1use crate::{CharClassifier, CharKind, CharScopeContext, LanguageScope};
2use anyhow::{Context, anyhow};
3use imara_diff::{
4 Algorithm, Sink, diff,
5 intern::{InternedInput, Interner, Token},
6 sources::lines_with_terminator,
7};
8use std::{fmt::Write, iter, ops::Range, sync::Arc};
9
10const MAX_WORD_DIFF_LEN: usize = 512;
11const MAX_WORD_DIFF_LINE_COUNT: usize = 8;
12
13/// Computes a diff between two strings, returning a unified diff string.
14pub fn unified_diff(old_text: &str, new_text: &str) -> String {
15 unified_diff_with_offsets(old_text, new_text, 0, 0)
16}
17
18/// Computes a diff between two strings, returning a unified diff string with
19/// hunk headers adjusted to reflect the given starting line numbers (zero-indexed).
20pub fn unified_diff_with_offsets(
21 old_text: &str,
22 new_text: &str,
23 old_start_line: u32,
24 new_start_line: u32,
25) -> String {
26 unified_diff_with_context(old_text, new_text, old_start_line, new_start_line, 3)
27}
28
29/// Computes a diff between two strings, returning a unified diff string with
30/// hunk headers adjusted to reflect the given starting line numbers (zero-indexed),
31/// and a configurable number of context lines around changes.
32pub fn unified_diff_with_context(
33 old_text: &str,
34 new_text: &str,
35 old_start_line: u32,
36 new_start_line: u32,
37 context_lines: u32,
38) -> String {
39 let input = InternedInput::new(old_text, new_text);
40 diff(
41 Algorithm::Histogram,
42 &input,
43 OffsetUnifiedDiffBuilder::new(&input, old_start_line, new_start_line, context_lines),
44 )
45}
46
47/// A unified diff builder that applies line number offsets to hunk headers.
48struct OffsetUnifiedDiffBuilder<'a> {
49 before: &'a [Token],
50 after: &'a [Token],
51 interner: &'a Interner<&'a str>,
52
53 pos: u32,
54 before_hunk_start: u32,
55 after_hunk_start: u32,
56 before_hunk_len: u32,
57 after_hunk_len: u32,
58
59 old_line_offset: u32,
60 new_line_offset: u32,
61 context_lines: u32,
62
63 buffer: String,
64 dst: String,
65}
66
67impl<'a> OffsetUnifiedDiffBuilder<'a> {
68 fn new(
69 input: &'a InternedInput<&'a str>,
70 old_line_offset: u32,
71 new_line_offset: u32,
72 context_lines: u32,
73 ) -> Self {
74 Self {
75 before_hunk_start: 0,
76 after_hunk_start: 0,
77 before_hunk_len: 0,
78 after_hunk_len: 0,
79 old_line_offset,
80 new_line_offset,
81 context_lines,
82 buffer: String::with_capacity(8),
83 dst: String::new(),
84 interner: &input.interner,
85 before: &input.before,
86 after: &input.after,
87 pos: 0,
88 }
89 }
90
91 fn print_tokens(&mut self, tokens: &[Token], prefix: char) {
92 for &token in tokens {
93 writeln!(&mut self.buffer, "{prefix}{}", self.interner[token]).unwrap();
94 }
95 }
96
97 fn flush(&mut self) {
98 if self.before_hunk_len == 0 && self.after_hunk_len == 0 {
99 return;
100 }
101
102 let end = (self.pos + self.context_lines).min(self.before.len() as u32);
103 self.update_pos(end, end);
104
105 writeln!(
106 &mut self.dst,
107 "@@ -{},{} +{},{} @@",
108 self.before_hunk_start + 1 + self.old_line_offset,
109 self.before_hunk_len,
110 self.after_hunk_start + 1 + self.new_line_offset,
111 self.after_hunk_len,
112 )
113 .unwrap();
114 write!(&mut self.dst, "{}", &self.buffer).unwrap();
115 self.buffer.clear();
116 self.before_hunk_len = 0;
117 self.after_hunk_len = 0;
118 }
119
120 fn update_pos(&mut self, print_to: u32, move_to: u32) {
121 self.print_tokens(&self.before[self.pos as usize..print_to as usize], ' ');
122 let len = print_to - self.pos;
123 self.pos = move_to;
124 self.before_hunk_len += len;
125 self.after_hunk_len += len;
126 }
127}
128
129impl Sink for OffsetUnifiedDiffBuilder<'_> {
130 type Out = String;
131
132 fn process_change(&mut self, before: Range<u32>, after: Range<u32>) {
133 if before.start - self.pos > self.context_lines * 2 {
134 self.flush();
135 }
136 if self.before_hunk_len == 0 && self.after_hunk_len == 0 {
137 self.pos = before.start.saturating_sub(self.context_lines);
138 self.before_hunk_start = self.pos;
139 self.after_hunk_start = after.start.saturating_sub(self.context_lines);
140 }
141 self.update_pos(before.start, before.end);
142 self.before_hunk_len += before.end - before.start;
143 self.after_hunk_len += after.end - after.start;
144 self.print_tokens(
145 &self.before[before.start as usize..before.end as usize],
146 '-',
147 );
148 self.print_tokens(&self.after[after.start as usize..after.end as usize], '+');
149 }
150
151 fn finish(mut self) -> Self::Out {
152 self.flush();
153 self.dst
154 }
155}
156
157/// Computes a diff between two strings, returning a vector of old and new row
158/// ranges.
159pub fn line_diff(old_text: &str, new_text: &str) -> Vec<(Range<u32>, Range<u32>)> {
160 let mut edits = Vec::new();
161 let input = InternedInput::new(
162 lines_with_terminator(old_text),
163 lines_with_terminator(new_text),
164 );
165 diff_internal(&input, &mut |_, _, old_rows, new_rows| {
166 edits.push((old_rows, new_rows));
167 });
168 edits
169}
170
171/// Computes a diff between two strings, returning a vector of edits.
172///
173/// The edits are represented as tuples of byte ranges and replacement strings.
174///
175/// Internally, this function first performs a line-based diff, and then performs a second
176/// word-based diff within hunks that replace small numbers of lines.
177pub fn text_diff(old_text: &str, new_text: &str) -> Vec<(Range<usize>, Arc<str>)> {
178 text_diff_with_options(old_text, new_text, DiffOptions::default())
179}
180
181/// Computes word-level diff ranges between two strings.
182///
183/// Returns a tuple of (old_ranges, new_ranges) where each vector contains
184/// the byte ranges of changed words in the respective text.
185pub fn word_diff_ranges(
186 old_text: &str,
187 new_text: &str,
188 options: DiffOptions,
189) -> (Vec<Range<usize>>, Vec<Range<usize>>) {
190 let mut input: InternedInput<&str> = InternedInput::default();
191 input.update_before(tokenize(old_text, options.language_scope.clone()));
192 input.update_after(tokenize(new_text, options.language_scope));
193
194 let mut old_ranges: Vec<Range<usize>> = Vec::new();
195 let mut new_ranges: Vec<Range<usize>> = Vec::new();
196
197 diff_internal(&input, &mut |old_byte_range, new_byte_range, _, _| {
198 if !old_byte_range.is_empty() {
199 if let Some(last) = old_ranges.last_mut()
200 && last.end >= old_byte_range.start
201 {
202 last.end = old_byte_range.end;
203 } else {
204 old_ranges.push(old_byte_range);
205 }
206 }
207
208 if !new_byte_range.is_empty() {
209 if let Some(last) = new_ranges.last_mut()
210 && last.end >= new_byte_range.start
211 {
212 last.end = new_byte_range.end;
213 } else {
214 new_ranges.push(new_byte_range);
215 }
216 }
217 });
218
219 (old_ranges, new_ranges)
220}
221
222/// Computes character-level diff between two strings.
223///
224/// Usually, you should use `text_diff`, which performs a word-wise diff.
225pub fn char_diff<'a>(old_text: &'a str, new_text: &'a str) -> Vec<(Range<usize>, &'a str)> {
226 let mut input: InternedInput<&str> = InternedInput::default();
227 input.update_before(tokenize_chars(old_text));
228 input.update_after(tokenize_chars(new_text));
229 let mut edits: Vec<(Range<usize>, &str)> = Vec::new();
230 diff_internal(&input, &mut |old_byte_range, new_byte_range, _, _| {
231 let replacement = if new_byte_range.is_empty() {
232 ""
233 } else {
234 &new_text[new_byte_range]
235 };
236 edits.push((old_byte_range, replacement));
237 });
238 edits
239}
240
241pub struct DiffOptions {
242 pub language_scope: Option<LanguageScope>,
243 pub max_word_diff_len: usize,
244 pub max_word_diff_line_count: usize,
245}
246
247impl Default for DiffOptions {
248 fn default() -> Self {
249 Self {
250 language_scope: Default::default(),
251 max_word_diff_len: MAX_WORD_DIFF_LEN,
252 max_word_diff_line_count: MAX_WORD_DIFF_LINE_COUNT,
253 }
254 }
255}
256
257/// Computes a diff between two strings, using a specific language scope's
258/// word characters for word-level diffing.
259pub fn text_diff_with_options(
260 old_text: &str,
261 new_text: &str,
262 options: DiffOptions,
263) -> Vec<(Range<usize>, Arc<str>)> {
264 let empty: Arc<str> = Arc::default();
265 let mut edits = Vec::new();
266 let mut hunk_input = InternedInput::default();
267 let input = InternedInput::new(
268 lines_with_terminator(old_text),
269 lines_with_terminator(new_text),
270 );
271 diff_internal(&input, &mut |old_byte_range,
272 new_byte_range,
273 old_rows,
274 new_rows| {
275 if should_perform_word_diff_within_hunk(
276 &old_rows,
277 &old_byte_range,
278 &new_rows,
279 &new_byte_range,
280 &options,
281 ) {
282 let old_offset = old_byte_range.start;
283 let new_offset = new_byte_range.start;
284 hunk_input.clear();
285 hunk_input.update_before(tokenize(
286 &old_text[old_byte_range],
287 options.language_scope.clone(),
288 ));
289 hunk_input.update_after(tokenize(
290 &new_text[new_byte_range],
291 options.language_scope.clone(),
292 ));
293 diff_internal(&hunk_input, &mut |old_byte_range, new_byte_range, _, _| {
294 let old_byte_range =
295 old_offset + old_byte_range.start..old_offset + old_byte_range.end;
296 let new_byte_range =
297 new_offset + new_byte_range.start..new_offset + new_byte_range.end;
298 let replacement_text = if new_byte_range.is_empty() {
299 empty.clone()
300 } else {
301 new_text[new_byte_range].into()
302 };
303 edits.push((old_byte_range, replacement_text));
304 });
305 } else {
306 let replacement_text = if new_byte_range.is_empty() {
307 empty.clone()
308 } else {
309 new_text[new_byte_range].into()
310 };
311 edits.push((old_byte_range, replacement_text));
312 }
313 });
314 edits
315}
316
317pub fn apply_diff_patch(base_text: &str, patch: &str) -> Result<String, anyhow::Error> {
318 let patch = diffy::Patch::from_str(patch).context("Failed to parse patch")?;
319 let result = diffy::apply(base_text, &patch);
320 result.map_err(|err| anyhow!(err))
321}
322
323pub fn apply_reversed_diff_patch(base_text: &str, patch: &str) -> Result<String, anyhow::Error> {
324 let patch = diffy::Patch::from_str(patch).context("Failed to parse patch")?;
325 let reversed = patch.reverse();
326 diffy::apply(base_text, &reversed).map_err(|err| anyhow!(err))
327}
328
329fn should_perform_word_diff_within_hunk(
330 old_row_range: &Range<u32>,
331 old_byte_range: &Range<usize>,
332 new_row_range: &Range<u32>,
333 new_byte_range: &Range<usize>,
334 options: &DiffOptions,
335) -> bool {
336 !old_byte_range.is_empty()
337 && !new_byte_range.is_empty()
338 && old_byte_range.len() <= options.max_word_diff_len
339 && new_byte_range.len() <= options.max_word_diff_len
340 && old_row_range.len() <= options.max_word_diff_line_count
341 && new_row_range.len() <= options.max_word_diff_line_count
342}
343
344fn diff_internal(
345 input: &InternedInput<&str>,
346 on_change: &mut dyn FnMut(Range<usize>, Range<usize>, Range<u32>, Range<u32>),
347) {
348 let mut old_offset = 0;
349 let mut new_offset = 0;
350 let mut old_token_ix = 0;
351 let mut new_token_ix = 0;
352 diff(
353 Algorithm::Histogram,
354 input,
355 |old_tokens: Range<u32>, new_tokens: Range<u32>| {
356 old_offset += token_len(
357 input,
358 &input.before[old_token_ix as usize..old_tokens.start as usize],
359 );
360 new_offset += token_len(
361 input,
362 &input.after[new_token_ix as usize..new_tokens.start as usize],
363 );
364 let old_len = token_len(
365 input,
366 &input.before[old_tokens.start as usize..old_tokens.end as usize],
367 );
368 let new_len = token_len(
369 input,
370 &input.after[new_tokens.start as usize..new_tokens.end as usize],
371 );
372 let old_byte_range = old_offset..old_offset + old_len;
373 let new_byte_range = new_offset..new_offset + new_len;
374 old_token_ix = old_tokens.end;
375 new_token_ix = new_tokens.end;
376 old_offset = old_byte_range.end;
377 new_offset = new_byte_range.end;
378 on_change(old_byte_range, new_byte_range, old_tokens, new_tokens);
379 },
380 );
381}
382
383fn tokenize_chars(text: &str) -> impl Iterator<Item = &str> {
384 let mut chars = text.char_indices().peekable();
385 iter::from_fn(move || {
386 let (start, c) = chars.next()?;
387 Some(&text[start..start + c.len_utf8()])
388 })
389}
390
391fn tokenize(text: &str, language_scope: Option<LanguageScope>) -> impl Iterator<Item = &str> {
392 let classifier =
393 CharClassifier::new(language_scope).scope_context(Some(CharScopeContext::Completion));
394 let mut chars = text.char_indices();
395 let mut prev = None;
396 let mut start_ix = 0;
397 iter::from_fn(move || {
398 for (ix, c) in chars.by_ref() {
399 let mut token = None;
400 let kind = classifier.kind(c);
401 if let Some((prev_char, prev_kind)) = prev
402 && (kind != prev_kind || (kind == CharKind::Punctuation && c != prev_char))
403 {
404 token = Some(&text[start_ix..ix]);
405 start_ix = ix;
406 }
407 prev = Some((c, kind));
408 if token.is_some() {
409 return token;
410 }
411 }
412 if start_ix < text.len() {
413 let token = &text[start_ix..];
414 start_ix = text.len();
415 return Some(token);
416 }
417 None
418 })
419}
420
421fn token_len(input: &InternedInput<&str>, tokens: &[Token]) -> usize {
422 tokens
423 .iter()
424 .map(|token| input.interner[*token].len())
425 .sum()
426}
427
428#[cfg(test)]
429mod tests {
430 use super::*;
431
432 #[test]
433 fn test_tokenize() {
434 let text = "";
435 assert_eq!(tokenize(text, None).collect::<Vec<_>>(), Vec::<&str>::new());
436
437 let text = " ";
438 assert_eq!(tokenize(text, None).collect::<Vec<_>>(), vec![" "]);
439
440 let text = "one";
441 assert_eq!(tokenize(text, None).collect::<Vec<_>>(), vec!["one"]);
442
443 let text = "one\n";
444 assert_eq!(tokenize(text, None).collect::<Vec<_>>(), vec!["one", "\n"]);
445
446 let text = "one.two(three)";
447 assert_eq!(
448 tokenize(text, None).collect::<Vec<_>>(),
449 vec!["one", ".", "two", "(", "three", ")"]
450 );
451
452 let text = "one two three()";
453 assert_eq!(
454 tokenize(text, None).collect::<Vec<_>>(),
455 vec!["one", " ", "two", " ", "three", "(", ")"]
456 );
457
458 let text = " one\n two three";
459 assert_eq!(
460 tokenize(text, None).collect::<Vec<_>>(),
461 vec![" ", "one", "\n ", "two", " ", "three"]
462 );
463 }
464
465 #[test]
466 fn test_text_diff() {
467 let old_text = "one two three";
468 let new_text = "one TWO three";
469 assert_eq!(text_diff(old_text, new_text), [(4..7, "TWO".into()),]);
470
471 let old_text = "one\ntwo\nthree\n";
472 let new_text = "one\ntwo\nAND\nTHEN\nthree\n";
473 assert_eq!(
474 text_diff(old_text, new_text),
475 [(8..8, "AND\nTHEN\n".into()),]
476 );
477
478 let old_text = "one two\nthree four five\nsix seven eight nine\nten\n";
479 let new_text = "one two\nthree FOUR five\nsix SEVEN eight nine\nten\nELEVEN\n";
480 assert_eq!(
481 text_diff(old_text, new_text),
482 [
483 (14..18, "FOUR".into()),
484 (28..33, "SEVEN".into()),
485 (49..49, "ELEVEN\n".into())
486 ]
487 );
488 }
489
490 #[test]
491 fn test_apply_diff_patch() {
492 let old_text = "one two\nthree four five\nsix seven eight nine\nten\n";
493 let new_text = "one two\nthree FOUR five\nsix SEVEN eight nine\nten\nELEVEN\n";
494 let patch = unified_diff(old_text, new_text);
495 assert_eq!(apply_diff_patch(old_text, &patch).unwrap(), new_text);
496 }
497
498 #[test]
499 fn test_apply_reversed_diff_patch() {
500 let old_text = "one two\nthree four five\nsix seven eight nine\nten\n";
501 let new_text = "one two\nthree FOUR five\nsix SEVEN eight nine\nten\nELEVEN\n";
502 let patch = unified_diff(old_text, new_text);
503 assert_eq!(
504 apply_reversed_diff_patch(new_text, &patch).unwrap(),
505 old_text
506 );
507 }
508
509 #[test]
510 fn test_char_diff() {
511 assert_eq!(char_diff("", ""), vec![]);
512 assert_eq!(char_diff("", "abc"), vec![(0..0, "abc")]);
513 assert_eq!(char_diff("abc", ""), vec![(0..3, "")]);
514 assert_eq!(char_diff("ac", "abc"), vec![(1..1, "b")]); // "b" inserted
515 assert_eq!(char_diff("abc", "ac"), vec![(1..2, "")]); // "b" deleted
516 assert_eq!(char_diff("abc", "adc"), vec![(1..2, "d")]); // "b" replaced with "d"
517 assert_eq!(char_diff("ζ₯", "ζ₯ζ¬θͺ"), vec![(3..3, "ζ¬θͺ")]); // "ζ¬θͺ" inserted
518 assert_eq!(char_diff("ζ₯ζ¬θͺ", "ζ₯"), vec![(3..9, "")]); // "ζ¬θͺ" deleted
519 assert_eq!(char_diff("π", "πππ"), vec![(4..4, "ππ")]); // "ππ" inserted
520 assert_eq!(
521 char_diff("testζ₯ζ¬", "testζ₯ζ¬θͺγ§γ"),
522 vec![(10..10, "θͺγ§γ")]
523 );
524 }
525
526 #[test]
527 fn test_unified_diff_with_offsets() {
528 let old_text = "foo\nbar\nbaz\n";
529 let new_text = "foo\nBAR\nbaz\n";
530
531 let expected_diff_body = " foo\n-bar\n+BAR\n baz\n";
532
533 let diff_no_offset = unified_diff(old_text, new_text);
534 assert_eq!(
535 diff_no_offset,
536 format!("@@ -1,3 +1,3 @@\n{}", expected_diff_body)
537 );
538
539 let diff_with_offset = unified_diff_with_offsets(old_text, new_text, 9, 11);
540 assert_eq!(
541 diff_with_offset,
542 format!("@@ -10,3 +12,3 @@\n{}", expected_diff_body)
543 );
544
545 let diff_with_offset = unified_diff_with_offsets(old_text, new_text, 99, 104);
546 assert_eq!(
547 diff_with_offset,
548 format!("@@ -100,3 +105,3 @@\n{}", expected_diff_body)
549 );
550 }
551
552 #[test]
553 fn test_unified_diff_with_context() {
554 // Test that full context includes all lines from the start
555 let old_text = "line1\nline2\nline3\nline4\nline5\nCHANGE_ME\nline7\nline8\n";
556 let new_text = "line1\nline2\nline3\nline4\nline5\nCHANGED\nline7\nline8\n";
557
558 // With default 3 lines of context, the diff starts at line 3
559 let diff_default = unified_diff_with_offsets(old_text, new_text, 0, 0);
560 assert_eq!(
561 diff_default,
562 "@@ -3,6 +3,6 @@\n line3\n line4\n line5\n-CHANGE_ME\n+CHANGED\n line7\n line8\n"
563 );
564
565 // With full context (8 lines), the diff starts at line 1
566 let diff_full_context = unified_diff_with_context(old_text, new_text, 0, 0, 8);
567 assert_eq!(
568 diff_full_context,
569 "@@ -1,8 +1,8 @@\n line1\n line2\n line3\n line4\n line5\n-CHANGE_ME\n+CHANGED\n line7\n line8\n"
570 );
571
572 // With 0 context, only the changed line is shown
573 let diff_no_context = unified_diff_with_context(old_text, new_text, 0, 0, 0);
574 assert_eq!(diff_no_context, "@@ -6,1 +6,1 @@\n-CHANGE_ME\n+CHANGED\n");
575 }
576}