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