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, |_, _, 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, |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
222pub struct DiffOptions {
223 pub language_scope: Option<LanguageScope>,
224 pub max_word_diff_len: usize,
225 pub max_word_diff_line_count: usize,
226}
227
228impl Default for DiffOptions {
229 fn default() -> Self {
230 Self {
231 language_scope: Default::default(),
232 max_word_diff_len: MAX_WORD_DIFF_LEN,
233 max_word_diff_line_count: MAX_WORD_DIFF_LINE_COUNT,
234 }
235 }
236}
237
238/// Computes a diff between two strings, using a specific language scope's
239/// word characters for word-level diffing.
240pub fn text_diff_with_options_internal(
241 old_text: &str,
242 new_text: &str,
243 options: DiffOptions,
244) -> Vec<text::Edit<usize>> {
245 let mut edits = Vec::new();
246 let mut hunk_input = InternedInput::default();
247 let input = InternedInput::new(
248 lines_with_terminator(old_text),
249 lines_with_terminator(new_text),
250 );
251 diff_internal(
252 &input,
253 |old_byte_range, new_byte_range, old_rows, new_rows| {
254 if should_perform_word_diff_within_hunk(
255 &old_rows,
256 &old_byte_range,
257 &new_rows,
258 &new_byte_range,
259 &options,
260 ) {
261 let old_offset = old_byte_range.start;
262 let new_offset = new_byte_range.start;
263 hunk_input.clear();
264 hunk_input.update_before(tokenize(
265 &old_text[old_byte_range],
266 options.language_scope.clone(),
267 ));
268 hunk_input.update_after(tokenize(
269 &new_text[new_byte_range],
270 options.language_scope.clone(),
271 ));
272 diff_internal(&hunk_input, |old_byte_range, new_byte_range, _, _| {
273 let old_byte_range =
274 old_offset + old_byte_range.start..old_offset + old_byte_range.end;
275 let new_byte_range =
276 new_offset + new_byte_range.start..new_offset + new_byte_range.end;
277 edits.push(text::Edit {
278 old: old_byte_range,
279 new: new_byte_range,
280 });
281 });
282 } else {
283 edits.push(text::Edit {
284 old: old_byte_range,
285 new: new_byte_range,
286 });
287 }
288 },
289 );
290 edits
291}
292
293pub fn text_diff_with_options(
294 old_text: &str,
295 new_text: &str,
296 options: DiffOptions,
297) -> Vec<(Range<usize>, Arc<str>)> {
298 text_diff_with_options_internal(old_text, new_text, options)
299 .into_iter()
300 .map(|edit| (edit.old, new_text[edit.new].into()))
301 .collect()
302}
303
304pub fn apply_diff_patch(base_text: &str, patch: &str) -> Result<String, anyhow::Error> {
305 let patch = diffy::Patch::from_str(patch).context("Failed to parse patch")?;
306 let result = diffy::apply(base_text, &patch);
307 result.map_err(|err| anyhow!(err))
308}
309
310pub fn apply_reversed_diff_patch(base_text: &str, patch: &str) -> Result<String, anyhow::Error> {
311 let patch = diffy::Patch::from_str(patch).context("Failed to parse patch")?;
312 let reversed = patch.reverse();
313 diffy::apply(base_text, &reversed).map_err(|err| anyhow!(err))
314}
315
316fn should_perform_word_diff_within_hunk(
317 old_row_range: &Range<u32>,
318 old_byte_range: &Range<usize>,
319 new_row_range: &Range<u32>,
320 new_byte_range: &Range<usize>,
321 options: &DiffOptions,
322) -> bool {
323 !old_byte_range.is_empty()
324 && !new_byte_range.is_empty()
325 && old_byte_range.len() <= options.max_word_diff_len
326 && new_byte_range.len() <= options.max_word_diff_len
327 && old_row_range.len() <= options.max_word_diff_line_count
328 && new_row_range.len() <= options.max_word_diff_line_count
329}
330
331fn diff_internal(
332 input: &InternedInput<&str>,
333 mut on_change: impl FnMut(Range<usize>, Range<usize>, Range<u32>, Range<u32>),
334) {
335 let mut old_offset = 0;
336 let mut new_offset = 0;
337 let mut old_token_ix = 0;
338 let mut new_token_ix = 0;
339 diff(
340 Algorithm::Histogram,
341 input,
342 |old_tokens: Range<u32>, new_tokens: Range<u32>| {
343 old_offset += token_len(
344 input,
345 &input.before[old_token_ix as usize..old_tokens.start as usize],
346 );
347 new_offset += token_len(
348 input,
349 &input.after[new_token_ix as usize..new_tokens.start as usize],
350 );
351 let old_len = token_len(
352 input,
353 &input.before[old_tokens.start as usize..old_tokens.end as usize],
354 );
355 let new_len = token_len(
356 input,
357 &input.after[new_tokens.start as usize..new_tokens.end as usize],
358 );
359 let old_byte_range = old_offset..old_offset + old_len;
360 let new_byte_range = new_offset..new_offset + new_len;
361 old_token_ix = old_tokens.end;
362 new_token_ix = new_tokens.end;
363 old_offset = old_byte_range.end;
364 new_offset = new_byte_range.end;
365 on_change(old_byte_range, new_byte_range, old_tokens, new_tokens);
366 },
367 );
368}
369
370fn tokenize(text: &str, language_scope: Option<LanguageScope>) -> impl Iterator<Item = &str> {
371 let classifier =
372 CharClassifier::new(language_scope).scope_context(Some(CharScopeContext::Completion));
373 let mut chars = text.char_indices();
374 let mut prev = None;
375 let mut start_ix = 0;
376 iter::from_fn(move || {
377 for (ix, c) in chars.by_ref() {
378 let mut token = None;
379 let kind = classifier.kind(c);
380 if let Some((prev_char, prev_kind)) = prev
381 && (kind != prev_kind || (kind == CharKind::Punctuation && c != prev_char))
382 {
383 token = Some(&text[start_ix..ix]);
384 start_ix = ix;
385 }
386 prev = Some((c, kind));
387 if token.is_some() {
388 return token;
389 }
390 }
391 if start_ix < text.len() {
392 let token = &text[start_ix..];
393 start_ix = text.len();
394 return Some(token);
395 }
396 None
397 })
398}
399
400fn token_len(input: &InternedInput<&str>, tokens: &[Token]) -> usize {
401 tokens
402 .iter()
403 .map(|token| input.interner[*token].len())
404 .sum()
405}
406
407#[cfg(test)]
408mod tests {
409 use super::*;
410
411 #[test]
412 fn test_tokenize() {
413 let text = "";
414 assert_eq!(tokenize(text, None).collect::<Vec<_>>(), Vec::<&str>::new());
415
416 let text = " ";
417 assert_eq!(tokenize(text, None).collect::<Vec<_>>(), vec![" "]);
418
419 let text = "one";
420 assert_eq!(tokenize(text, None).collect::<Vec<_>>(), vec!["one"]);
421
422 let text = "one\n";
423 assert_eq!(tokenize(text, None).collect::<Vec<_>>(), vec!["one", "\n"]);
424
425 let text = "one.two(three)";
426 assert_eq!(
427 tokenize(text, None).collect::<Vec<_>>(),
428 vec!["one", ".", "two", "(", "three", ")"]
429 );
430
431 let text = "one two three()";
432 assert_eq!(
433 tokenize(text, None).collect::<Vec<_>>(),
434 vec!["one", " ", "two", " ", "three", "(", ")"]
435 );
436
437 let text = " one\n two three";
438 assert_eq!(
439 tokenize(text, None).collect::<Vec<_>>(),
440 vec![" ", "one", "\n ", "two", " ", "three"]
441 );
442 }
443
444 #[test]
445 fn test_text_diff() {
446 let old_text = "one two three";
447 let new_text = "one TWO three";
448 assert_eq!(text_diff(old_text, new_text), [(4..7, "TWO".into()),]);
449
450 let old_text = "one\ntwo\nthree\n";
451 let new_text = "one\ntwo\nAND\nTHEN\nthree\n";
452 assert_eq!(
453 text_diff(old_text, new_text),
454 [(8..8, "AND\nTHEN\n".into()),]
455 );
456
457 let old_text = "one two\nthree four five\nsix seven eight nine\nten\n";
458 let new_text = "one two\nthree FOUR five\nsix SEVEN eight nine\nten\nELEVEN\n";
459 assert_eq!(
460 text_diff(old_text, new_text),
461 [
462 (14..18, "FOUR".into()),
463 (28..33, "SEVEN".into()),
464 (49..49, "ELEVEN\n".into())
465 ]
466 );
467 }
468
469 #[test]
470 fn test_apply_diff_patch() {
471 let old_text = "one two\nthree four five\nsix seven eight nine\nten\n";
472 let new_text = "one two\nthree FOUR five\nsix SEVEN eight nine\nten\nELEVEN\n";
473 let patch = unified_diff(old_text, new_text);
474 assert_eq!(apply_diff_patch(old_text, &patch).unwrap(), new_text);
475 }
476
477 #[test]
478 fn test_apply_reversed_diff_patch() {
479 let old_text = "one two\nthree four five\nsix seven eight nine\nten\n";
480 let new_text = "one two\nthree FOUR five\nsix SEVEN eight nine\nten\nELEVEN\n";
481 let patch = unified_diff(old_text, new_text);
482 assert_eq!(
483 apply_reversed_diff_patch(new_text, &patch).unwrap(),
484 old_text
485 );
486 }
487
488 #[test]
489 fn test_unified_diff_with_offsets() {
490 let old_text = "foo\nbar\nbaz\n";
491 let new_text = "foo\nBAR\nbaz\n";
492
493 let expected_diff_body = " foo\n-bar\n+BAR\n baz\n";
494
495 let diff_no_offset = unified_diff(old_text, new_text);
496 assert_eq!(
497 diff_no_offset,
498 format!("@@ -1,3 +1,3 @@\n{}", expected_diff_body)
499 );
500
501 let diff_with_offset = unified_diff_with_offsets(old_text, new_text, 9, 11);
502 assert_eq!(
503 diff_with_offset,
504 format!("@@ -10,3 +12,3 @@\n{}", expected_diff_body)
505 );
506
507 let diff_with_offset = unified_diff_with_offsets(old_text, new_text, 99, 104);
508 assert_eq!(
509 diff_with_offset,
510 format!("@@ -100,3 +105,3 @@\n{}", expected_diff_body)
511 );
512 }
513
514 #[test]
515 fn test_unified_diff_with_context() {
516 // Test that full context includes all lines from the start
517 let old_text = "line1\nline2\nline3\nline4\nline5\nCHANGE_ME\nline7\nline8\n";
518 let new_text = "line1\nline2\nline3\nline4\nline5\nCHANGED\nline7\nline8\n";
519
520 // With default 3 lines of context, the diff starts at line 3
521 let diff_default = unified_diff_with_offsets(old_text, new_text, 0, 0);
522 assert_eq!(
523 diff_default,
524 "@@ -3,6 +3,6 @@\n line3\n line4\n line5\n-CHANGE_ME\n+CHANGED\n line7\n line8\n"
525 );
526
527 // With full context (8 lines), the diff starts at line 1
528 let diff_full_context = unified_diff_with_context(old_text, new_text, 0, 0, 8);
529 assert_eq!(
530 diff_full_context,
531 "@@ -1,8 +1,8 @@\n line1\n line2\n line3\n line4\n line5\n-CHANGE_ME\n+CHANGED\n line7\n line8\n"
532 );
533
534 // With 0 context, only the changed line is shown
535 let diff_no_context = unified_diff_with_context(old_text, new_text, 0, 0, 0);
536 assert_eq!(diff_no_context, "@@ -6,1 +6,1 @@\n-CHANGE_ME\n+CHANGED\n");
537 }
538}