1use std::ops::Range;
2use std::path::Path;
3use std::sync::Arc;
4
5use edit_prediction::udiff::apply_diff_to_string;
6use language::{char_diff, text_diff};
7
8use zeta_prompt::ZetaPromptInput;
9
10fn apply_diff_to_string_lenient(diff_str: &str, text: &str) -> String {
11 let hunks = parse_diff_hunks(diff_str);
12 let mut result = text.to_string();
13
14 for hunk in hunks {
15 let hunk_diff = format!("--- a/file\n+++ b/file\n{}", format_hunk(&hunk));
16 if let Ok(updated) = apply_diff_to_string(&hunk_diff, &result) {
17 result = updated;
18 }
19 }
20
21 result
22}
23
24#[derive(Debug, Clone, PartialEq, Eq)]
25struct ParsedHunk {
26 old_start: u32,
27 old_count: u32,
28 new_start: u32,
29 new_count: u32,
30 lines: Vec<HunkLine>,
31}
32
33#[derive(Debug, Clone, PartialEq, Eq)]
34enum HunkLine {
35 Context(String),
36 Addition(String),
37 Deletion(String),
38}
39
40fn parse_hunk_header(line: &str) -> Option<(u32, u32, u32, u32)> {
41 let line = line.strip_prefix("@@ -")?;
42 let (old_part, rest) = line.split_once(' ')?;
43 let rest = rest.strip_prefix('+')?;
44 let (new_part, _) = rest.split_once(" @@")?;
45
46 let (old_start, old_count) = if let Some((start, count)) = old_part.split_once(',') {
47 (start.parse().ok()?, count.parse().ok()?)
48 } else {
49 (old_part.parse().ok()?, 1)
50 };
51
52 let (new_start, new_count) = if let Some((start, count)) = new_part.split_once(',') {
53 (start.parse().ok()?, count.parse().ok()?)
54 } else {
55 (new_part.parse().ok()?, 1)
56 };
57
58 Some((old_start, old_count, new_start, new_count))
59}
60
61fn parse_diff_hunks(diff: &str) -> Vec<ParsedHunk> {
62 let mut hunks = Vec::new();
63 let mut current_hunk: Option<ParsedHunk> = None;
64
65 for line in diff.lines() {
66 if let Some((old_start, old_count, new_start, new_count)) = parse_hunk_header(line) {
67 if let Some(hunk) = current_hunk.take() {
68 hunks.push(hunk);
69 }
70 current_hunk = Some(ParsedHunk {
71 old_start,
72 old_count,
73 new_start,
74 new_count,
75 lines: Vec::new(),
76 });
77 } else if let Some(ref mut hunk) = current_hunk {
78 if let Some(stripped) = line.strip_prefix('+') {
79 hunk.lines.push(HunkLine::Addition(stripped.to_string()));
80 } else if let Some(stripped) = line.strip_prefix('-') {
81 hunk.lines.push(HunkLine::Deletion(stripped.to_string()));
82 } else if let Some(stripped) = line.strip_prefix(' ') {
83 hunk.lines.push(HunkLine::Context(stripped.to_string()));
84 } else if line.is_empty() {
85 hunk.lines.push(HunkLine::Context(String::new()));
86 }
87 }
88 }
89
90 if let Some(hunk) = current_hunk {
91 hunks.push(hunk);
92 }
93
94 hunks
95}
96
97fn format_hunk(hunk: &ParsedHunk) -> String {
98 let mut result = format!(
99 "@@ -{},{} +{},{} @@\n",
100 hunk.old_start, hunk.old_count, hunk.new_start, hunk.new_count
101 );
102 for line in &hunk.lines {
103 match line {
104 HunkLine::Context(text) => {
105 result.push(' ');
106 result.push_str(text);
107 result.push('\n');
108 }
109 HunkLine::Addition(text) => {
110 result.push('+');
111 result.push_str(text);
112 result.push('\n');
113 }
114 HunkLine::Deletion(text) => {
115 result.push('-');
116 result.push_str(text);
117 result.push('\n');
118 }
119 }
120 }
121 result
122}
123
124fn filter_diff_hunks_by_excerpt(
125 diff: &str,
126 excerpt_start_row: u32,
127 excerpt_row_count: u32,
128) -> (String, i32) {
129 let hunks = parse_diff_hunks(diff);
130 let excerpt_start_0based = excerpt_start_row;
131 let excerpt_end_0based = excerpt_start_row + excerpt_row_count;
132
133 let mut filtered_hunks = Vec::new();
134 let mut cumulative_line_offset: i32 = 0;
135
136 for hunk in hunks {
137 let hunk_start_0based = hunk.new_start.saturating_sub(1);
138 let hunk_end_0based = hunk_start_0based + hunk.new_count;
139
140 let additions: i32 = hunk
141 .lines
142 .iter()
143 .filter(|l| matches!(l, HunkLine::Addition(_)))
144 .count() as i32;
145 let deletions: i32 = hunk
146 .lines
147 .iter()
148 .filter(|l| matches!(l, HunkLine::Deletion(_)))
149 .count() as i32;
150 let hunk_line_delta = additions - deletions;
151
152 if hunk_end_0based <= excerpt_start_0based {
153 cumulative_line_offset += hunk_line_delta;
154 continue;
155 }
156
157 if hunk_start_0based >= excerpt_end_0based {
158 continue;
159 }
160
161 let mut filtered_lines = Vec::new();
162 let mut current_row_0based = hunk_start_0based;
163 let mut filtered_old_count = 0u32;
164 let mut filtered_new_count = 0u32;
165 let mut first_included_row: Option<u32> = None;
166
167 for line in &hunk.lines {
168 match line {
169 HunkLine::Context(text) => {
170 if current_row_0based >= excerpt_start_0based
171 && current_row_0based < excerpt_end_0based
172 {
173 if first_included_row.is_none() {
174 first_included_row = Some(current_row_0based);
175 }
176 filtered_lines.push(HunkLine::Context(text.clone()));
177 filtered_old_count += 1;
178 filtered_new_count += 1;
179 }
180 current_row_0based += 1;
181 }
182 HunkLine::Addition(text) => {
183 if current_row_0based >= excerpt_start_0based
184 && current_row_0based < excerpt_end_0based
185 {
186 if first_included_row.is_none() {
187 first_included_row = Some(current_row_0based);
188 }
189 filtered_lines.push(HunkLine::Addition(text.clone()));
190 filtered_new_count += 1;
191 }
192 current_row_0based += 1;
193 }
194 HunkLine::Deletion(text) => {
195 if current_row_0based >= excerpt_start_0based
196 && current_row_0based < excerpt_end_0based
197 {
198 if first_included_row.is_none() {
199 first_included_row = Some(current_row_0based);
200 }
201 filtered_lines.push(HunkLine::Deletion(text.clone()));
202 filtered_old_count += 1;
203 }
204 }
205 }
206 }
207
208 if !filtered_lines.is_empty() {
209 let first_row = first_included_row.unwrap_or(excerpt_start_0based);
210 let new_start_1based = (first_row - excerpt_start_0based) + 1;
211
212 filtered_hunks.push(ParsedHunk {
213 old_start: new_start_1based,
214 old_count: filtered_old_count,
215 new_start: new_start_1based,
216 new_count: filtered_new_count,
217 lines: filtered_lines,
218 });
219 }
220
221 cumulative_line_offset += hunk_line_delta;
222 }
223
224 let mut result = String::new();
225 for hunk in &filtered_hunks {
226 result.push_str(&format_hunk(hunk));
227 }
228
229 (result, cumulative_line_offset)
230}
231
232fn compute_excerpt_aware_reversal_overlap(
233 edit_history_diffs: &[&str],
234 excerpt_content: &str,
235 excerpt_start_row: u32,
236 predicted_content: &str,
237) -> ReversalOverlap {
238 let mut current_content = excerpt_content.to_string();
239 let mut current_excerpt_start_row = excerpt_start_row;
240
241 for diff in edit_history_diffs.iter().rev() {
242 if diff.is_empty() {
243 continue;
244 }
245
246 let current_row_count = current_content.lines().count() as u32;
247 let (filtered_diff, _line_offset) =
248 filter_diff_hunks_by_excerpt(diff, current_excerpt_start_row, current_row_count.max(1));
249
250 if filtered_diff.is_empty() {
251 let hunks = parse_diff_hunks(diff);
252 for hunk in hunks {
253 let hunk_end = hunk.new_start.saturating_sub(1) + hunk.new_count;
254 if hunk_end <= current_excerpt_start_row {
255 let additions: u32 = hunk
256 .lines
257 .iter()
258 .filter(|l| matches!(l, HunkLine::Addition(_)))
259 .count() as u32;
260 let deletions: u32 = hunk
261 .lines
262 .iter()
263 .filter(|l| matches!(l, HunkLine::Deletion(_)))
264 .count() as u32;
265 if additions >= deletions {
266 current_excerpt_start_row =
267 current_excerpt_start_row.saturating_sub(additions - deletions);
268 } else {
269 current_excerpt_start_row += deletions - additions;
270 }
271 }
272 }
273 continue;
274 }
275
276 let reversed = reverse_diff(&format!("--- a/file\n+++ b/file\n{}", filtered_diff));
277 match apply_diff_to_string(&reversed, ¤t_content) {
278 Ok(updated) => {
279 current_content = updated;
280 }
281 Err(_) => {
282 continue;
283 }
284 }
285
286 let hunks = parse_diff_hunks(diff);
287 for hunk in hunks {
288 let hunk_end = hunk.new_start.saturating_sub(1) + hunk.new_count;
289 if hunk_end <= current_excerpt_start_row {
290 let additions: u32 = hunk
291 .lines
292 .iter()
293 .filter(|l| matches!(l, HunkLine::Addition(_)))
294 .count() as u32;
295 let deletions: u32 = hunk
296 .lines
297 .iter()
298 .filter(|l| matches!(l, HunkLine::Deletion(_)))
299 .count() as u32;
300 if additions >= deletions {
301 current_excerpt_start_row =
302 current_excerpt_start_row.saturating_sub(additions - deletions);
303 } else {
304 current_excerpt_start_row += deletions - additions;
305 }
306 }
307 }
308 }
309
310 compute_reversal_overlap(¤t_content, excerpt_content, predicted_content)
311}
312
313fn reverse_diff(diff: &str) -> String {
314 let mut result: String = diff
315 .lines()
316 .map(|line| {
317 if line.starts_with("--- ") {
318 line.replacen("--- ", "+++ ", 1)
319 } else if line.starts_with("+++ ") {
320 line.replacen("+++ ", "--- ", 1)
321 } else if line.starts_with('+') && !line.starts_with("+++") {
322 format!("-{}", &line[1..])
323 } else if line.starts_with('-') && !line.starts_with("---") {
324 format!("+{}", &line[1..])
325 } else {
326 line.to_string()
327 }
328 })
329 .collect::<Vec<_>>()
330 .join("\n");
331 if diff.ends_with('\n') {
332 result.push('\n');
333 }
334 result
335}
336
337#[derive(Debug, Clone, PartialEq, Eq)]
338struct GranularEdit {
339 range: Range<usize>,
340 old_text: String,
341 new_text: String,
342}
343
344fn compute_granular_edits(old_text: &str, new_text: &str) -> Vec<GranularEdit> {
345 text_diff(old_text, new_text)
346 .into_iter()
347 .map(|(range, new_text)| GranularEdit {
348 old_text: old_text[range.clone()].to_string(),
349 range,
350 new_text: new_text.to_string(),
351 })
352 .collect()
353}
354
355#[derive(Debug, Clone)]
356struct HistoryAdditionRange {
357 range_in_current: Range<usize>,
358}
359
360#[derive(Debug, Clone)]
361struct HistoryDeletionRange {
362 deleted_text: String,
363 position_in_current: usize,
364}
365
366fn compute_history_addition_ranges(history_edits: &[GranularEdit]) -> Vec<HistoryAdditionRange> {
367 let mut result = Vec::new();
368 let mut offset_delta: isize = 0;
369
370 for edit in history_edits {
371 if !edit.new_text.is_empty() {
372 let new_start = (edit.range.start as isize + offset_delta) as usize;
373 let new_end = new_start + edit.new_text.len();
374 result.push(HistoryAdditionRange {
375 range_in_current: new_start..new_end,
376 });
377 }
378
379 offset_delta += edit.new_text.len() as isize - edit.old_text.len() as isize;
380 }
381
382 result
383}
384
385fn compute_history_deletion_ranges(history_edits: &[GranularEdit]) -> Vec<HistoryDeletionRange> {
386 let mut result = Vec::new();
387 let mut offset_delta: isize = 0;
388
389 for edit in history_edits {
390 if !edit.old_text.is_empty() {
391 let position_in_current = (edit.range.start as isize + offset_delta) as usize;
392 result.push(HistoryDeletionRange {
393 deleted_text: edit.old_text.clone(),
394 position_in_current,
395 });
396 }
397
398 offset_delta += edit.new_text.len() as isize - edit.old_text.len() as isize;
399 }
400
401 result
402}
403
404#[derive(Debug, Clone, Default, PartialEq, Eq)]
405struct ReversalOverlap {
406 chars_reversing_user_edits: usize,
407 total_chars_in_prediction: usize,
408}
409
410impl ReversalOverlap {
411 fn ratio(&self) -> f32 {
412 if self.total_chars_in_prediction == 0 {
413 0.0
414 } else {
415 self.chars_reversing_user_edits as f32 / self.total_chars_in_prediction as f32
416 }
417 }
418}
419
420/// Normalize edits where `old_text` appears as a subsequence within `new_text` (extension),
421/// or where `new_text` appears as a subsequence within `old_text` (reduction).
422///
423/// For extensions: when the user's text is preserved (in order) within the prediction,
424/// we only count the newly inserted characters, not the preserved ones.
425/// E.g., "epr" → "eprintln!()" becomes 8 inserted chars ("intln!()")
426/// E.g., "test_my_function" → "a_test_for_my_special_function_plz" becomes 18 inserted chars
427///
428/// For reductions: when the prediction's text is preserved (in order) within the original,
429/// we only count the deleted characters, not the preserved ones.
430/// E.g., "ifrom" → "from" becomes 1 deleted char ("i")
431fn normalize_extension_edits(edits: Vec<GranularEdit>) -> Vec<GranularEdit> {
432 edits
433 .into_iter()
434 .flat_map(|edit| {
435 if edit.old_text.is_empty() || edit.new_text.is_empty() {
436 return vec![edit];
437 }
438
439 // Use character-wise diff to find exact byte ranges of changes
440 let char_edits = char_diff(&edit.old_text, &edit.new_text);
441
442 let all_deletions = !char_edits.is_empty()
443 && char_edits
444 .iter()
445 .all(|(range, replacement)| !range.is_empty() && replacement.is_empty());
446 let all_insertions = !char_edits.is_empty()
447 && char_edits
448 .iter()
449 .all(|(range, replacement)| range.is_empty() && !replacement.is_empty());
450 if all_deletions || all_insertions {
451 return char_edits
452 .into_iter()
453 .map(|(range, replacement)| GranularEdit {
454 range: edit.range.start + range.start..edit.range.start + range.end,
455 old_text: edit.old_text[range].to_string(),
456 new_text: replacement.to_string(),
457 })
458 .collect();
459 }
460
461 // Otherwise, keep the original edit (mixed changes)
462 vec![edit]
463 })
464 .collect()
465}
466
467fn compute_reversal_overlap(
468 original_content: &str,
469 current_content: &str,
470 predicted_content: &str,
471) -> ReversalOverlap {
472 let history_edits =
473 normalize_extension_edits(compute_granular_edits(original_content, current_content));
474 let prediction_edits =
475 normalize_extension_edits(compute_granular_edits(current_content, predicted_content));
476
477 let history_addition_ranges = compute_history_addition_ranges(&history_edits);
478 let history_deletion_ranges = compute_history_deletion_ranges(&history_edits);
479
480 let reversed_additions =
481 compute_reversed_additions(&history_addition_ranges, &prediction_edits);
482 let restored_deletions =
483 compute_restored_deletions(&history_deletion_ranges, &prediction_edits);
484
485 let total_chars_in_prediction: usize = prediction_edits
486 .iter()
487 .map(|e| e.new_text.chars().count() + e.old_text.chars().count())
488 .sum();
489
490 ReversalOverlap {
491 chars_reversing_user_edits: reversed_additions + restored_deletions,
492 total_chars_in_prediction,
493 }
494}
495
496fn compute_reversed_additions(
497 history_addition_ranges: &[HistoryAdditionRange],
498 prediction_edits: &[GranularEdit],
499) -> usize {
500 let mut reversed_chars = 0;
501
502 for pred_edit in prediction_edits {
503 for history_addition in history_addition_ranges {
504 let overlap_start = pred_edit
505 .range
506 .start
507 .max(history_addition.range_in_current.start);
508 let overlap_end = pred_edit
509 .range
510 .end
511 .min(history_addition.range_in_current.end);
512
513 if overlap_start < overlap_end {
514 let relative_start = overlap_start - pred_edit.range.start;
515 let relative_end = overlap_end - pred_edit.range.start;
516 let overlap_text = &pred_edit.old_text[relative_start..relative_end];
517 reversed_chars += overlap_text.chars().count();
518 }
519 }
520 }
521
522 reversed_chars
523}
524
525fn compute_restored_deletions(
526 history_deletion_ranges: &[HistoryDeletionRange],
527 prediction_edits: &[GranularEdit],
528) -> usize {
529 let mut restored = 0;
530
531 for pred_edit in prediction_edits {
532 if pred_edit.new_text.is_empty() {
533 continue;
534 }
535
536 for deletion in history_deletion_ranges {
537 if pred_edit.range.contains(&deletion.position_in_current)
538 || deletion.position_in_current == pred_edit.range.start
539 {
540 restored += compute_lcs_length(&deletion.deleted_text, &pred_edit.new_text);
541 }
542 }
543 }
544
545 restored
546}
547
548fn compute_lcs_length(a: &str, b: &str) -> usize {
549 let a_chars: Vec<char> = a.chars().collect();
550 let b_chars: Vec<char> = b.chars().collect();
551 let m = a_chars.len();
552 let n = b_chars.len();
553
554 if m == 0 || n == 0 {
555 return 0;
556 }
557
558 let mut prev = vec![0; n + 1];
559 let mut curr = vec![0; n + 1];
560
561 for i in 1..=m {
562 for j in 1..=n {
563 if a_chars[i - 1] == b_chars[j - 1] {
564 curr[j] = prev[j - 1] + 1;
565 } else {
566 curr[j] = prev[j].max(curr[j - 1]);
567 }
568 }
569 std::mem::swap(&mut prev, &mut curr);
570 curr.fill(0);
571 }
572
573 prev[n]
574}
575
576fn filter_edit_history_by_path<'a>(
577 edit_history: &'a [Arc<zeta_prompt::Event>],
578 cursor_path: &std::path::Path,
579) -> Vec<&'a zeta_prompt::Event> {
580 edit_history
581 .iter()
582 .filter(|event| match event.as_ref() {
583 zeta_prompt::Event::BufferChange { path, .. } => {
584 let event_path = path.as_ref();
585 if event_path == cursor_path {
586 return true;
587 }
588 let stripped = event_path
589 .components()
590 .skip(1)
591 .collect::<std::path::PathBuf>();
592 stripped == cursor_path
593 }
594 })
595 .map(|arc| arc.as_ref())
596 .collect()
597}
598
599fn extract_diff_from_event(event: &zeta_prompt::Event) -> &str {
600 match event {
601 zeta_prompt::Event::BufferChange { diff, .. } => diff.as_str(),
602 }
603}
604
605fn is_predicted_event(event: &zeta_prompt::Event) -> bool {
606 match event {
607 zeta_prompt::Event::BufferChange { predicted, .. } => *predicted,
608 }
609}
610
611pub fn compute_prediction_reversal_ratio(
612 prompt_inputs: &ZetaPromptInput,
613 predicted_content: &str,
614 cursor_path: &Path,
615) -> f32 {
616 let current_content: &str = prompt_inputs.cursor_excerpt.as_ref();
617
618 let edit_history: &[Arc<zeta_prompt::Event>] = &prompt_inputs.events;
619 let relevant_events = filter_edit_history_by_path(edit_history, cursor_path);
620
621 let most_recent = match relevant_events.last() {
622 Some(event) if !is_predicted_event(event) => *event,
623 _ => return 0.0,
624 };
625
626 let diff = extract_diff_from_event(most_recent);
627 if diff.is_empty() {
628 return 0.0;
629 }
630
631 if let Some(excerpt_start_row) = prompt_inputs.excerpt_start_row {
632 let diffs = vec![diff];
633 let overlap = compute_excerpt_aware_reversal_overlap(
634 &diffs,
635 current_content,
636 excerpt_start_row,
637 predicted_content,
638 );
639 return overlap.ratio();
640 }
641
642 let reversed = reverse_diff(diff);
643 let with_headers = format!("--- a/file\n+++ b/file\n{}", reversed);
644 let original_content = match apply_diff_to_string(&with_headers, current_content) {
645 Ok(updated_content) => updated_content,
646 Err(_) => apply_diff_to_string_lenient(&reversed, current_content),
647 };
648
649 let overlap = compute_reversal_overlap(&original_content, current_content, predicted_content);
650 overlap.ratio()
651}
652
653#[cfg(test)]
654mod tests {
655 use super::*;
656 use edit_prediction::udiff::apply_diff_to_string;
657 use indoc::indoc;
658 use zeta_prompt::ExcerptRanges;
659
660 fn make_test_prompt_inputs(
661 content: &str,
662 events: Vec<Arc<zeta_prompt::Event>>,
663 excerpt_start_row: Option<u32>,
664 ) -> ZetaPromptInput {
665 ZetaPromptInput {
666 cursor_path: Arc::from(Path::new("src/test.rs")),
667 cursor_excerpt: content.into(),
668 cursor_offset_in_excerpt: 0,
669 excerpt_start_row,
670 events,
671 related_files: Vec::new(),
672 excerpt_ranges: ExcerptRanges {
673 editable_150: 0..content.len(),
674 editable_180: 0..content.len(),
675 editable_350: 0..content.len(),
676 editable_150_context_350: 0..content.len(),
677 editable_180_context_350: 0..content.len(),
678 editable_350_context_150: 0..content.len(),
679 ..Default::default()
680 },
681 experiment: None,
682 in_open_source_repo: false,
683 can_collect_data: false,
684 }
685 }
686
687 #[test]
688 fn test_reversal_overlap() {
689 struct Case {
690 name: &'static str,
691 original: &'static str,
692 current: &'static str,
693 predicted: &'static str,
694 expected_reversal_chars: usize,
695 expected_total_chars: usize,
696 }
697
698 let cases = [
699 Case {
700 name: "user_adds_line_prediction_removes_it",
701 original: indoc! {"
702 a
703 b
704 c"},
705 current: indoc! {"
706 a
707 new line
708 b
709 c"},
710 predicted: indoc! {"
711 a
712 b
713 c"},
714 expected_reversal_chars: 9,
715 expected_total_chars: 9,
716 },
717 Case {
718 name: "user_deletes_line_prediction_restores_it",
719 original: indoc! {"
720 a
721 deleted
722 b"},
723 current: indoc! {"
724 a
725 b"},
726 predicted: indoc! {"
727 a
728 deleted
729 b"},
730 expected_reversal_chars: 8,
731 expected_total_chars: 8,
732 },
733 Case {
734 name: "user_deletes_text_prediction_restores_partial",
735 original: "hello beautiful world",
736 current: "hello world",
737 predicted: "hello beautiful world",
738 expected_reversal_chars: 10,
739 expected_total_chars: 10,
740 },
741 Case {
742 name: "user_deletes_foo_prediction_adds_bar",
743 original: "foo",
744 current: "",
745 predicted: "bar",
746 expected_reversal_chars: 0,
747 expected_total_chars: 3,
748 },
749 Case {
750 name: "independent_edits_different_locations",
751 original: indoc! {"
752 line1
753 line2
754 line3"},
755 current: indoc! {"
756 LINE1
757 line2
758 line3"},
759 predicted: indoc! {"
760 LINE1
761 line2
762 LINE3"},
763 expected_reversal_chars: 0,
764 expected_total_chars: 10,
765 },
766 Case {
767 name: "no_history_edits",
768 original: "same",
769 current: "same",
770 predicted: "different",
771 expected_reversal_chars: 0,
772 expected_total_chars: 13,
773 },
774 Case {
775 name: "user_replaces_text_prediction_reverses",
776 original: indoc! {"
777 keep
778 delete_me
779 keep2"},
780 current: indoc! {"
781 keep
782 added
783 keep2"},
784 predicted: indoc! {"
785 keep
786 delete_me
787 keep2"},
788 expected_reversal_chars: 14,
789 expected_total_chars: 14,
790 },
791 Case {
792 name: "user_modifies_word_prediction_modifies_differently",
793 original: "the quick brown fox",
794 current: "the slow brown fox",
795 predicted: "the fast brown fox",
796 expected_reversal_chars: 4,
797 expected_total_chars: 8,
798 },
799 Case {
800 name: "user finishes function name (suffix)",
801 original: "",
802 current: "epr",
803 predicted: "eprintln!()",
804 expected_reversal_chars: 0,
805 expected_total_chars: 8,
806 },
807 Case {
808 name: "user starts function name (prefix)",
809 original: "",
810 current: "my_function()",
811 predicted: "test_my_function()",
812 expected_reversal_chars: 0,
813 expected_total_chars: 5,
814 },
815 Case {
816 name: "user types partial, prediction extends in multiple places",
817 original: "",
818 current: "test_my_function",
819 predicted: "a_test_for_my_special_function_plz",
820 expected_reversal_chars: 0,
821 expected_total_chars: 18,
822 },
823 // Edge cases for subsequence matching
824 Case {
825 name: "subsequence with interleaved underscores",
826 original: "",
827 current: "a_b_c",
828 predicted: "_a__b__c__",
829 expected_reversal_chars: 0,
830 expected_total_chars: 5,
831 },
832 Case {
833 name: "not a subsequence - different characters",
834 original: "",
835 current: "abc",
836 predicted: "xyz",
837 expected_reversal_chars: 3,
838 expected_total_chars: 6,
839 },
840 Case {
841 name: "not a subsequence - wrong order",
842 original: "",
843 current: "abc",
844 predicted: "cba",
845 expected_reversal_chars: 3,
846 expected_total_chars: 6,
847 },
848 Case {
849 name: "partial subsequence - only some chars match",
850 original: "",
851 current: "abcd",
852 predicted: "axbx",
853 expected_reversal_chars: 4,
854 expected_total_chars: 8,
855 },
856 // Common completion patterns
857 Case {
858 name: "completing a method call",
859 original: "",
860 current: "vec.pu",
861 predicted: "vec.push(item)",
862 expected_reversal_chars: 0,
863 expected_total_chars: 8,
864 },
865 Case {
866 name: "completing an import statement",
867 original: "",
868 current: "use std::col",
869 predicted: "use std::collections::HashMap",
870 expected_reversal_chars: 0,
871 expected_total_chars: 17,
872 },
873 Case {
874 name: "completing a struct field",
875 original: "",
876 current: "name: St",
877 predicted: "name: String",
878 expected_reversal_chars: 0,
879 expected_total_chars: 4,
880 },
881 Case {
882 name: "prediction replaces with completely different text",
883 original: "",
884 current: "hello",
885 predicted: "world",
886 expected_reversal_chars: 5,
887 expected_total_chars: 10,
888 },
889 Case {
890 name: "empty prediction removes user text",
891 original: "",
892 current: "mistake",
893 predicted: "",
894 expected_reversal_chars: 7,
895 expected_total_chars: 7,
896 },
897 Case {
898 name: "fixing typo is not reversal",
899 original: "",
900 current: "<dv",
901 predicted: "<div>",
902 expected_reversal_chars: 0,
903 expected_total_chars: 2,
904 },
905 Case {
906 name: "infix insertion not reversal",
907 original: indoc! {"
908 from my_project import Foo
909 "},
910 current: indoc! {"
911 ifrom my_project import Foo
912 "},
913 predicted: indoc! {"
914 import
915 from my_project import Foo
916 "},
917 expected_reversal_chars: 0,
918 expected_total_chars: 6,
919 },
920 Case {
921 name: "non-word based reversal",
922 original: "from",
923 current: "ifrom",
924 predicted: "from",
925 expected_reversal_chars: 1,
926 expected_total_chars: 1,
927 },
928 Case {
929 name: "multiple insertions no reversal",
930 original: "print(\"Hello, World!\")",
931 current: "sys.(\"Hello, World!\")",
932 predicted: "sys.stdout.write(\"Hello, World!\\n\")",
933 expected_reversal_chars: 0,
934 expected_total_chars: 14,
935 },
936 ];
937
938 for case in &cases {
939 let overlap = compute_reversal_overlap(case.original, case.current, case.predicted);
940 assert_eq!(
941 overlap.chars_reversing_user_edits, case.expected_reversal_chars,
942 "Test '{}': expected {} reversal chars, got {}",
943 case.name, case.expected_reversal_chars, overlap.chars_reversing_user_edits
944 );
945 assert_eq!(
946 overlap.total_chars_in_prediction, case.expected_total_chars,
947 "Test '{}': expected {} total chars, got {}",
948 case.name, case.expected_total_chars, overlap.total_chars_in_prediction
949 );
950 }
951 }
952
953 #[test]
954 fn test_reverse_diff() {
955 let forward_diff = indoc! {"
956 --- a/file.rs
957 +++ b/file.rs
958 @@ -1,3 +1,4 @@
959 fn main() {
960 + let x = 42;
961 println!(\"hello\");
962 }"};
963
964 let reversed = reverse_diff(forward_diff);
965
966 assert!(
967 reversed.contains("+++ a/file.rs"),
968 "Should have +++ for old path"
969 );
970 assert!(
971 reversed.contains("--- b/file.rs"),
972 "Should have --- for new path"
973 );
974 assert!(
975 reversed.contains("- let x = 42;"),
976 "Added line should become deletion"
977 );
978 assert!(
979 reversed.contains(" fn main()"),
980 "Context lines should be unchanged"
981 );
982 }
983
984 #[test]
985 fn test_reverse_diff_roundtrip() {
986 // Applying a diff and then its reverse should get back to original
987 let original = indoc! {"
988 first line
989 hello world
990 last line
991 "};
992 let modified = indoc! {"
993 first line
994 hello beautiful world
995 last line
996 "};
997
998 // unified_diff doesn't include file headers, but apply_diff_to_string needs them
999 let diff_body = language::unified_diff(original, modified);
1000 let forward_diff = format!("--- a/file\n+++ b/file\n{}", diff_body);
1001 let reversed_diff = reverse_diff(&forward_diff);
1002
1003 // Apply forward diff to original
1004 let after_forward = apply_diff_to_string(&forward_diff, original).unwrap();
1005 assert_eq!(after_forward, modified);
1006
1007 // Apply reversed diff to modified
1008 let after_reverse = apply_diff_to_string(&reversed_diff, &after_forward).unwrap();
1009 assert_eq!(after_reverse, original);
1010 }
1011
1012 #[test]
1013 fn test_filter_edit_history_by_path() {
1014 // Test that filter_edit_history_by_path correctly matches paths when
1015 // the edit history has paths with a repo prefix (e.g., "repo/src/file.rs")
1016 // but the cursor_path doesn't have the repo prefix (e.g., "src/file.rs")
1017 let events = vec![
1018 Arc::new(zeta_prompt::Event::BufferChange {
1019 path: Arc::from(Path::new("myrepo/src/file.rs")),
1020 old_path: Arc::from(Path::new("myrepo/src/file.rs")),
1021 diff: indoc! {"
1022 @@ -1 +1 @@
1023 -old
1024 +new"}
1025 .into(),
1026 predicted: false,
1027 in_open_source_repo: true,
1028 }),
1029 Arc::new(zeta_prompt::Event::BufferChange {
1030 path: Arc::from(Path::new("myrepo/other.rs")),
1031 old_path: Arc::from(Path::new("myrepo/other.rs")),
1032 diff: indoc! {"
1033 @@ -1 +1 @@
1034 -a
1035 +b"}
1036 .into(),
1037 predicted: false,
1038 in_open_source_repo: true,
1039 }),
1040 Arc::new(zeta_prompt::Event::BufferChange {
1041 path: Arc::from(Path::new("src/file.rs")),
1042 old_path: Arc::from(Path::new("src/file.rs")),
1043 diff: indoc! {"
1044 @@ -1 +1 @@
1045 -x
1046 +y"}
1047 .into(),
1048 predicted: false,
1049 in_open_source_repo: true,
1050 }),
1051 ];
1052
1053 // "myrepo/src/file.rs" stripped -> "src/file.rs" matches cursor_path
1054 // "src/file.rs" exact match
1055 let cursor_path = Path::new("src/file.rs");
1056 let filtered = filter_edit_history_by_path(&events, cursor_path);
1057 assert_eq!(
1058 filtered.len(),
1059 2,
1060 "Should match myrepo/src/file.rs (stripped) and src/file.rs (exact)"
1061 );
1062
1063 // "myrepo/src/file.rs" stripped -> "src/file.rs" != "file.rs"
1064 // "src/file.rs" stripped -> "file.rs" == "file.rs"
1065 let cursor_path = Path::new("file.rs");
1066 let filtered = filter_edit_history_by_path(&events, cursor_path);
1067 assert_eq!(
1068 filtered.len(),
1069 1,
1070 "Should only match src/file.rs (stripped to file.rs)"
1071 );
1072
1073 // "myrepo/other.rs" stripped -> "other.rs" == "other.rs"
1074 let cursor_path = Path::new("other.rs");
1075 let filtered = filter_edit_history_by_path(&events, cursor_path);
1076 assert_eq!(filtered.len(), 1, "Should match only myrepo/other.rs");
1077 }
1078
1079 #[test]
1080 fn test_reverse_diff_preserves_trailing_newline() {
1081 let diff_with_trailing_newline = indoc! {"
1082 --- a/file
1083 +++ b/file
1084 @@ -1 +1 @@
1085 -old
1086 +new
1087 "};
1088 let reversed = reverse_diff(diff_with_trailing_newline);
1089 assert!(
1090 reversed.ends_with('\n'),
1091 "Reversed diff should preserve trailing newline"
1092 );
1093
1094 let diff_without_trailing_newline = indoc! {"
1095 --- a/file
1096 +++ b/file
1097 @@ -1 +1 @@
1098 -old
1099 +new"};
1100 let reversed = reverse_diff(diff_without_trailing_newline);
1101 assert!(
1102 !reversed.ends_with('\n'),
1103 "Reversed diff should not add trailing newline if original didn't have one"
1104 );
1105 }
1106
1107 #[test]
1108 fn test_filter_hunks_by_excerpt_region() {
1109 struct Case {
1110 name: &'static str,
1111 diff: &'static str,
1112 excerpt_start_row: u32,
1113 excerpt_row_count: u32,
1114 expected_filtered_diff: &'static str,
1115 expected_line_offset: i32,
1116 }
1117
1118 let cases = [
1119 Case {
1120 name: "hunk_entirely_before_excerpt",
1121 diff: indoc! {"
1122 @@ -1,3 +1,4 @@
1123 line1
1124 +inserted
1125 line2
1126 line3
1127 "},
1128 excerpt_start_row: 10,
1129 excerpt_row_count: 5,
1130 expected_filtered_diff: "",
1131 expected_line_offset: 1,
1132 },
1133 Case {
1134 name: "hunk_entirely_inside_excerpt",
1135 diff: indoc! {"
1136 @@ -12,3 +12,4 @@
1137 line12
1138 +inserted
1139 line13
1140 line14
1141 "},
1142 excerpt_start_row: 10,
1143 excerpt_row_count: 10,
1144 expected_filtered_diff: indoc! {"
1145 @@ -2,3 +2,4 @@
1146 line12
1147 +inserted
1148 line13
1149 line14
1150 "},
1151 expected_line_offset: 1,
1152 },
1153 Case {
1154 name: "hunk_entirely_after_excerpt",
1155 diff: indoc! {"
1156 @@ -50,3 +50,4 @@
1157 line50
1158 +inserted
1159 line51
1160 line52
1161 "},
1162 excerpt_start_row: 10,
1163 excerpt_row_count: 5,
1164 expected_filtered_diff: "",
1165 expected_line_offset: 0,
1166 },
1167 Case {
1168 name: "hunk_straddles_excerpt_start",
1169 diff: indoc! {"
1170 @@ -8,5 +8,6 @@
1171 line8
1172 line9
1173 +inserted
1174 line10
1175 line11
1176 line12
1177 "},
1178 excerpt_start_row: 10,
1179 excerpt_row_count: 10,
1180 expected_filtered_diff: indoc! {"
1181 @@ -1,3 +1,3 @@
1182 line10
1183 line11
1184 line12
1185 "},
1186 expected_line_offset: 1,
1187 },
1188 Case {
1189 name: "hunk_straddles_excerpt_end",
1190 diff: indoc! {"
1191 @@ -18,5 +18,6 @@
1192 line18
1193 line19
1194 +inserted
1195 line20
1196 line21
1197 line22
1198 "},
1199 excerpt_start_row: 10,
1200 excerpt_row_count: 10,
1201 expected_filtered_diff: indoc! {"
1202 @@ -8,2 +8,3 @@
1203 line18
1204 line19
1205 +inserted
1206 "},
1207 expected_line_offset: 1,
1208 },
1209 Case {
1210 name: "multiple_hunks_mixed",
1211 diff: indoc! {"
1212 @@ -1,2 +1,3 @@
1213 line1
1214 +before_excerpt
1215 line2
1216 @@ -12,2 +13,3 @@
1217 line12
1218 +inside_excerpt
1219 line13
1220 @@ -50,2 +52,3 @@
1221 line50
1222 +after_excerpt
1223 line51
1224 "},
1225 excerpt_start_row: 10,
1226 excerpt_row_count: 10,
1227 expected_filtered_diff: indoc! {"
1228 @@ -3,2 +3,3 @@
1229 line12
1230 +inside_excerpt
1231 line13
1232 "},
1233 expected_line_offset: 2,
1234 },
1235 Case {
1236 name: "deletion_before_excerpt",
1237 diff: indoc! {"
1238 @@ -1,4 +1,3 @@
1239 line1
1240 -deleted
1241 line2
1242 line3
1243 "},
1244 excerpt_start_row: 10,
1245 excerpt_row_count: 5,
1246 expected_filtered_diff: "",
1247 expected_line_offset: -1,
1248 },
1249 Case {
1250 name: "deletion_inside_excerpt",
1251 diff: indoc! {"
1252 @@ -12,4 +12,3 @@
1253 line12
1254 -deleted
1255 line13
1256 line14
1257 "},
1258 excerpt_start_row: 10,
1259 excerpt_row_count: 10,
1260 expected_filtered_diff: indoc! {"
1261 @@ -2,4 +2,3 @@
1262 line12
1263 -deleted
1264 line13
1265 line14
1266 "},
1267 expected_line_offset: -1,
1268 },
1269 Case {
1270 name: "empty_diff",
1271 diff: "",
1272 excerpt_start_row: 10,
1273 excerpt_row_count: 5,
1274 expected_filtered_diff: "",
1275 expected_line_offset: 0,
1276 },
1277 Case {
1278 name: "hunk_spans_entire_excerpt",
1279 diff: indoc! {"
1280 @@ -8,10 +8,12 @@
1281 line8
1282 line9
1283 line10
1284 line11
1285 +inserted1
1286 line12
1287 line13
1288 +inserted2
1289 line14
1290 line15
1291 line16
1292 line17
1293 "},
1294 excerpt_start_row: 10,
1295 excerpt_row_count: 5,
1296 expected_filtered_diff: indoc! {"
1297 @@ -1,3 +1,5 @@
1298 line11
1299 +inserted1
1300 line12
1301 line13
1302 +inserted2
1303 "},
1304 expected_line_offset: 2,
1305 },
1306 Case {
1307 name: "replacement_inside_excerpt",
1308 diff: indoc! {"
1309 @@ -12,3 +12,3 @@
1310 line12
1311 -old_text
1312 +new_text
1313 line14
1314 "},
1315 excerpt_start_row: 10,
1316 excerpt_row_count: 10,
1317 expected_filtered_diff: indoc! {"
1318 @@ -2,3 +2,3 @@
1319 line12
1320 -old_text
1321 +new_text
1322 line14
1323 "},
1324 expected_line_offset: 0,
1325 },
1326 ];
1327
1328 for case in &cases {
1329 let (filtered, line_offset) = filter_diff_hunks_by_excerpt(
1330 case.diff,
1331 case.excerpt_start_row,
1332 case.excerpt_row_count,
1333 );
1334 assert_eq!(
1335 filtered, case.expected_filtered_diff,
1336 "Test '{}': filtered diff mismatch.\nExpected:\n{}\nGot:\n{}",
1337 case.name, case.expected_filtered_diff, filtered
1338 );
1339 assert_eq!(
1340 line_offset, case.expected_line_offset,
1341 "Test '{}': line offset mismatch. Expected {}, got {}",
1342 case.name, case.expected_line_offset, line_offset
1343 );
1344 }
1345 }
1346
1347 #[test]
1348 fn test_excerpt_aware_reversal_tracking() {
1349 struct Case {
1350 name: &'static str,
1351 edit_history_diffs: Vec<&'static str>,
1352 excerpt_content: &'static str,
1353 excerpt_start_row: u32,
1354 predicted_content: &'static str,
1355 expected_reversal_chars: usize,
1356 expected_total_chars: usize,
1357 }
1358
1359 let cases = [
1360 Case {
1361 name: "edit_outside_excerpt_no_reversal",
1362 edit_history_diffs: vec![indoc! {"
1363 @@ -1,2 +1,3 @@
1364 line1
1365 +added_outside
1366 line2
1367 "}],
1368 excerpt_content: indoc! {"
1369 line10
1370 line11
1371 line12
1372 "},
1373 excerpt_start_row: 10,
1374 predicted_content: indoc! {"
1375 line10
1376 modified
1377 line12
1378 "},
1379 expected_reversal_chars: 0,
1380 expected_total_chars: 14,
1381 },
1382 Case {
1383 name: "edit_inside_excerpt_with_reversal",
1384 edit_history_diffs: vec![indoc! {"
1385 @@ -10,3 +10,4 @@
1386 line10
1387 +user_added
1388 line11
1389 line12
1390 "}],
1391 excerpt_content: indoc! {"
1392 line10
1393 user_added
1394 line11
1395 line12
1396 "},
1397 excerpt_start_row: 10,
1398 predicted_content: indoc! {"
1399 line10
1400 line11
1401 line12
1402 "},
1403 expected_reversal_chars: 11,
1404 expected_total_chars: 11,
1405 },
1406 Case {
1407 name: "straddling_edit_partial_reversal",
1408 edit_history_diffs: vec![indoc! {"
1409 @@ -8,6 +8,8 @@
1410 line8
1411 line9
1412 +before_excerpt
1413 line10
1414 +inside_excerpt
1415 line11
1416 line12
1417 line13
1418 "}],
1419 excerpt_content: indoc! {"
1420 line10
1421 inside_excerpt
1422 line11
1423 line12
1424 line13
1425 "},
1426 excerpt_start_row: 10,
1427 predicted_content: indoc! {"
1428 line10
1429 line11
1430 line12
1431 line13
1432 "},
1433 expected_reversal_chars: 15,
1434 expected_total_chars: 15,
1435 },
1436 Case {
1437 name: "multiple_edits_mixed_locations",
1438 edit_history_diffs: vec![
1439 indoc! {"
1440 @@ -1,2 +1,3 @@
1441 line1
1442 +outside1
1443 line2
1444 "},
1445 indoc! {"
1446 @@ -11,2 +12,3 @@
1447 line11
1448 +inside1
1449 line12
1450 "},
1451 ],
1452 excerpt_content: indoc! {"
1453 line10
1454 line11
1455 inside1
1456 line12
1457 line13
1458 "},
1459 excerpt_start_row: 10,
1460 predicted_content: indoc! {"
1461 line10
1462 line11
1463 line12
1464 line13
1465 "},
1466 expected_reversal_chars: 8,
1467 expected_total_chars: 8,
1468 },
1469 Case {
1470 name: "no_edit_history",
1471 edit_history_diffs: vec![],
1472 excerpt_content: indoc! {"
1473 line10
1474 line11
1475 line12
1476 "},
1477 excerpt_start_row: 10,
1478 predicted_content: indoc! {"
1479 line10
1480 modified
1481 line12
1482 "},
1483 expected_reversal_chars: 0,
1484 expected_total_chars: 14,
1485 },
1486 Case {
1487 name: "edit_after_excerpt_no_effect",
1488 edit_history_diffs: vec![indoc! {"
1489 @@ -50,2 +50,3 @@
1490 line50
1491 +added_after
1492 line51
1493 "}],
1494 excerpt_content: indoc! {"
1495 line10
1496 line11
1497 line12
1498 "},
1499 excerpt_start_row: 10,
1500 predicted_content: indoc! {"
1501 line10
1502 changed
1503 line12
1504 "},
1505 expected_reversal_chars: 0,
1506 expected_total_chars: 13,
1507 },
1508 Case {
1509 name: "line_offset_tracking_across_hunks",
1510 edit_history_diffs: vec![
1511 indoc! {"
1512 @@ -1,2 +1,4 @@
1513 line1
1514 +added1
1515 +added2
1516 line2
1517 "},
1518 indoc! {"
1519 @@ -12,2 +14,3 @@
1520 line12
1521 +inside_after_offset
1522 line13
1523 "},
1524 ],
1525 excerpt_content: indoc! {"
1526 line10
1527 line11
1528 line12
1529 inside_after_offset
1530 line13
1531 "},
1532 excerpt_start_row: 10,
1533 predicted_content: indoc! {"
1534 line10
1535 line11
1536 line12
1537 line13
1538 "},
1539 expected_reversal_chars: 20,
1540 expected_total_chars: 20,
1541 },
1542 ];
1543
1544 for case in &cases {
1545 let overlap = compute_excerpt_aware_reversal_overlap(
1546 &case.edit_history_diffs,
1547 case.excerpt_content,
1548 case.excerpt_start_row,
1549 case.predicted_content,
1550 );
1551 assert_eq!(
1552 overlap.chars_reversing_user_edits, case.expected_reversal_chars,
1553 "Test '{}': expected {} reversal chars, got {}",
1554 case.name, case.expected_reversal_chars, overlap.chars_reversing_user_edits
1555 );
1556 assert_eq!(
1557 overlap.total_chars_in_prediction, case.expected_total_chars,
1558 "Test '{}': expected {} total chars, got {}",
1559 case.name, case.expected_total_chars, overlap.total_chars_in_prediction
1560 );
1561 }
1562 }
1563
1564 #[test]
1565 fn test_lenient_diff_application() {
1566 struct Case {
1567 name: &'static str,
1568 diff: &'static str,
1569 content: &'static str,
1570 expected_result: &'static str,
1571 }
1572
1573 let cases = [
1574 Case {
1575 name: "hunk_context_not_found_skipped",
1576 diff: indoc! {"
1577 @@ -1,3 +1,4 @@
1578 context_not_in_content
1579 +added_line
1580 more_context
1581 final_context
1582 "},
1583 content: indoc! {"
1584 completely
1585 different
1586 content
1587 "},
1588 expected_result: indoc! {"
1589 completely
1590 different
1591 content
1592 "},
1593 },
1594 Case {
1595 name: "hunk_context_found_applied",
1596 diff: indoc! {"
1597 @@ -1,3 +1,4 @@
1598 line1
1599 +inserted
1600 line2
1601 line3
1602 "},
1603 content: indoc! {"
1604 line1
1605 line2
1606 line3
1607 "},
1608 expected_result: indoc! {"
1609 line1
1610 inserted
1611 line2
1612 line3
1613 "},
1614 },
1615 Case {
1616 name: "multiple_hunks_partial_match",
1617 diff: indoc! {"
1618 @@ -1,2 +1,3 @@
1619 not_found
1620 +skipped
1621 also_not_found
1622 @@ -5,2 +6,3 @@
1623 line5
1624 +applied
1625 line6
1626 "},
1627 content: indoc! {"
1628 line1
1629 line2
1630 line3
1631 line4
1632 line5
1633 line6
1634 "},
1635 expected_result: indoc! {"
1636 line1
1637 line2
1638 line3
1639 line4
1640 line5
1641 applied
1642 line6
1643 "},
1644 },
1645 Case {
1646 name: "empty_diff",
1647 diff: "",
1648 content: indoc! {"
1649 unchanged
1650 content
1651 "},
1652 expected_result: indoc! {"
1653 unchanged
1654 content
1655 "},
1656 },
1657 ];
1658
1659 for case in &cases {
1660 let result = apply_diff_to_string_lenient(case.diff, case.content);
1661 assert_eq!(
1662 result, case.expected_result,
1663 "Test '{}': expected:\n{}\ngot:\n{}",
1664 case.name, case.expected_result, result
1665 );
1666 }
1667 }
1668
1669 #[test]
1670 fn test_unicode_reversal_overlap() {
1671 struct Case {
1672 name: &'static str,
1673 original: &'static str,
1674 current: &'static str,
1675 predicted: &'static str,
1676 expected_reversal_chars: usize,
1677 expected_total_chars: usize,
1678 }
1679
1680 let cases = [
1681 Case {
1682 name: "unicode_extension_cjk",
1683 original: "",
1684 current: "日", // 1 char
1685 predicted: "日本語", // 3 chars, adds 2 chars
1686 expected_reversal_chars: 0,
1687 expected_total_chars: 2, // "本語" = 2 chars added
1688 },
1689 Case {
1690 name: "unicode_extension_emoji",
1691 original: "",
1692 current: "🎉", // 1 char
1693 predicted: "🎉🎊🎈", // 3 chars, adds 2 chars
1694 expected_reversal_chars: 0,
1695 expected_total_chars: 2, // "🎊🎈" = 2 chars added
1696 },
1697 Case {
1698 name: "unicode_deletion_restored",
1699 original: "héllo wörld", // 11 chars
1700 current: "héllo", // 5 chars
1701 predicted: "héllo wörld", // restores " wörld" = 6 chars
1702 expected_reversal_chars: 6, // LCS(" wörld", " wörld") = 6 chars
1703 expected_total_chars: 6,
1704 },
1705 Case {
1706 name: "unicode_addition_reversed",
1707 original: "café", // 4 chars
1708 current: "café latté", // 10 chars, added " latté" = 6 chars
1709 predicted: "café", // removes " latté"
1710 expected_reversal_chars: 6, // 6 chars removed
1711 expected_total_chars: 6,
1712 },
1713 Case {
1714 name: "mixed_ascii_unicode",
1715 original: "",
1716 current: "test日本", // 6 chars
1717 predicted: "test日本語です", // 9 chars
1718 expected_reversal_chars: 0,
1719 expected_total_chars: 3, // 3 new chars after subsequence normalization
1720 },
1721 Case {
1722 name: "unicode_replacement_not_subsequence",
1723 original: "",
1724 current: "日本", // 2 chars
1725 predicted: "中国", // 2 chars, different
1726 expected_reversal_chars: 2, // removes "日本" = 2 chars
1727 expected_total_chars: 4, // 2 removed + 2 added
1728 },
1729 ];
1730
1731 for case in &cases {
1732 let overlap = compute_reversal_overlap(case.original, case.current, case.predicted);
1733 assert_eq!(
1734 overlap.chars_reversing_user_edits, case.expected_reversal_chars,
1735 "Test '{}': expected {} reversal chars, got {}",
1736 case.name, case.expected_reversal_chars, overlap.chars_reversing_user_edits
1737 );
1738 assert_eq!(
1739 overlap.total_chars_in_prediction, case.expected_total_chars,
1740 "Test '{}': expected {} total chars, got {}",
1741 case.name, case.expected_total_chars, overlap.total_chars_in_prediction
1742 );
1743 }
1744 }
1745
1746 #[test]
1747 fn test_compute_lcs_length() {
1748 assert_eq!(compute_lcs_length("", ""), 0);
1749 assert_eq!(compute_lcs_length("abc", ""), 0);
1750 assert_eq!(compute_lcs_length("", "abc"), 0);
1751 assert_eq!(compute_lcs_length("abc", "abc"), 3);
1752 assert_eq!(compute_lcs_length("abc", "def"), 0);
1753 assert_eq!(compute_lcs_length("abcdef", "ace"), 3);
1754 assert_eq!(compute_lcs_length("AGGTAB", "GXTXAYB"), 4);
1755 assert_eq!(compute_lcs_length("日本語", "日語"), 2);
1756 }
1757
1758 #[test]
1759 fn test_compute_prediction_reversal_ratio_full_file() {
1760 let prompt_inputs = make_test_prompt_inputs(
1761 indoc! {"
1762 line1
1763 user_added
1764 line2
1765 "},
1766 vec![Arc::new(zeta_prompt::Event::BufferChange {
1767 path: Arc::from(Path::new("src/test.rs")),
1768 old_path: Arc::from(Path::new("src/test.rs")),
1769 diff: indoc! {"
1770 @@ -1,2 +1,3 @@
1771 line1
1772 +user_added
1773 line2
1774 "}
1775 .into(),
1776 predicted: false,
1777 in_open_source_repo: false,
1778 })],
1779 None,
1780 );
1781
1782 let predicted = indoc! {"
1783 line1
1784 line2
1785 "};
1786 let ratio =
1787 compute_prediction_reversal_ratio(&prompt_inputs, predicted, Path::new("src/test.rs"));
1788
1789 assert!(
1790 ratio > 0.9,
1791 "Expected high reversal ratio when prediction removes user addition, got {}",
1792 ratio
1793 );
1794 }
1795
1796 #[test]
1797 fn test_compute_prediction_reversal_ratio_with_excerpt() {
1798 let prompt_inputs = make_test_prompt_inputs(
1799 indoc! {"
1800 line10
1801 user_added
1802 line11
1803 "},
1804 vec![Arc::new(zeta_prompt::Event::BufferChange {
1805 path: Arc::from(Path::new("src/test.rs")),
1806 old_path: Arc::from(Path::new("src/test.rs")),
1807 diff: indoc! {"
1808 @@ -10,2 +10,3 @@
1809 line10
1810 +user_added
1811 line11
1812 "}
1813 .into(),
1814 predicted: false,
1815 in_open_source_repo: false,
1816 })],
1817 Some(10),
1818 );
1819
1820 let predicted = indoc! {"
1821 line10
1822 line11
1823 "};
1824 let ratio =
1825 compute_prediction_reversal_ratio(&prompt_inputs, predicted, Path::new("src/test.rs"));
1826
1827 assert!(
1828 ratio > 0.9,
1829 "Expected high reversal ratio for excerpt-aware computation, got {}",
1830 ratio
1831 );
1832 }
1833
1834 #[test]
1835 fn test_compute_prediction_reversal_ratio_no_history() {
1836 let prompt_inputs = make_test_prompt_inputs(
1837 indoc! {"
1838 original content
1839 "},
1840 vec![],
1841 None,
1842 );
1843
1844 let predicted = indoc! {"
1845 completely different
1846 "};
1847 let ratio =
1848 compute_prediction_reversal_ratio(&prompt_inputs, predicted, Path::new("src/test.rs"));
1849
1850 assert_eq!(
1851 ratio, 0.0,
1852 "Expected zero reversal ratio with no edit history"
1853 );
1854 }
1855
1856 #[test]
1857 fn test_compute_prediction_reversal_ratio_path_filtering() {
1858 let prompt_inputs = make_test_prompt_inputs(
1859 indoc! {"
1860 line1
1861 user_added
1862 line2
1863 "},
1864 vec![Arc::new(zeta_prompt::Event::BufferChange {
1865 path: Arc::from(Path::new("src/other.rs")),
1866 old_path: Arc::from(Path::new("src/other.rs")),
1867 diff: indoc! {"
1868 @@ -1,2 +1,3 @@
1869 line1
1870 +user_added
1871 line2
1872 "}
1873 .into(),
1874 predicted: false,
1875 in_open_source_repo: false,
1876 })],
1877 None,
1878 );
1879
1880 let predicted = indoc! {"
1881 line1
1882 line2
1883 "};
1884 let ratio =
1885 compute_prediction_reversal_ratio(&prompt_inputs, predicted, Path::new("src/test.rs"));
1886
1887 assert_eq!(
1888 ratio, 0.0,
1889 "Expected zero reversal when edit history is for different file"
1890 );
1891 }
1892
1893 #[test]
1894 fn test_compute_prediction_reversal_ratio_lenient_fallback() {
1895 let prompt_inputs = make_test_prompt_inputs(
1896 indoc! {"
1897 actual_line1
1898 user_added
1899 actual_line2
1900 "},
1901 vec![Arc::new(zeta_prompt::Event::BufferChange {
1902 path: Arc::from(Path::new("src/test.rs")),
1903 old_path: Arc::from(Path::new("src/test.rs")),
1904 diff: indoc! {"
1905 @@ -1,2 +1,3 @@
1906 wrong_context
1907 +user_added
1908 more_wrong
1909 "}
1910 .into(),
1911 predicted: false,
1912 in_open_source_repo: false,
1913 })],
1914 None,
1915 );
1916
1917 let predicted = indoc! {"
1918 actual_line1
1919 actual_line2
1920 "};
1921 let ratio =
1922 compute_prediction_reversal_ratio(&prompt_inputs, predicted, Path::new("src/test.rs"));
1923
1924 assert!(
1925 ratio >= 0.0 && ratio <= 1.0,
1926 "Ratio should be valid even with lenient fallback, got {}",
1927 ratio
1928 );
1929 }
1930
1931 #[test]
1932 fn test_excerpt_aware_reversal_error_recovery() {
1933 let diffs = vec![indoc! {"
1934 @@ -1,2 +1,3 @@
1935 nonexistent_context
1936 +added
1937 more_nonexistent
1938 "}];
1939 let excerpt_content = indoc! {"
1940 completely
1941 different
1942 content
1943 "};
1944 let predicted_content = indoc! {"
1945 completely
1946 modified
1947 content
1948 "};
1949
1950 let overlap =
1951 compute_excerpt_aware_reversal_overlap(&diffs, excerpt_content, 0, predicted_content);
1952
1953 assert!(
1954 overlap.ratio() >= 0.0 && overlap.ratio() <= 1.0,
1955 "Should handle failed diff application gracefully"
1956 );
1957 }
1958
1959 #[test]
1960 fn test_only_most_recent_edit_tracked() {
1961 let prompt_inputs = make_test_prompt_inputs(
1962 indoc! {"
1963 line1
1964 first_add
1965 second_add
1966 line2
1967 "},
1968 vec![
1969 Arc::new(zeta_prompt::Event::BufferChange {
1970 path: Arc::from(Path::new("src/test.rs")),
1971 old_path: Arc::from(Path::new("src/test.rs")),
1972 diff: indoc! {"
1973 @@ -1,2 +1,3 @@
1974 line1
1975 +first_add
1976 line2
1977 "}
1978 .into(),
1979 predicted: false,
1980 in_open_source_repo: false,
1981 }),
1982 Arc::new(zeta_prompt::Event::BufferChange {
1983 path: Arc::from(Path::new("src/test.rs")),
1984 old_path: Arc::from(Path::new("src/test.rs")),
1985 diff: indoc! {"
1986 @@ -2,2 +2,3 @@
1987 first_add
1988 +second_add
1989 line2
1990 "}
1991 .into(),
1992 predicted: false,
1993 in_open_source_repo: false,
1994 }),
1995 ],
1996 None,
1997 );
1998
1999 let predicted = indoc! {"
2000 line1
2001 first_add
2002 line2
2003 "};
2004 let ratio =
2005 compute_prediction_reversal_ratio(&prompt_inputs, predicted, Path::new("src/test.rs"));
2006
2007 assert!(
2008 ratio > 0.9,
2009 "Expected high reversal ratio when prediction exactly reverses the most recent edit, got {}",
2010 ratio
2011 );
2012 }
2013}