_count = 0;
report_progress(evaluated_count, iterations);
let mut failed_count = 0;
report_progress(evaluated_count, failed_count, iterations);
let (tx, rx) = mpsc::channel();
@ -1048,7 +1109,6 @@ fn eval(iterations: usize, expected_pass_ratio: f32, mut eval: EvalInput) {
}
drop(tx);
let mut failed_count = 0;
let mut failed_evals = HashMap::default();
let mut errored_evals = HashMap::default();
let mut eval_outputs = Vec::new();
@ -1073,7 +1133,7 @@ fn eval(iterations: usize, expected_pass_ratio: f32, mut eval: EvalInput) {
}
evaluated_count += 1;
report_progress(evaluated_count, iterations);
report_progress(evaluated_count, failed_count, iterations);
}
let actual_pass_ratio = (iterations - failed_count) as f32 / iterations as f32;
@ -1144,8 +1204,19 @@ impl Display for EvalOutput {
}
}
fn report_progress(evaluated_count: usize, iterations: usize) {
print!("\r\x1b[KEvaluated {}/{}", evaluated_count, iterations);
fn report_progress(evaluated_count: usize, failed_count: usize, iterations: usize) {
let passed_count = evaluated_count - failed_count;
let passed_ratio = if evaluated_count == 0 {
0.0
} else {
passed_count as f64 / evaluated_count as f64
};
print!(
"\r\x1b[KEvaluated {}/{} ({:.2}%)",
evaluated_count,
iterations,
passed_ratio * 100.0
);
std::io::stdout().flush().unwrap();
}
@ -1158,25 +1229,30 @@ struct EditAgentTest {
impl EditAgentTest {
async fn new(cx: &mut TestAppContext) -> Self {
cx.executor().allow_parking();
cx.update(settings::init);
cx.update(Project::init_settings);
cx.update(language::init);
cx.update(gpui_tokio::init);
cx.update(client::init_settings);
let fs = FakeFs::new(cx.executor().clone());
cx.update(|cx| {
settings::init(cx);
gpui_tokio::init(cx);
let http_client = Arc::new(ReqwestClient::user_agent("agent tests").unwrap());
cx.set_http_client(http_client);
client::init_settings(cx);
let client = Client::production(cx);
let user_store = cx.new(|cx| UserStore::new(client.clone(), cx));
settings::init(cx);
Project::init_settings(cx);
language::init(cx);
language_model::init(client.clone(), cx);
language_models::init(user_store.clone(), client.clone(), fs.clone(), cx);
crate::init(client.http_client(), cx);
});
fs.insert_tree("/root", json!({})).await;
let project = Project::test(fs.clone(), [path!("/root").as_ref()], cx).await;
let (agent_model, judge_model) = cx
.update(|cx| {
let http_client = ReqwestClient::user_agent("agent tests").unwrap();
cx.set_http_client(Arc::new(http_client));
let client = Client::production(cx);
let user_store = cx.new(|cx| UserStore::new(client.clone(), cx));
language_model::init(client.clone(), cx);
language_models::init(user_store.clone(), client.clone(), fs.clone(), cx);
cx.spawn(async move |cx| {
let agent_model =
Self::load_model("anthropic", "claude-3-7-sonnet-latest", cx).await;
@ -1225,12 +1301,32 @@ impl EditAgentTest {
.update(cx, |project, cx| project.open_buffer(path, cx))
.await
.unwrap();
let conversation = LanguageModelRequest {
messages: eval.conversation,
tools: cx.update(|cx| {
ToolRegistry::default_global(cx)
.tools()
.into_iter()
.filter_map(|tool| {
let input_schema = tool
.input_schema(self.agent.model.tool_input_format())
.ok()?;
Some(LanguageModelRequestTool {
name: tool.name(),
description: tool.description(),
input_schema,
})
})
.collect()
}),
..Default::default()
};
let edit_output = if let Some(input_content) = eval.input_content.as_deref() {
buffer.update(cx, |buffer, cx| buffer.set_text(input_content, cx));
let (edit_output, _) = self.agent.edit(
buffer.clone(),
eval.edit_description,
eval.conversation,
&conversation,
&mut cx.to_async(),
);
edit_output.await?
@ -1238,7 +1334,7 @@ impl EditAgentTest {
let (edit_output, _) = self.agent.overwrite(
buffer.clone(),
eval.edit_description,
eval.conversation,
&conversation,
&mut cx.to_async(),
);
edit_output.await?

21343
crates/assistant_tools/src/edit_agent/evals/fixtures/disable_cursor_blinking/after.rs
View file

File diff suppressed because it is too large Load diff

339
crates/assistant_tools/src/edit_agent/evals/fixtures/translate_doc_comments/before.rs Normal file
View file

@ -0,0 +1,339 @@
// font-kit/src/canvas.rs
//
// Copyright © 2018 The Pathfinder Project Developers.
//
// Licensed under the Apache License, Version 2.0 <LICENSE-APACHE or
// http://www.apache.org/licenses/LICENSE-2.0> or the MIT license
// <LICENSE-MIT or http://opensource.org/licenses/MIT>, at your
// option. This file may not be copied, modified, or distributed
// except according to those terms.
//! An in-memory bitmap surface for glyph rasterization.
use lazy_static::lazy_static;
use pathfinder_geometry::rect::RectI;
use pathfinder_geometry::vector::Vector2I;
use std::cmp;
use std::fmt;
use crate::utils;
lazy_static! {
static ref BITMAP_1BPP_TO_8BPP_LUT: [[u8; 8]; 256] = {
let mut lut = [[0; 8]; 256];
for byte in 0..0x100 {
let mut value = [0; 8];
for bit in 0..8 {
if (byte & (0x80 >> bit)) != 0 {
value[bit] = 0xff;
}
}
lut[byte] = value
}
lut
};
}
/// An in-memory bitmap surface for glyph rasterization.
pub struct Canvas {
/// The raw pixel data.
pub pixels: Vec<u8>,
/// The size of the buffer, in pixels.
pub size: Vector2I,
/// The number of *bytes* between successive rows.
pub stride: usize,
/// The image format of the canvas.
pub format: Format,
}
impl Canvas {
/// Creates a new blank canvas with the given pixel size and format.
///
/// Stride is automatically calculated from width.
///
/// The canvas is initialized with transparent black (all values 0).
#[inline]
pub fn new(size: Vector2I, format: Format) -> Canvas {
Canvas::with_stride(
size,
size.x() as usize * format.bytes_per_pixel() as usize,
format,
)
}
/// Creates a new blank canvas with the given pixel size, stride (number of bytes between
/// successive rows), and format.
///
/// The canvas is initialized with transparent black (all values 0).
pub fn with_stride(size: Vector2I, stride: usize, format: Format) -> Canvas {
Canvas {
pixels: vec![0; stride * size.y() as usize],
size,
stride,
format,
}
}
#[allow(dead_code)]
pub(crate) fn blit_from_canvas(&mut self, src: &Canvas) {
self.blit_from(
Vector2I::default(),
&src.pixels,
src.size,
src.stride,
src.format,
)
}
/// Blits to a rectangle with origin at `dst_point` and size according to `src_size`.
/// If the target area overlaps the boundaries of the canvas, only the drawable region is blitted.
/// `dst_point` and `src_size` are specified in pixels. `src_stride` is specified in bytes.
/// `src_stride` must be equal or larger than the actual data length.
#[allow(dead_code)]
pub(crate) fn blit_from(
&mut self,
dst_point: Vector2I,
src_bytes: &[u8],
src_size: Vector2I,
src_stride: usize,
src_format: Format,
) {
assert_eq!(
src_stride * src_size.y() as usize,
src_bytes.len(),
"Number of pixels in src_bytes does not match stride and size."
);
assert!(
src_stride >= src_size.x() as usize * src_format.bytes_per_pixel() as usize,
"src_stride must be >= than src_size.x()"
);
let dst_rect = RectI::new(dst_point, src_size);
let dst_rect = dst_rect.intersection(RectI::new(Vector2I::default(), self.size));
let dst_rect = match dst_rect {
Some(dst_rect) => dst_rect,
None => return,
};
match (self.format, src_format) {
(Format::A8, Format::A8)
| (Format::Rgb24, Format::Rgb24)
| (Format::Rgba32, Format::Rgba32) => {
self.blit_from_with::<BlitMemcpy>(dst_rect, src_bytes, src_stride, src_format)
}
(Format::A8, Format::Rgb24) => {
self.blit_from_with::<BlitRgb24ToA8>(dst_rect, src_bytes, src_stride, src_format)
}
(Format::Rgb24, Format::A8) => {
self.blit_from_with::<BlitA8ToRgb24>(dst_rect, src_bytes, src_stride, src_format)
}
(Format::Rgb24, Format::Rgba32) => self
.blit_from_with::<BlitRgba32ToRgb24>(dst_rect, src_bytes, src_stride, src_format),
(Format::Rgba32, Format::Rgb24) => self
.blit_from_with::<BlitRgb24ToRgba32>(dst_rect, src_bytes, src_stride, src_format),
(Format::Rgba32, Format::A8) | (Format::A8, Format::Rgba32) => unimplemented!(),
}
}
#[allow(dead_code)]
pub(crate) fn blit_from_bitmap_1bpp(
&mut self,
dst_point: Vector2I,
src_bytes: &[u8],
src_size: Vector2I,
src_stride: usize,
) {
if self.format != Format::A8 {
unimplemented!()
}
let dst_rect = RectI::new(dst_point, src_size);
let dst_rect = dst_rect.intersection(RectI::new(Vector2I::default(), self.size));
let dst_rect = match dst_rect {
Some(dst_rect) => dst_rect,
None => return,
};
let size = dst_rect.size();
let dest_bytes_per_pixel = self.format.bytes_per_pixel() as usize;
let dest_row_stride = size.x() as usize * dest_bytes_per_pixel;
let src_row_stride = utils::div_round_up(size.x() as usize, 8);
for y in 0..size.y() {
let (dest_row_start, src_row_start) = (
(y + dst_rect.origin_y()) as usize * self.stride
+ dst_rect.origin_x() as usize * dest_bytes_per_pixel,
y as usize * src_stride,
);
let dest_row_end = dest_row_start + dest_row_stride;
let src_row_end = src_row_start + src_row_stride;
let dest_row_pixels = &mut self.pixels[dest_row_start..dest_row_end];
let src_row_pixels = &src_bytes[src_row_start..src_row_end];
for x in 0..src_row_stride {
let pattern = &BITMAP_1BPP_TO_8BPP_LUT[src_row_pixels[x] as usize];
let dest_start = x * 8;
let dest_end = cmp::min(dest_start + 8, dest_row_stride);
let src = &pattern[0..(dest_end - dest_start)];
dest_row_pixels[dest_start..dest_end].clone_from_slice(src);
}
}
}
/// Blits to area `rect` using the data given in the buffer `src_bytes`.
/// `src_stride` must be specified in bytes.
/// The dimensions of `rect` must be in pixels.
fn blit_from_with<B: Blit>(
&mut self,
rect: RectI,
src_bytes: &[u8],
src_stride: usize,
src_format: Format,
) {
let src_bytes_per_pixel = src_format.bytes_per_pixel() as usize;
let dest_bytes_per_pixel = self.format.bytes_per_pixel() as usize;
for y in 0..rect.height() {
let (dest_row_start, src_row_start) = (
(y + rect.origin_y()) as usize * self.stride
+ rect.origin_x() as usize * dest_bytes_per_pixel,
y as usize * src_stride,
);
let dest_row_end = dest_row_start + rect.width() as usize * dest_bytes_per_pixel;
let src_row_end = src_row_start + rect.width() as usize * src_bytes_per_pixel;
let dest_row_pixels = &mut self.pixels[dest_row_start..dest_row_end];
let src_row_pixels = &src_bytes[src_row_start..src_row_end];
B::blit(dest_row_pixels, src_row_pixels)
}
}
}
impl fmt::Debug for Canvas {
fn fmt(&self, f: &mut fmt::Formatter) -> fmt::Result {
f.debug_struct("Canvas")
.field("pixels", &self.pixels.len()) // Do not dump a vector content.
.field("size", &self.size)
.field("stride", &self.stride)
.field("format", &self.format)
.finish()
}
}
/// The image format for the canvas.
#[derive(Clone, Copy, Debug, PartialEq)]
pub enum Format {
/// Premultiplied R8G8B8A8, little-endian.
Rgba32,
/// R8G8B8, little-endian.
Rgb24,
/// A8.
A8,
}
impl Format {
/// Returns the number of bits per pixel that this image format corresponds to.
#[inline]
pub fn bits_per_pixel(self) -> u8 {
match self {
Format::Rgba32 => 32,
Format::Rgb24 => 24,
Format::A8 => 8,
}
}
/// Returns the number of color channels per pixel that this image format corresponds to.
#[inline]
pub fn components_per_pixel(self) -> u8 {
match self {
Format::Rgba32 => 4,
Format::Rgb24 => 3,
Format::A8 => 1,
}
}
/// Returns the number of bits per color channel that this image format contains.
#[inline]
pub fn bits_per_component(self) -> u8 {
self.bits_per_pixel() / self.components_per_pixel()
}
/// Returns the number of bytes per pixel that this image format corresponds to.
#[inline]
pub fn bytes_per_pixel(self) -> u8 {
self.bits_per_pixel() / 8
}
}
/// The antialiasing strategy that should be used when rasterizing glyphs.
#[derive(Clone, Copy, Debug, PartialEq)]
pub enum RasterizationOptions {
/// "Black-and-white" rendering. Each pixel is either entirely on or off.
Bilevel,
/// Grayscale antialiasing. Only one channel is used.
GrayscaleAa,
/// Subpixel RGB antialiasing, for LCD screens.
SubpixelAa,
}
trait Blit {
fn blit(dest: &mut [u8], src: &[u8]);
}
struct BlitMemcpy;
impl Blit for BlitMemcpy {
#[inline]
fn blit(dest: &mut [u8], src: &[u8]) {
dest.clone_from_slice(src)
}
}
struct BlitRgb24ToA8;
impl Blit for BlitRgb24ToA8 {
#[inline]
fn blit(dest: &mut [u8], src: &[u8]) {
// TODO(pcwalton): SIMD.
for (dest, src) in dest.iter_mut().zip(src.chunks(3)) {
*dest = src[1]
}
}
}
struct BlitA8ToRgb24;
impl Blit for BlitA8ToRgb24 {
#[inline]
fn blit(dest: &mut [u8], src: &[u8]) {
for (dest, src) in dest.chunks_mut(3).zip(src.iter()) {
dest[0] = *src;
dest[1] = *src;
dest[2] = *src;
}
}
}
struct BlitRgba32ToRgb24;
impl Blit for BlitRgba32ToRgb24 {
#[inline]
fn blit(dest: &mut [u8], src: &[u8]) {
// TODO(pcwalton): SIMD.
for (dest, src) in dest.chunks_mut(3).zip(src.chunks(4)) {
dest.copy_from_slice(&src[0..3])
}
}
}
struct BlitRgb24ToRgba32;
impl Blit for BlitRgb24ToRgba32 {
fn blit(dest: &mut [u8], src: &[u8]) {
for (dest, src) in dest.chunks_mut(4).zip(src.chunks(3)) {
dest[0] = src[0];
dest[1] = src[1];
dest[2] = src[2];
dest[3] = 255;
}
}
}