edit_file: Add diff-fenced output format (#32737)

This format is enabled for Google models as they seem to prefer it.
A relevant unit eval's pass rate has increased from 0.77 to 0.98.

Diff-fenced format looks like this (markdown fences and a line hint are
optional):

```diff
<<<<<<< SEARCH line=42
...
=======
...
>>>>>>> REPLACE
```

Release Notes:

- Agent: Gemini models now use the diff-fenced format when making edits
This commit is contained in:
Oleksiy Syvokon 2025-06-16 17:28:18 +03:00 committed by GitHub
parent 8df6ce2aac
commit fceba6c795
No known key found for this signature in database
GPG key ID: B5690EEEBB952194
8 changed files with 667 additions and 79 deletions

View file

@ -8,6 +8,7 @@ use crate::{Template, Templates};
use anyhow::Result;
use assistant_tool::ActionLog;
use create_file_parser::{CreateFileParser, CreateFileParserEvent};
pub use edit_parser::EditFormat;
use edit_parser::{EditParser, EditParserEvent, EditParserMetrics};
use futures::{
Stream, StreamExt,
@ -41,13 +42,23 @@ impl Template for CreateFilePromptTemplate {
}
#[derive(Serialize)]
struct EditFilePromptTemplate {
struct EditFileXmlPromptTemplate {
path: Option<PathBuf>,
edit_description: String,
}
impl Template for EditFilePromptTemplate {
const TEMPLATE_NAME: &'static str = "edit_file_prompt.hbs";
impl Template for EditFileXmlPromptTemplate {
const TEMPLATE_NAME: &'static str = "edit_file_prompt_xml.hbs";
}
#[derive(Serialize)]
struct EditFileDiffFencedPromptTemplate {
path: Option<PathBuf>,
edit_description: String,
}
impl Template for EditFileDiffFencedPromptTemplate {
const TEMPLATE_NAME: &'static str = "edit_file_prompt_diff_fenced.hbs";
}
#[derive(Clone, Debug, PartialEq, Eq)]
@ -70,6 +81,7 @@ pub struct EditAgent {
action_log: Entity<ActionLog>,
project: Entity<Project>,
templates: Arc<Templates>,
edit_format: EditFormat,
}
impl EditAgent {
@ -78,12 +90,14 @@ impl EditAgent {
project: Entity<Project>,
action_log: Entity<ActionLog>,
templates: Arc<Templates>,
edit_format: EditFormat,
) -> Self {
EditAgent {
model,
project,
action_log,
templates,
edit_format,
}
}
@ -209,14 +223,23 @@ impl EditAgent {
let this = self.clone();
let (events_tx, events_rx) = mpsc::unbounded();
let conversation = conversation.clone();
let edit_format = self.edit_format;
let output = cx.spawn(async move |cx| {
let snapshot = buffer.read_with(cx, |buffer, _| buffer.snapshot())?;
let path = cx.update(|cx| snapshot.resolve_file_path(cx, true))?;
let prompt = EditFilePromptTemplate {
path,
edit_description,
}
.render(&this.templates)?;
let prompt = match edit_format {
EditFormat::XmlTags => EditFileXmlPromptTemplate {
path,
edit_description,
}
.render(&this.templates)?,
EditFormat::DiffFenced => EditFileDiffFencedPromptTemplate {
path,
edit_description,
}
.render(&this.templates)?,
};
let edit_chunks = this
.request(conversation, CompletionIntent::EditFile, prompt, cx)
.await?;
@ -236,7 +259,7 @@ impl EditAgent {
self.action_log
.update(cx, |log, cx| log.buffer_read(buffer.clone(), cx))?;
let (output, edit_events) = Self::parse_edit_chunks(edit_chunks, cx);
let (output, edit_events) = Self::parse_edit_chunks(edit_chunks, self.edit_format, cx);
let mut edit_events = edit_events.peekable();
while let Some(edit_event) = Pin::new(&mut edit_events).peek().await {
// Skip events until we're at the start of a new edit.
@ -350,6 +373,7 @@ impl EditAgent {
fn parse_edit_chunks(
chunks: impl 'static + Send + Stream<Item = Result<String, LanguageModelCompletionError>>,
edit_format: EditFormat,
cx: &mut AsyncApp,
) -> (
Task<Result<EditAgentOutput>>,
@ -359,7 +383,7 @@ impl EditAgent {
let output = cx.background_spawn(async move {
pin_mut!(chunks);
let mut parser = EditParser::new();
let mut parser = EditParser::new(edit_format);
let mut raw_edits = String::new();
while let Some(chunk) = chunks.next().await {
match chunk {
@ -1355,7 +1379,13 @@ mod tests {
let project = Project::test(FakeFs::new(cx.executor()), [], cx).await;
let model = Arc::new(FakeLanguageModel::default());
let action_log = cx.new(|_| ActionLog::new(project.clone()));
EditAgent::new(model, project, action_log, Templates::new())
EditAgent::new(
model,
project,
action_log,
Templates::new(),
EditFormat::XmlTags,
)
}
#[gpui::test(iterations = 10)]

View file

@ -1,13 +1,18 @@
use anyhow::bail;
use derive_more::{Add, AddAssign};
use language_model::LanguageModel;
use regex::Regex;
use schemars::JsonSchema;
use serde::{Deserialize, Serialize};
use smallvec::SmallVec;
use std::{mem, ops::Range};
use std::{mem, ops::Range, str::FromStr, sync::Arc};
const OLD_TEXT_END_TAG: &str = "</old_text>";
const NEW_TEXT_END_TAG: &str = "</new_text>";
const EDITS_END_TAG: &str = "</edits>";
const SEARCH_MARKER: &str = "<<<<<<< SEARCH";
const SEPARATOR_MARKER: &str = "=======";
const REPLACE_MARKER: &str = ">>>>>>> REPLACE";
const END_TAGS: [&str; 3] = [OLD_TEXT_END_TAG, NEW_TEXT_END_TAG, EDITS_END_TAG];
#[derive(Debug)]
@ -31,44 +36,153 @@ pub struct EditParserMetrics {
pub mismatched_tags: usize,
}
#[derive(Debug, Clone, Copy, PartialEq, Eq, Serialize, Deserialize)]
#[serde(rename_all = "snake_case")]
pub enum EditFormat {
/// XML-like tags:
/// <old_text>...</old_text>
/// <new_text>...</new_text>
XmlTags,
/// Diff-fenced format, in which:
/// - Text before the SEARCH marker is ignored
/// - Fences are optional
/// - Line hint is optional.
///
/// Example:
///
/// ```diff
/// <<<<<<< SEARCH line=42
/// ...
/// =======
/// ...
/// >>>>>>> REPLACE
/// ```
DiffFenced,
}
impl FromStr for EditFormat {
type Err = anyhow::Error;
fn from_str(s: &str) -> anyhow::Result<Self> {
match s.to_lowercase().as_str() {
"xml_tags" | "xml" => Ok(EditFormat::XmlTags),
"diff_fenced" | "diff-fenced" | "diff" => Ok(EditFormat::DiffFenced),
_ => bail!("Unknown EditFormat: {}", s),
}
}
}
impl EditFormat {
/// Return an optimal edit format for the language model
pub fn from_model(model: Arc<dyn LanguageModel>) -> anyhow::Result<Self> {
if model.provider_id().0 == "google" {
Ok(EditFormat::DiffFenced)
} else {
Ok(EditFormat::XmlTags)
}
}
/// Return an optimal edit format for the language model,
/// with the ability to override it by setting the
/// `ZED_EDIT_FORMAT` environment variable
#[allow(dead_code)]
pub fn from_env(model: Arc<dyn LanguageModel>) -> anyhow::Result<Self> {
let default = EditFormat::from_model(model)?;
std::env::var("ZED_EDIT_FORMAT").map_or(Ok(default), |s| EditFormat::from_str(&s))
}
}
pub trait EditFormatParser: Send + std::fmt::Debug {
fn push(&mut self, chunk: &str) -> SmallVec<[EditParserEvent; 1]>;
fn take_metrics(&mut self) -> EditParserMetrics;
}
#[derive(Debug)]
pub struct EditParser {
state: EditParserState,
pub struct XmlEditParser {
state: XmlParserState,
buffer: String,
metrics: EditParserMetrics,
}
#[derive(Debug, PartialEq)]
enum EditParserState {
enum XmlParserState {
Pending,
WithinOldText { start: bool, line_hint: Option<u32> },
AfterOldText,
WithinNewText { start: bool },
}
impl EditParser {
#[derive(Debug)]
pub struct DiffFencedEditParser {
state: DiffParserState,
buffer: String,
metrics: EditParserMetrics,
}
#[derive(Debug, PartialEq)]
enum DiffParserState {
Pending,
WithinSearch { start: bool, line_hint: Option<u32> },
WithinReplace { start: bool },
}
/// Main parser that delegates to format-specific parsers
pub struct EditParser {
parser: Box<dyn EditFormatParser>,
}
impl XmlEditParser {
pub fn new() -> Self {
EditParser {
state: EditParserState::Pending,
XmlEditParser {
state: XmlParserState::Pending,
buffer: String::new(),
metrics: EditParserMetrics::default(),
}
}
pub fn push(&mut self, chunk: &str) -> SmallVec<[EditParserEvent; 1]> {
fn find_end_tag(&self) -> Option<Range<usize>> {
let (tag, start_ix) = END_TAGS
.iter()
.flat_map(|tag| Some((tag, self.buffer.find(tag)?)))
.min_by_key(|(_, ix)| *ix)?;
Some(start_ix..start_ix + tag.len())
}
fn ends_with_tag_prefix(&self) -> bool {
let mut end_prefixes = END_TAGS
.iter()
.flat_map(|tag| (1..tag.len()).map(move |i| &tag[..i]))
.chain(["\n"]);
end_prefixes.any(|prefix| self.buffer.ends_with(&prefix))
}
fn parse_line_hint(&self, tag: &str) -> Option<u32> {
use std::sync::LazyLock;
static LINE_HINT_REGEX: LazyLock<Regex> =
LazyLock::new(|| Regex::new(r#"line=(?:"?)(\d+)"#).unwrap());
LINE_HINT_REGEX
.captures(tag)
.and_then(|caps| caps.get(1))
.and_then(|m| m.as_str().parse::<u32>().ok())
}
}
impl EditFormatParser for XmlEditParser {
fn push(&mut self, chunk: &str) -> SmallVec<[EditParserEvent; 1]> {
self.buffer.push_str(chunk);
let mut edit_events = SmallVec::new();
loop {
match &mut self.state {
EditParserState::Pending => {
XmlParserState::Pending => {
if let Some(start) = self.buffer.find("<old_text") {
if let Some(tag_end) = self.buffer[start..].find('>') {
let tag_end = start + tag_end + 1;
let tag = &self.buffer[start..tag_end];
let line_hint = self.parse_line_hint(tag);
self.buffer.drain(..tag_end);
self.state = EditParserState::WithinOldText {
self.state = XmlParserState::WithinOldText {
start: true,
line_hint,
};
@ -79,7 +193,7 @@ impl EditParser {
break;
}
}
EditParserState::WithinOldText { start, line_hint } => {
XmlParserState::WithinOldText { start, line_hint } => {
if !self.buffer.is_empty() {
if *start && self.buffer.starts_with('\n') {
self.buffer.remove(0);
@ -100,7 +214,7 @@ impl EditParser {
}
self.buffer.drain(..tag_range.end);
self.state = EditParserState::AfterOldText;
self.state = XmlParserState::AfterOldText;
edit_events.push(EditParserEvent::OldTextChunk {
chunk,
done: true,
@ -117,15 +231,15 @@ impl EditParser {
break;
}
}
EditParserState::AfterOldText => {
XmlParserState::AfterOldText => {
if let Some(start) = self.buffer.find("<new_text>") {
self.buffer.drain(..start + "<new_text>".len());
self.state = EditParserState::WithinNewText { start: true };
self.state = XmlParserState::WithinNewText { start: true };
} else {
break;
}
}
EditParserState::WithinNewText { start } => {
XmlParserState::WithinNewText { start } => {
if !self.buffer.is_empty() {
if *start && self.buffer.starts_with('\n') {
self.buffer.remove(0);
@ -145,7 +259,7 @@ impl EditParser {
}
self.buffer.drain(..tag_range.end);
self.state = EditParserState::Pending;
self.state = XmlParserState::Pending;
edit_events.push(EditParserEvent::NewTextChunk { chunk, done: true });
} else {
if !self.ends_with_tag_prefix() {
@ -162,34 +276,163 @@ impl EditParser {
edit_events
}
fn find_end_tag(&self) -> Option<Range<usize>> {
let (tag, start_ix) = END_TAGS
.iter()
.flat_map(|tag| Some((tag, self.buffer.find(tag)?)))
.min_by_key(|(_, ix)| *ix)?;
Some(start_ix..start_ix + tag.len())
fn take_metrics(&mut self) -> EditParserMetrics {
std::mem::take(&mut self.metrics)
}
}
impl DiffFencedEditParser {
pub fn new() -> Self {
DiffFencedEditParser {
state: DiffParserState::Pending,
buffer: String::new(),
metrics: EditParserMetrics::default(),
}
}
fn ends_with_tag_prefix(&self) -> bool {
let mut end_prefixes = END_TAGS
fn ends_with_diff_marker_prefix(&self) -> bool {
let diff_markers = [SEPARATOR_MARKER, REPLACE_MARKER];
let mut diff_prefixes = diff_markers
.iter()
.flat_map(|tag| (1..tag.len()).map(move |i| &tag[..i]))
.flat_map(|marker| (1..marker.len()).map(move |i| &marker[..i]))
.chain(["\n"]);
end_prefixes.any(|prefix| self.buffer.ends_with(&prefix))
diff_prefixes.any(|prefix| self.buffer.ends_with(&prefix))
}
fn parse_line_hint(&self, tag: &str) -> Option<u32> {
static LINE_HINT_REGEX: std::sync::LazyLock<Regex> =
std::sync::LazyLock::new(|| Regex::new(r#"line=(?:"?)(\d+)"#).unwrap());
fn parse_line_hint(&self, search_line: &str) -> Option<u32> {
use regex::Regex;
use std::sync::LazyLock;
static LINE_HINT_REGEX: LazyLock<Regex> =
LazyLock::new(|| Regex::new(r#"line=(?:"?)(\d+)"#).unwrap());
LINE_HINT_REGEX
.captures(tag)
.captures(search_line)
.and_then(|caps| caps.get(1))
.and_then(|m| m.as_str().parse::<u32>().ok())
}
}
pub fn finish(self) -> EditParserMetrics {
self.metrics
impl EditFormatParser for DiffFencedEditParser {
fn push(&mut self, chunk: &str) -> SmallVec<[EditParserEvent; 1]> {
self.buffer.push_str(chunk);
let mut edit_events = SmallVec::new();
loop {
match &mut self.state {
DiffParserState::Pending => {
if let Some(diff) = self.buffer.find(SEARCH_MARKER) {
let search_end = diff + SEARCH_MARKER.len();
if let Some(newline_pos) = self.buffer[search_end..].find('\n') {
let search_line = &self.buffer[diff..search_end + newline_pos];
let line_hint = self.parse_line_hint(search_line);
self.buffer.drain(..search_end + newline_pos + 1);
self.state = DiffParserState::WithinSearch {
start: true,
line_hint,
};
} else {
break;
}
} else {
break;
}
}
DiffParserState::WithinSearch { start, line_hint } => {
if !self.buffer.is_empty() {
if *start && self.buffer.starts_with('\n') {
self.buffer.remove(0);
}
*start = false;
}
let line_hint = *line_hint;
if let Some(separator_pos) = self.buffer.find(SEPARATOR_MARKER) {
let mut chunk = self.buffer[..separator_pos].to_string();
if chunk.ends_with('\n') {
chunk.pop();
}
let separator_end = separator_pos + SEPARATOR_MARKER.len();
if let Some(newline_pos) = self.buffer[separator_end..].find('\n') {
self.buffer.drain(..separator_end + newline_pos + 1);
self.state = DiffParserState::WithinReplace { start: true };
edit_events.push(EditParserEvent::OldTextChunk {
chunk,
done: true,
line_hint,
});
} else {
break;
}
} else {
if !self.ends_with_diff_marker_prefix() {
edit_events.push(EditParserEvent::OldTextChunk {
chunk: mem::take(&mut self.buffer),
done: false,
line_hint,
});
}
break;
}
}
DiffParserState::WithinReplace { start } => {
if !self.buffer.is_empty() {
if *start && self.buffer.starts_with('\n') {
self.buffer.remove(0);
}
*start = false;
}
if let Some(replace_pos) = self.buffer.find(REPLACE_MARKER) {
let mut chunk = self.buffer[..replace_pos].to_string();
if chunk.ends_with('\n') {
chunk.pop();
}
self.buffer.drain(..replace_pos + REPLACE_MARKER.len());
if let Some(newline_pos) = self.buffer.find('\n') {
self.buffer.drain(..newline_pos + 1);
} else {
self.buffer.clear();
}
self.state = DiffParserState::Pending;
edit_events.push(EditParserEvent::NewTextChunk { chunk, done: true });
} else {
if !self.ends_with_diff_marker_prefix() {
edit_events.push(EditParserEvent::NewTextChunk {
chunk: mem::take(&mut self.buffer),
done: false,
});
}
break;
}
}
}
}
edit_events
}
fn take_metrics(&mut self) -> EditParserMetrics {
std::mem::take(&mut self.metrics)
}
}
impl EditParser {
pub fn new(format: EditFormat) -> Self {
let parser: Box<dyn EditFormatParser> = match format {
EditFormat::XmlTags => Box::new(XmlEditParser::new()),
EditFormat::DiffFenced => Box::new(DiffFencedEditParser::new()),
};
EditParser { parser }
}
pub fn push(&mut self, chunk: &str) -> SmallVec<[EditParserEvent; 1]> {
self.parser.push(chunk)
}
pub fn finish(mut self) -> EditParserMetrics {
self.parser.take_metrics()
}
}
@ -201,8 +444,8 @@ mod tests {
use std::cmp;
#[gpui::test(iterations = 1000)]
fn test_single_edit(mut rng: StdRng) {
let mut parser = EditParser::new();
fn test_xml_single_edit(mut rng: StdRng) {
let mut parser = EditParser::new(EditFormat::XmlTags);
assert_eq!(
parse_random_chunks(
"<old_text>original</old_text><new_text>updated</new_text>",
@ -225,8 +468,8 @@ mod tests {
}
#[gpui::test(iterations = 1000)]
fn test_multiple_edits(mut rng: StdRng) {
let mut parser = EditParser::new();
fn test_xml_multiple_edits(mut rng: StdRng) {
let mut parser = EditParser::new(EditFormat::XmlTags);
assert_eq!(
parse_random_chunks(
indoc! {"
@ -263,8 +506,8 @@ mod tests {
}
#[gpui::test(iterations = 1000)]
fn test_edits_with_extra_text(mut rng: StdRng) {
let mut parser = EditParser::new();
fn test_xml_edits_with_extra_text(mut rng: StdRng) {
let mut parser = EditParser::new(EditFormat::XmlTags);
assert_eq!(
parse_random_chunks(
indoc! {"
@ -305,8 +548,8 @@ mod tests {
}
#[gpui::test(iterations = 1000)]
fn test_nested_tags(mut rng: StdRng) {
let mut parser = EditParser::new();
fn test_xml_nested_tags(mut rng: StdRng) {
let mut parser = EditParser::new(EditFormat::XmlTags);
assert_eq!(
parse_random_chunks(
"<old_text>code with <tag>nested</tag> elements</old_text><new_text>new <code>content</code></new_text>",
@ -329,8 +572,8 @@ mod tests {
}
#[gpui::test(iterations = 1000)]
fn test_empty_old_and_new_text(mut rng: StdRng) {
let mut parser = EditParser::new();
fn test_xml_empty_old_and_new_text(mut rng: StdRng) {
let mut parser = EditParser::new(EditFormat::XmlTags);
assert_eq!(
parse_random_chunks(
"<old_text></old_text><new_text></new_text>",
@ -353,8 +596,8 @@ mod tests {
}
#[gpui::test(iterations = 100)]
fn test_multiline_content(mut rng: StdRng) {
let mut parser = EditParser::new();
fn test_xml_multiline_content(mut rng: StdRng) {
let mut parser = EditParser::new(EditFormat::XmlTags);
assert_eq!(
parse_random_chunks(
"<old_text>line1\nline2\nline3</old_text><new_text>line1\nmodified line2\nline3</new_text>",
@ -377,8 +620,8 @@ mod tests {
}
#[gpui::test(iterations = 1000)]
fn test_mismatched_tags(mut rng: StdRng) {
let mut parser = EditParser::new();
fn test_xml_mismatched_tags(mut rng: StdRng) {
let mut parser = EditParser::new(EditFormat::XmlTags);
assert_eq!(
parse_random_chunks(
// Reduced from an actual Sonnet 3.7 output
@ -428,7 +671,7 @@ mod tests {
}
);
let mut parser = EditParser::new();
let mut parser = EditParser::new(EditFormat::XmlTags);
assert_eq!(
parse_random_chunks(
// Reduced from an actual Opus 4 output
@ -459,10 +702,230 @@ mod tests {
);
}
#[gpui::test(iterations = 1000)]
fn test_diff_fenced_single_edit(mut rng: StdRng) {
let mut parser = EditParser::new(EditFormat::DiffFenced);
assert_eq!(
parse_random_chunks(
indoc! {"
<<<<<<< SEARCH
original text
=======
updated text
>>>>>>> REPLACE
"},
&mut parser,
&mut rng
),
vec![Edit {
old_text: "original text".to_string(),
new_text: "updated text".to_string(),
line_hint: None,
}]
);
assert_eq!(
parser.finish(),
EditParserMetrics {
tags: 0,
mismatched_tags: 0
}
);
}
#[gpui::test(iterations = 100)]
fn test_line_hints(mut rng: StdRng) {
fn test_diff_fenced_with_markdown_fences(mut rng: StdRng) {
let mut parser = EditParser::new(EditFormat::DiffFenced);
assert_eq!(
parse_random_chunks(
indoc! {"
```diff
<<<<<<< SEARCH
from flask import Flask
=======
import math
from flask import Flask
>>>>>>> REPLACE
```
"},
&mut parser,
&mut rng
),
vec![Edit {
old_text: "from flask import Flask".to_string(),
new_text: "import math\nfrom flask import Flask".to_string(),
line_hint: None,
}]
);
assert_eq!(
parser.finish(),
EditParserMetrics {
tags: 0,
mismatched_tags: 0
}
);
}
#[gpui::test(iterations = 100)]
fn test_diff_fenced_multiple_edits(mut rng: StdRng) {
let mut parser = EditParser::new(EditFormat::DiffFenced);
assert_eq!(
parse_random_chunks(
indoc! {"
<<<<<<< SEARCH
first old
=======
first new
>>>>>>> REPLACE
<<<<<<< SEARCH
second old
=======
second new
>>>>>>> REPLACE
"},
&mut parser,
&mut rng
),
vec![
Edit {
old_text: "first old".to_string(),
new_text: "first new".to_string(),
line_hint: None,
},
Edit {
old_text: "second old".to_string(),
new_text: "second new".to_string(),
line_hint: None,
},
]
);
assert_eq!(
parser.finish(),
EditParserMetrics {
tags: 0,
mismatched_tags: 0
}
);
}
#[gpui::test(iterations = 100)]
fn test_mixed_formats(mut rng: StdRng) {
// Test XML format parser only parses XML tags
let mut xml_parser = EditParser::new(EditFormat::XmlTags);
assert_eq!(
parse_random_chunks(
indoc! {"
<old_text>xml style old</old_text><new_text>xml style new</new_text>
<<<<<<< SEARCH
diff style old
=======
diff style new
>>>>>>> REPLACE
"},
&mut xml_parser,
&mut rng
),
vec![Edit {
old_text: "xml style old".to_string(),
new_text: "xml style new".to_string(),
line_hint: None,
},]
);
assert_eq!(
xml_parser.finish(),
EditParserMetrics {
tags: 2,
mismatched_tags: 0
}
);
// Test diff-fenced format parser only parses diff markers
let mut diff_parser = EditParser::new(EditFormat::DiffFenced);
assert_eq!(
parse_random_chunks(
indoc! {"
<old_text>xml style old</old_text><new_text>xml style new</new_text>
<<<<<<< SEARCH
diff style old
=======
diff style new
>>>>>>> REPLACE
"},
&mut diff_parser,
&mut rng
),
vec![Edit {
old_text: "diff style old".to_string(),
new_text: "diff style new".to_string(),
line_hint: None,
},]
);
assert_eq!(
diff_parser.finish(),
EditParserMetrics {
tags: 0,
mismatched_tags: 0
}
);
}
#[gpui::test(iterations = 100)]
fn test_diff_fenced_empty_sections(mut rng: StdRng) {
let mut parser = EditParser::new(EditFormat::DiffFenced);
assert_eq!(
parse_random_chunks(
indoc! {"
<<<<<<< SEARCH
=======
>>>>>>> REPLACE
"},
&mut parser,
&mut rng
),
vec![Edit {
old_text: "".to_string(),
new_text: "".to_string(),
line_hint: None,
}]
);
assert_eq!(
parser.finish(),
EditParserMetrics {
tags: 0,
mismatched_tags: 0
}
);
}
#[gpui::test(iterations = 100)]
fn test_diff_fenced_with_line_hint(mut rng: StdRng) {
let mut parser = EditParser::new(EditFormat::DiffFenced);
let edits = parse_random_chunks(
indoc! {"
<<<<<<< SEARCH line=42
original text
=======
updated text
>>>>>>> REPLACE
"},
&mut parser,
&mut rng,
);
assert_eq!(
edits,
vec![Edit {
old_text: "original text".to_string(),
line_hint: Some(42),
new_text: "updated text".to_string(),
}]
);
}
#[gpui::test(iterations = 100)]
fn test_xml_line_hints(mut rng: StdRng) {
// Line hint is a single quoted line number
let mut parser = EditParser::new();
let mut parser = EditParser::new(EditFormat::XmlTags);
let edits = parse_random_chunks(
r#"
@ -478,7 +941,7 @@ mod tests {
assert_eq!(edits[0].new_text, "updated code");
// Line hint is a single unquoted line number
let mut parser = EditParser::new();
let mut parser = EditParser::new(EditFormat::XmlTags);
let edits = parse_random_chunks(
r#"
@ -494,7 +957,7 @@ mod tests {
assert_eq!(edits[0].new_text, "updated code");
// Line hint is a range
let mut parser = EditParser::new();
let mut parser = EditParser::new(EditFormat::XmlTags);
let edits = parse_random_chunks(
r#"
@ -510,7 +973,7 @@ mod tests {
assert_eq!(edits[0].new_text, "updated code");
// No line hint
let mut parser = EditParser::new();
let mut parser = EditParser::new(EditFormat::XmlTags);
let edits = parse_random_chunks(
r#"
<old_text>old</old_text>

View file

@ -41,7 +41,7 @@ fn eval_extract_handle_command_output() {
// ----------------------------|----------
// claude-3.7-sonnet | 0.99 (2025-06-14)
// claude-sonnet-4 | 0.97 (2025-06-14)
// gemini-2.5-pro-06-05 | 0.77 (2025-05-22)
// gemini-2.5-pro-06-05 | 0.98 (2025-06-16)
// gemini-2.5-flash | 0.11 (2025-05-22)
// gpt-4.1 | 1.00 (2025-05-22)
@ -59,7 +59,7 @@ fn eval_extract_handle_command_output() {
let edit_description = "Extract `handle_command_output` method from `run_git_blame`.";
eval(
100,
0.7, // Taking the lower bar for Gemini
0.95,
0.05,
EvalInput::from_conversation(
vec![
@ -116,7 +116,7 @@ fn eval_delete_run_git_blame() {
// ----------------------------|----------
// claude-3.7-sonnet | 1.0 (2025-06-14)
// claude-sonnet-4 | 0.96 (2025-06-14)
// gemini-2.5-pro-06-05 |
// gemini-2.5-pro-06-05 | 1.0 (2025-06-16)
// gemini-2.5-flash |
// gpt-4.1 |
let input_file_path = "root/blame.rs";
@ -241,7 +241,7 @@ fn eval_use_wasi_sdk_in_compile_parser_to_wasm() {
//
// claude-3.7-sonnet | 0.96 (2025-06-14)
// claude-sonnet-4 | 0.11 (2025-06-14)
// gemini-2.5-pro-preview-03-25 | 0.99 (2025-05-22)
// gemini-2.5-pro-preview-latest | 0.99 (2025-06-16)
// gemini-2.5-flash-preview-04-17 |
// gpt-4.1 |
let input_file_path = "root/lib.rs";
@ -366,7 +366,7 @@ fn eval_disable_cursor_blinking() {
//
// claude-3.7-sonnet | 0.99 (2025-06-14)
// claude-sonnet-4 | 0.85 (2025-06-14)
// gemini-2.5-pro-preview-03-25 | 1.0 (2025-05-22)
// gemini-2.5-pro-preview-latest | 0.97 (2025-06-16)
// gemini-2.5-flash-preview-04-17 |
// gpt-4.1 |
let input_file_path = "root/editor.rs";
@ -453,12 +453,11 @@ fn eval_from_pixels_constructor() {
// (e.g., at the beginning of the file), yet the evaluation may still
// rate it highly.
//
// Model | Pass rate
// ============================================
//
// claude-4.0-sonnet | 0.99
// claude-3.7-sonnet | 0.88
// gemini-2.5-pro-preview-03-25 | 0.96
// Model | Date | Pass rate
// =========================================================
// claude-4.0-sonnet | 2025-06-14 | 0.99
// claude-3.7-sonnet | 2025-06-14 | 0.88
// gemini-2.5-pro-preview-06-05 | 2025-06-16 | 0.98
// gpt-4.1 |
let input_file_path = "root/canvas.rs";
let input_file_content = include_str!("evals/fixtures/from_pixels_constructor/before.rs");
@ -1498,8 +1497,16 @@ impl EditAgentTest {
.await;
let action_log = cx.new(|_| ActionLog::new(project.clone()));
let edit_format = EditFormat::from_env(agent_model.clone()).unwrap();
Self {
agent: EditAgent::new(agent_model, project.clone(), action_log, Templates::new()),
agent: EditAgent::new(
agent_model,
project.clone(),
action_log,
Templates::new(),
edit_format,
),
project,
judge_model,
}

View file

@ -44,6 +44,10 @@ impl StreamingFuzzyMatcher {
/// Returns `Some(range)` if a match has been found with the accumulated
/// query so far, or `None` if no suitable match exists yet.
pub fn push(&mut self, chunk: &str, line_hint: Option<u32>) -> Option<Range<usize>> {
if line_hint.is_some() {
self.line_hint = line_hint;
}
// Add the chunk to our incomplete line buffer
self.incomplete_line.push_str(chunk);
self.line_hint = line_hint;

View file

@ -1,6 +1,6 @@
use crate::{
Templates,
edit_agent::{EditAgent, EditAgentOutput, EditAgentOutputEvent},
edit_agent::{EditAgent, EditAgentOutput, EditAgentOutputEvent, EditFormat},
schema::json_schema_for,
ui::{COLLAPSED_LINES, ToolOutputPreview},
};
@ -201,8 +201,14 @@ impl Tool for EditFileTool {
let card_clone = card.clone();
let action_log_clone = action_log.clone();
let task = cx.spawn(async move |cx: &mut AsyncApp| {
let edit_agent =
EditAgent::new(model, project.clone(), action_log_clone, Templates::new());
let edit_format = EditFormat::from_model(model.clone())?;
let edit_agent = EditAgent::new(
model,
project.clone(),
action_log_clone,
Templates::new(),
edit_format,
);
let buffer = project
.update(cx, |project, cx| {

View file

@ -0,0 +1,77 @@
You MUST respond with a series of edits to a file, using the following diff format:
```
<<<<<<< SEARCH line=1
from flask import Flask
=======
import math
from flask import Flask
>>>>>>> REPLACE
<<<<<<< SEARCH line=325
return 0
=======
print("Done")
return 0
>>>>>>> REPLACE
```
# File Editing Instructions
- Use the SEARCH/REPLACE diff format shown above
- The SEARCH section must exactly match existing file content, including indentation
- The SEARCH section must come from the actual file, not an outline
- The SEARCH section cannot be empty
- `line` should be a starting line number for the text to be replaced
- Be minimal with replacements:
- For unique lines, include only those lines
- For non-unique lines, include enough context to identify them
- Do not escape quotes, newlines, or other characters
- For multiple occurrences, repeat the same diff block for each instance
- Edits are sequential - each assumes previous edits are already applied
- Only edit the specified file
# Example
```
<<<<<<< SEARCH line=3
struct User {
name: String,
email: String,
}
=======
struct User {
name: String,
email: String,
active: bool,
}
>>>>>>> REPLACE
<<<<<<< SEARCH line=25
let user = User {
name: String::from("John"),
email: String::from("john@example.com"),
};
=======
let user = User {
name: String::from("John"),
email: String::from("john@example.com"),
active: true,
};
>>>>>>> REPLACE
```
# Final instructions
Tool calls have been disabled. You MUST respond using the SEARCH/REPLACE diff format only.
<file_to_edit>
{{path}}
</file_to_edit>
<edit_description>
{{edit_description}}
</edit_description>

View file

@ -65,7 +65,8 @@ const PROMPT_PATHS = [
"crates/agent/src/prompts/summarize_thread_detailed_prompt.txt",
"crates/agent/src/prompts/summarize_thread_prompt.txt",
"crates/assistant_tools/src/templates/create_file_prompt.hbs",
"crates/assistant_tools/src/templates/edit_file_prompt.hbs",
"crates/assistant_tools/src/templates/edit_file_prompt_xml.hbs",
"crates/assistant_tools/src/templates/edit_file_prompt_diff_fenced.hbs",
"crates/git_ui/src/commit_message_prompt.txt",
];