Fuzzy-match lines when applying edits from the assistant (#12056)
This uses Jaro-Winkler similarity for now, which seemed to produce pretty good results in my tests. We can easily swap it with something else if needed. Release Notes: - N/A
This commit is contained in:
Antonio Scandurra
GitHub
parent
0b8c1680fb
commit
3a79aa85f4
4 changed files with 104 additions and 75 deletions
|
|
@ -40,6 +40,7 @@ serde.workspace = true
|
|||
serde_json.workspace = true
|
||||
settings.workspace = true
|
||||
smol.workspace = true
|
||||
strsim = "0.11"
|
||||
telemetry_events.workspace = true
|
||||
theme.workspace = true
|
||||
tiktoken-rs.workspace = true
|
||||
|
|
|
|||
|
|
@ -3058,9 +3058,9 @@ impl ConversationEditor {
|
|||
.entry(buffer)
|
||||
.or_insert(Vec::<(Range<language::Anchor>, _)>::new());
|
||||
for suggestion in suggestions {
|
||||
let ranges =
|
||||
fuzzy_search_lines(snapshot.as_rope(), &suggestion.old_text);
|
||||
if let Some(range) = ranges.first() {
|
||||
if let Some(range) =
|
||||
fuzzy_search_lines(snapshot.as_rope(), &suggestion.old_text)
|
||||
{
|
||||
let edit_start = snapshot.anchor_after(range.start);
|
||||
let edit_end = snapshot.anchor_before(range.end);
|
||||
if let Err(ix) = edits.binary_search_by(|(range, _)| {
|
||||
|
|
|
|||
|
|
@ -6,51 +6,75 @@ use std::ops::Range;
|
|||
///
|
||||
/// Returns a vector of ranges of byte offsets in the buffer corresponding
|
||||
/// to the entire lines of the buffer.
|
||||
pub fn fuzzy_search_lines(haystack: &Rope, needle: &str) -> Vec<Range<usize>> {
|
||||
let mut matches = Vec::new();
|
||||
pub fn fuzzy_search_lines(haystack: &Rope, needle: &str) -> Option<Range<usize>> {
|
||||
const SIMILARITY_THRESHOLD: f64 = 0.8;
|
||||
|
||||
let mut best_match: Option<(Range<usize>, f64)> = None; // (range, score)
|
||||
let mut haystack_lines = haystack.chunks().lines();
|
||||
let mut haystack_line_start = 0;
|
||||
while let Some(haystack_line) = haystack_lines.next() {
|
||||
while let Some(mut haystack_line) = haystack_lines.next() {
|
||||
let next_haystack_line_start = haystack_line_start + haystack_line.len() + 1;
|
||||
let mut trimmed_needle_lines = needle.lines().map(|line| line.trim());
|
||||
if Some(haystack_line.trim()) == trimmed_needle_lines.next() {
|
||||
let match_start = haystack_line_start;
|
||||
let mut match_end = next_haystack_line_start;
|
||||
let matched = loop {
|
||||
match (haystack_lines.next(), trimmed_needle_lines.next()) {
|
||||
(Some(haystack_line), Some(needle_line)) => {
|
||||
// Haystack line differs from needle line: not a match.
|
||||
if haystack_line.trim() == needle_line {
|
||||
match_end = haystack_lines.offset();
|
||||
} else {
|
||||
break false;
|
||||
}
|
||||
let mut advanced_to_next_haystack_line = false;
|
||||
|
||||
let mut matched = true;
|
||||
let match_start = haystack_line_start;
|
||||
let mut match_end = next_haystack_line_start;
|
||||
let mut match_score = 0.0;
|
||||
let mut needle_lines = needle.lines().peekable();
|
||||
while let Some(needle_line) = needle_lines.next() {
|
||||
let similarity = line_similarity(haystack_line, needle_line);
|
||||
if similarity >= SIMILARITY_THRESHOLD {
|
||||
match_end = haystack_lines.offset();
|
||||
match_score += similarity;
|
||||
|
||||
if needle_lines.peek().is_some() {
|
||||
if let Some(next_haystack_line) = haystack_lines.next() {
|
||||
advanced_to_next_haystack_line = true;
|
||||
haystack_line = next_haystack_line;
|
||||
} else {
|
||||
matched = false;
|
||||
break;
|
||||
}
|
||||
// We exhausted the haystack but not the query: not a match.
|
||||
(None, Some(_)) => break false,
|
||||
// We exhausted the query: it's a match.
|
||||
(_, None) => break true,
|
||||
} else {
|
||||
break;
|
||||
}
|
||||
};
|
||||
|
||||
if matched {
|
||||
matches.push(match_start..match_end)
|
||||
} else {
|
||||
matched = false;
|
||||
break;
|
||||
}
|
||||
|
||||
// Advance to the next line.
|
||||
haystack_lines.seek(next_haystack_line_start);
|
||||
}
|
||||
|
||||
if matched
|
||||
&& best_match
|
||||
.as_ref()
|
||||
.map(|(_, best_score)| match_score > *best_score)
|
||||
.unwrap_or(true)
|
||||
{
|
||||
best_match = Some((match_start..match_end, match_score));
|
||||
}
|
||||
|
||||
if advanced_to_next_haystack_line {
|
||||
haystack_lines.seek(next_haystack_line_start);
|
||||
}
|
||||
haystack_line_start = next_haystack_line_start;
|
||||
}
|
||||
matches
|
||||
|
||||
best_match.map(|(range, _)| range)
|
||||
}
|
||||
|
||||
/// Calculates the similarity between two lines, ignoring leading and trailing whitespace,
|
||||
/// using the Jaro-Winkler distance.
|
||||
///
|
||||
/// Returns a value between 0.0 and 1.0, where 1.0 indicates an exact match.
|
||||
fn line_similarity(line1: &str, line2: &str) -> f64 {
|
||||
strsim::jaro_winkler(line1.trim(), line2.trim())
|
||||
}
|
||||
|
||||
#[cfg(test)]
|
||||
mod test {
|
||||
use super::*;
|
||||
use gpui::{AppContext, Context as _};
|
||||
use language::{Buffer, OffsetRangeExt};
|
||||
use language::Buffer;
|
||||
use unindent::Unindent as _;
|
||||
use util::test::marked_text_ranges;
|
||||
|
||||
|
|
@ -79,17 +103,11 @@ mod test {
|
|||
);
|
||||
»
|
||||
|
||||
assert_eq!(
|
||||
« assert_eq!(
|
||||
"something",
|
||||
"else",
|
||||
);
|
||||
|
||||
if b {
|
||||
« assert_eq!(
|
||||
1 + 2,
|
||||
3,
|
||||
);
|
||||
» }
|
||||
»
|
||||
}
|
||||
"#
|
||||
.unindent(),
|
||||
|
|
@ -99,7 +117,7 @@ mod test {
|
|||
let buffer = cx.new_model(|cx| Buffer::local(&text, cx));
|
||||
let snapshot = buffer.read_with(cx, |buffer, _| buffer.snapshot());
|
||||
|
||||
let actual_ranges = fuzzy_search_lines(
|
||||
let actual_range = fuzzy_search_lines(
|
||||
snapshot.as_rope(),
|
||||
&"
|
||||
assert_eq!(
|
||||
|
|
@ -108,43 +126,46 @@ mod test {
|
|||
);
|
||||
"
|
||||
.unindent(),
|
||||
);
|
||||
assert_eq!(
|
||||
actual_ranges,
|
||||
expected_ranges,
|
||||
"actual: {:?}, expected: {:?}",
|
||||
actual_ranges
|
||||
.iter()
|
||||
.map(|range| range.to_point(&snapshot))
|
||||
.collect::<Vec<_>>(),
|
||||
expected_ranges
|
||||
.iter()
|
||||
.map(|range| range.to_point(&snapshot))
|
||||
.collect::<Vec<_>>()
|
||||
);
|
||||
)
|
||||
.unwrap();
|
||||
assert_eq!(actual_range, expected_ranges[0]);
|
||||
|
||||
let actual_ranges = fuzzy_search_lines(
|
||||
let actual_range = fuzzy_search_lines(
|
||||
snapshot.as_rope(),
|
||||
&"
|
||||
assert_eq!(
|
||||
1 + 2,
|
||||
3,
|
||||
);
|
||||
);
|
||||
"
|
||||
.unindent(),
|
||||
)
|
||||
.unwrap();
|
||||
assert_eq!(actual_range, expected_ranges[0]);
|
||||
|
||||
let actual_range = fuzzy_search_lines(
|
||||
snapshot.as_rope(),
|
||||
&"
|
||||
asst_eq!(
|
||||
\"something\",
|
||||
\"els\"
|
||||
)
|
||||
"
|
||||
.unindent(),
|
||||
)
|
||||
.unwrap();
|
||||
assert_eq!(actual_range, expected_ranges[1]);
|
||||
|
||||
let actual_range = fuzzy_search_lines(
|
||||
snapshot.as_rope(),
|
||||
&"
|
||||
assert_eq!(
|
||||
2 + 1,
|
||||
3,
|
||||
);
|
||||
"
|
||||
.unindent(),
|
||||
);
|
||||
assert_eq!(
|
||||
actual_ranges,
|
||||
expected_ranges,
|
||||
"actual: {:?}, expected: {:?}",
|
||||
actual_ranges
|
||||
.iter()
|
||||
.map(|range| range.to_point(&snapshot))
|
||||
.collect::<Vec<_>>(),
|
||||
expected_ranges
|
||||
.iter()
|
||||
.map(|range| range.to_point(&snapshot))
|
||||
.collect::<Vec<_>>()
|
||||
);
|
||||
assert_eq!(actual_range, None);
|
||||
}
|
||||
}
|
||||
|
|
|
|||
Loading…
Add table
Add a link
Reference in a new issue