| @@ -1,16 +1,18 @@ |
| 1 | +use std::{collections::BTreeMap, hash::Hash, path::Path}; |
| 2 | + |
| 3 | +use gix_diff::blob::{self, Algorithm, InternedInput, diff_with_slider_heuristics}; |
| 1 | 4 | use gix_object::bstr::ByteSlice; |
| 2 | 5 | use pretty_assertions::StrComparison; |
| 3 | 6 | |
| 4 | 7 | #[test] |
| 5 | 8 | fn baseline() -> gix_testtools::Result { |
| 6 | | − use gix_diff::blob::{self, Algorithm, InternedInput, diff_with_slider_heuristics}; |
| 7 | | − |
| 8 | 9 | let worktree_path = crate::scripted_fixture_read_only("make_diff_for_sliders_repo.sh")?; |
| 9 | 10 | let asset_dir = worktree_path.join("assets"); |
| 10 | 11 | |
| 11 | 12 | let dir = std::fs::read_dir(&worktree_path)?; |
| 13 | + let should_print_extended_report = std::env::var_os("GIX_DIFF_SLIDER_REPORT").is_some(); |
| 12 | 14 | |
| 13 | | − let mut diffs = Vec::new(); |
| 15 | + let mut cases = Vec::new(); |
| 14 | 16 | |
| 15 | 17 | for entry in dir { |
| 16 | 18 | let entry = entry?; |
| @@ -28,44 +30,130 @@ fn baseline() -> gix_testtools::Result { · 11 unmodified lines |
| 28 | 30 | old_data.to_str().expect("BUG: we don't have non-ascii here"), |
| 29 | 31 | new_data.to_str().expect("BUG: we don't have non-ascii here"), |
| 30 | 32 | ); |
| 31 | | − let diff = diff_with_slider_heuristics( |
| 32 | | − match algorithm { |
| 33 | | − Algorithm::Myers => Algorithm::Myers, |
| 34 | | − Algorithm::Histogram => Algorithm::Histogram, |
| 35 | | − Algorithm::MyersMinimal => Algorithm::MyersMinimal, |
| 36 | | − }, |
| 37 | | − &input, |
| 38 | | − ); |
| 39 | 33 | |
| 40 | | − let actual = blob::UnifiedDiff::new( |
| 41 | | − &diff, |
| 42 | | − &input, |
| 43 | | − blob::unified_diff::ConsumeBinaryHunk::new(String::new(), "\n"), |
| 44 | | − blob::unified_diff::ContextSize::symmetrical(3), |
| 45 | | − ) |
| 46 | | − .consume()?; |
| 34 | + let gix_no_postprocess = { |
| 35 | + let diff = blob::Diff::compute(algorithm, &input); |
| 36 | + render_unidiff(&diff, &input)? |
| 37 | + }; |
| 38 | + |
| 39 | + let gix_postprocess_no_heuristic = { |
| 40 | + let mut diff = blob::Diff::compute(algorithm, &input); |
| 41 | + diff.postprocess_no_heuristic(&input); |
| 42 | + render_unidiff(&diff, &input)? |
| 43 | + }; |
| 44 | + |
| 45 | + let gix_postprocess_slider_heuristics = { |
| 46 | + let diff = diff_with_slider_heuristics(algorithm, &input); |
| 47 | + render_unidiff(&diff, &input)? |
| 48 | + }; |
| 47 | 49 | |
| 48 | 50 | let baseline_path = worktree_path.join(&file_name); |
| 49 | 51 | let baseline = std::fs::read(baseline_path)?; |
| 50 | 52 | let baseline = crate::blob::skip_header_and_fold_to_unidiff(&baseline); |
| 53 | + let git_no_indent_heuristic = if should_print_extended_report { |
| 54 | + read_no_indent_baseline(&worktree_path, &file_name)? |
| 55 | + } else { |
| 56 | + None |
| 57 | + }; |
| 51 | 58 | |
| 52 | | − let actual_matches_baseline = actual == baseline; |
| 53 | | − diffs.push((actual, baseline, actual_matches_baseline, file_name)); |
| 59 | + cases.push(Case { |
| 60 | + file_name, |
| 61 | + algorithm, |
| 62 | + git_postprocess_indent_heuristic: baseline, |
| 63 | + git_no_indent_heuristic, |
| 64 | + gix_no_postprocess, |
| 65 | + gix_postprocess_no_heuristic, |
| 66 | + gix_postprocess_slider_heuristics, |
| 67 | + }); |
| 54 | 68 | } |
| 55 | 69 | |
| 56 | | − if diffs.is_empty() { |
| 57 | | − eprintln!("Slider baseline isn't setup - look at ./gix-diff/tests/README.md for instructions"); |
| 70 | + if cases.is_empty() { |
| 71 | + eprintln!("Slider baseline isn't set up – see ./gix-diff/tests/README.md for instructions"); |
| 72 | + } |
| 73 | + |
| 74 | + if should_print_extended_report { |
| 75 | + print_extended_report(&cases); |
| 76 | + } else { |
| 77 | + assert_diffs(&cases); |
| 58 | 78 | } |
| 59 | 79 | |
| 60 | | − assert_diffs(&diffs); |
| 61 | 80 | Ok(()) |
| 62 | 81 | } |
| 63 | 82 | |
| 64 | | −fn assert_diffs(diffs: &[(String, String, bool, String)]) { |
| 65 | | − let total_diffs = diffs.len(); |
| 66 | | − let matching_diffs = diffs |
| 83 | +struct Case { |
| 84 | + file_name: String, |
| 85 | + algorithm: Algorithm, |
| 86 | + git_postprocess_indent_heuristic: String, |
| 87 | + git_no_indent_heuristic: Option<String>, |
| 88 | + gix_no_postprocess: String, |
| 89 | + gix_postprocess_no_heuristic: String, |
| 90 | + gix_postprocess_slider_heuristics: String, |
| 91 | +} |
| 92 | + |
| 93 | +#[derive(Debug, Clone, Copy, PartialEq, Eq, PartialOrd, Ord)] |
| 94 | +enum Classification { |
| 95 | + Exact, |
| 96 | + NoPostprocessMatchesGitDefault, |
| 97 | + PostprocessNoHeuristicMatchesGitDefault, |
| 98 | + PostprocessNoHeuristicMatchesGitNoIndent, |
| 99 | + LikelySliderOnlyMismatch, |
| 100 | + OtherMismatch, |
| 101 | +} |
| 102 | + |
| 103 | +impl Case { |
| 104 | + fn classify(&self) -> Classification { |
| 105 | + if self.gix_postprocess_slider_heuristics == self.git_postprocess_indent_heuristic { |
| 106 | + Classification::Exact |
| 107 | + } else if self.gix_no_postprocess == self.git_postprocess_indent_heuristic { |
| 108 | + Classification::NoPostprocessMatchesGitDefault |
| 109 | + } else if self.gix_postprocess_no_heuristic == self.git_postprocess_indent_heuristic { |
| 110 | + Classification::PostprocessNoHeuristicMatchesGitDefault |
| 111 | + } else if self |
| 112 | + .git_no_indent_heuristic |
| 113 | + .as_ref() |
| 114 | + .is_some_and(|git_no_indent_heuristic| &self.gix_postprocess_no_heuristic == git_no_indent_heuristic) |
| 115 | + { |
| 116 | + Classification::PostprocessNoHeuristicMatchesGitNoIndent |
| 117 | + } else if has_same_changed_line_sequence( |
| 118 | + &self.gix_postprocess_slider_heuristics, |
| 119 | + &self.git_postprocess_indent_heuristic, |
| 120 | + ) { |
| 121 | + Classification::LikelySliderOnlyMismatch |
| 122 | + } else { |
| 123 | + Classification::OtherMismatch |
| 124 | + } |
| 125 | + } |
| 126 | +} |
| 127 | + |
| 128 | +fn render_unidiff<T: AsRef<[u8]> + Hash + Eq>(diff: &blob::Diff, input: &InternedInput<T>) -> std::io::Result<String> { |
| 129 | + blob::UnifiedDiff::new( |
| 130 | + diff, |
| 131 | + input, |
| 132 | + blob::unified_diff::ConsumeBinaryHunk::new(String::new(), "\n"), |
| 133 | + blob::unified_diff::ContextSize::symmetrical(3), |
| 134 | + ) |
| 135 | + .consume() |
| 136 | +} |
| 137 | + |
| 138 | +fn read_no_indent_baseline(worktree_path: &Path, primary_file_name: &str) -> std::io::Result<Option<String>> { |
| 139 | + let Some(stem) = primary_file_name.strip_suffix(".baseline") else { |
| 140 | + return Ok(None); |
| 141 | + }; |
| 142 | + |
| 143 | + let path = worktree_path.join(format!("{stem}.no-indent.baseline")); |
| 144 | + let baseline = match std::fs::read(path) { |
| 145 | + Ok(baseline) => baseline, |
| 146 | + Err(err) if err.kind() == std::io::ErrorKind::NotFound => return Ok(None), |
| 147 | + Err(err) => return Err(err), |
| 148 | + }; |
| 149 | + Ok(Some(super::skip_header_and_fold_to_unidiff(&baseline))) |
| 150 | +} |
| 151 | + |
| 152 | +fn assert_diffs(cases: &[Case]) { |
| 153 | + let total_diffs = cases.len(); |
| 154 | + let matching_diffs = cases |
| 67 | 155 | .iter() |
| 68 | | − .filter(|(_, _, actual_matches_baseline, _)| *actual_matches_baseline) |
| 156 | + .filter(|case| case.gix_postprocess_slider_heuristics == case.git_postprocess_indent_heuristic) |
| 69 | 157 | .count(); |
| 70 | 158 | |
| 71 | 159 | assert_eq!( |
| @@ -76,20 +164,253 @@ fn assert_diffs(diffs: &[(String, String, bool, String)]) { · 4 unmodified lines |
| 76 | 164 | total_diffs, |
| 77 | 165 | ((matching_diffs as f32) / (total_diffs as f32) * 100.0), |
| 78 | 166 | { |
| 79 | | − let first_non_matching_diff = diffs |
| 167 | + let first_non_matching_diff = cases |
| 80 | 168 | .iter() |
| 81 | | − .find(|(_, _, actual_matches_baseline, _)| !actual_matches_baseline) |
| 169 | + .find(|case| case.gix_postprocess_slider_heuristics != case.git_postprocess_indent_heuristic) |
| 82 | 170 | .expect("at least one non-matching diff to be there"); |
| 83 | 171 | |
| 84 | 172 | format!( |
| 85 | 173 | "affected baseline: `{}`\n\n{}", |
| 86 | | − first_non_matching_diff.3, |
| 87 | | − StrComparison::new(&first_non_matching_diff.0, &first_non_matching_diff.1) |
| 174 | + first_non_matching_diff.file_name, |
| 175 | + StrComparison::new( |
| 176 | + &first_non_matching_diff.gix_postprocess_slider_heuristics, |
| 177 | + &first_non_matching_diff.git_postprocess_indent_heuristic |
| 178 | + ) |
| 88 | 179 | ) |
| 89 | 180 | } |
| 90 | 181 | ); |
| 91 | 182 | } |
| 92 | 183 | |
| 184 | +// This function intentionally only is capable to detect a subset of diffs that differ by slider |
| 185 | +// placement only. It can only detect sliders that have added/deleted lines in the exact same |
| 186 | +// order. I plan on adding more accurate classification in the future, though. |
| 187 | +fn has_same_changed_line_sequence(lhs: &str, rhs: &str) -> bool { |
| 188 | + fn changed_lines(diff: &str) -> Vec<&str> { |
| 189 | + diff.lines() |
| 190 | + .filter(|line| line.starts_with('+') || line.starts_with('-')) |
| 191 | + .collect() |
| 192 | + } |
| 193 | + |
| 194 | + changed_lines(lhs) == changed_lines(rhs) |
| 195 | +} |
| 196 | + |
| 197 | +fn slider_detail(gix: &str, git: &str) -> String { |
| 198 | + let gix_hunks = parse_hunks(gix); |
| 199 | + let git_hunks = parse_hunks(git); |
| 200 | + |
| 201 | + if gix_hunks.is_empty() || git_hunks.is_empty() { |
| 202 | + return format!( |
| 203 | + "diagnostic-error/unparsed-or-empty-hunks/gix-{}/git-{}", |
| 204 | + gix_hunks.len(), |
| 205 | + git_hunks.len() |
| 206 | + ); |
| 207 | + } |
| 208 | + |
| 209 | + if gix_hunks.len() != git_hunks.len() { |
| 210 | + return format!( |
| 211 | + "multi-hunk/different-hunk-count/gix-{}/git-{}", |
| 212 | + gix_hunks.len(), |
| 213 | + git_hunks.len() |
| 214 | + ); |
| 215 | + } |
| 216 | + |
| 217 | + // At this point, we know that `gix_hunks.len() == git_hunks.len()`. |
| 218 | + if gix_hunks.len() > 1 { |
| 219 | + let directions: Vec<_> = gix_hunks |
| 220 | + .iter() |
| 221 | + .zip(&git_hunks) |
| 222 | + .map(|(gix_hunk, git_hunk)| movement_direction(gix_hunk, git_hunk)) |
| 223 | + .collect(); |
| 224 | + let same_direction = directions |
| 225 | + .first() |
| 226 | + .is_some_and(|first| directions.iter().all(|direction| direction == first)); |
| 227 | + let direction = if same_direction { |
| 228 | + "same-direction" |
| 229 | + } else { |
| 230 | + "mixed-direction" |
| 231 | + }; |
| 232 | + |
| 233 | + return format!("multi-hunk/same-count/{direction}"); |
| 234 | + } |
| 235 | + |
| 236 | + let gix_hunk = &gix_hunks[0]; |
| 237 | + let git_hunk = &git_hunks[0]; |
| 238 | + |
| 239 | + let gix_is_empty = gix_hunk.removed == 0 && gix_hunk.added == 0; |
| 240 | + let git_is_empty = git_hunk.removed == 0 && git_hunk.added == 0; |
| 241 | + |
| 242 | + match (gix_is_empty, git_is_empty) { |
| 243 | + (true, true) => return "diagnostic-error/empty-hunks/both".to_owned(), |
| 244 | + (true, false) => return "diagnostic-error/empty-hunks/gix".to_owned(), |
| 245 | + (false, true) => return "diagnostic-error/empty-hunks/git".to_owned(), |
| 246 | + (false, false) => {} |
| 247 | + } |
| 248 | + |
| 249 | + let kind = match (gix_hunk.removed > 0, gix_hunk.added > 0) { |
| 250 | + (false, true) => "pure-insertion", |
| 251 | + (true, false) => "pure-deletion", |
| 252 | + (true, true) => "modification", |
| 253 | + (false, false) => unreachable!("BUG: we verified that both `added` and `removed` are non-zero"), |
| 254 | + }; |
| 255 | + let direction = movement_direction(gix_hunk, git_hunk); |
| 256 | + let distance = movement_distance_bucket(gix_hunk, git_hunk); |
| 257 | + |
| 258 | + format!("single-hunk/{kind}/{direction}/{distance}") |
| 259 | +} |
| 260 | + |
| 261 | +struct ParsedHunk { |
| 262 | + before_start: u32, |
| 263 | + after_start: u32, |
| 264 | + removed: usize, |
| 265 | + added: usize, |
| 266 | +} |
| 267 | + |
| 268 | +fn parse_hunks(diff: &str) -> Vec<ParsedHunk> { |
| 269 | + let mut hunks = Vec::<ParsedHunk>::new(); |
| 270 | + |
| 271 | + for line in diff.lines() { |
| 272 | + if line.starts_with("@@ ") { |
| 273 | + if let Some((before_start, after_start)) = parse_hunk_header(line) { |
| 274 | + hunks.push(ParsedHunk { |
| 275 | + before_start, |
| 276 | + after_start, |
| 277 | + removed: 0, |
| 278 | + added: 0, |
| 279 | + }); |
| 280 | + } |
| 281 | + } else if let Some(hunk) = hunks.last_mut() { |
| 282 | + if line.starts_with('-') { |
| 283 | + hunk.removed += 1; |
| 284 | + } else if line.starts_with('+') { |
| 285 | + hunk.added += 1; |
| 286 | + } |
| 287 | + } |
| 288 | + } |
| 289 | + |
| 290 | + hunks |
| 291 | +} |
| 292 | + |
| 293 | +fn parse_hunk_header(line: &str) -> Option<(u32, u32)> { |
| 294 | + // Supported forms include both `@@ -10,3 +10,4 @@` and `@@ -10 +10 @@`. |
| 295 | + let line = line.strip_prefix("@@ ")?; |
| 296 | + let (header, _section) = line.split_once(" @@")?; |
| 297 | + |
| 298 | + let mut parts = header.split_ascii_whitespace(); |
| 299 | + let before = parts.next()?.strip_prefix('-')?; |
| 300 | + let after = parts.next()?.strip_prefix('+')?; |
| 301 | + |
| 302 | + Some((parse_range_start(before)?, parse_range_start(after)?)) |
| 303 | +} |
| 304 | + |
| 305 | +fn parse_range_start(range: &str) -> Option<u32> { |
| 306 | + let start = range.split_once(',').map_or(range, |(start, _len)| start); |
| 307 | + |
| 308 | + start.parse().ok() |
| 309 | +} |
| 310 | + |
| 311 | +fn movement_direction(gix_hunk: &ParsedHunk, git_hunk: &ParsedHunk) -> &'static str { |
| 312 | + match gix_hunk.before_start.cmp(&git_hunk.before_start) { |
| 313 | + std::cmp::Ordering::Less => "gix-before-start-earlier", |
| 314 | + std::cmp::Ordering::Greater => "gix-before-start-later", |
| 315 | + std::cmp::Ordering::Equal => match gix_hunk.after_start.cmp(&git_hunk.after_start) { |
| 316 | + std::cmp::Ordering::Less => "same-before-start/gix-after-start-earlier", |
| 317 | + std::cmp::Ordering::Greater => "same-before-start/gix-after-start-later", |
| 318 | + std::cmp::Ordering::Equal => "same-before-start", |
| 319 | + }, |
| 320 | + } |
| 321 | +} |
| 322 | + |
| 323 | +fn movement_distance_bucket(gix_hunk: &ParsedHunk, git_hunk: &ParsedHunk) -> &'static str { |
| 324 | + let before_delta = gix_hunk.before_start.abs_diff(git_hunk.before_start); |
| 325 | + let after_delta = gix_hunk.after_start.abs_diff(git_hunk.after_start); |
| 326 | + |
| 327 | + match before_delta.max(after_delta) { |
| 328 | + 0 => "0", |
| 329 | + 1 => "1", |
| 330 | + 2..=5 => "2-5", |
| 331 | + 6..=20 => "6-20", |
| 332 | + _ => ">20", |
| 333 | + } |
| 334 | +} |
| 335 | + |
| 336 | +fn print_extended_report(cases: &[Case]) { |
| 337 | + let mut counts = BTreeMap::<Classification, usize>::new(); |
| 338 | + let mut likely_slider_only_counts = BTreeMap::<String, usize>::new(); |
| 339 | + |
| 340 | + for case in cases { |
| 341 | + let classification = case.classify(); |
| 342 | + *counts.entry(classification).or_default() += 1; |
| 343 | + |
| 344 | + if classification == Classification::LikelySliderOnlyMismatch { |
| 345 | + *likely_slider_only_counts |
| 346 | + .entry(slider_detail( |
| 347 | + &case.gix_postprocess_slider_heuristics, |
| 348 | + &case.git_postprocess_indent_heuristic, |
| 349 | + )) |
| 350 | + .or_default() += 1; |
| 351 | + } |
| 352 | + } |
| 353 | + |
| 354 | + eprintln!("slider report"); |
| 355 | + eprintln!(); |
| 356 | + |
| 357 | + if cases.is_empty() { |
| 358 | + eprintln!("no cases to report"); |
| 359 | + |
| 360 | + return; |
| 361 | + } |
| 362 | + |
| 363 | + eprintln!("total cases: {}", cases.len()); |
| 364 | + eprintln!( |
| 365 | + "git no-indent baselines: {}", |
| 366 | + cases |
| 367 | + .iter() |
| 368 | + .filter(|case| case.git_no_indent_heuristic.is_some()) |
| 369 | + .count() |
| 370 | + ); |
| 371 | + eprintln!(); |
| 372 | + |
| 373 | + for (classification, count) in counts { |
| 374 | + let percentage = count as f32 / cases.len() as f32 * 100.0; |
| 375 | + |
| 376 | + eprintln!("{classification:?}: {count} [{percentage:.2}%]"); |
| 377 | + } |
| 378 | + |
| 379 | + eprintln!(); |
| 380 | + eprintln!("likely slider-only details"); |
| 381 | + eprintln!(); |
| 382 | + |
| 383 | + let likely_slider_only_total: usize = likely_slider_only_counts.values().sum(); |
| 384 | + |
| 385 | + if likely_slider_only_total == 0 { |
| 386 | + eprintln!("no likely slider-only mismatches"); |
| 387 | + } else { |
| 388 | + let mut details: Vec<_> = likely_slider_only_counts.into_iter().collect(); |
| 389 | + details.sort_by(|(lhs_detail, lhs_count), (rhs_detail, rhs_count)| { |
| 390 | + rhs_count.cmp(lhs_count).then_with(|| lhs_detail.cmp(rhs_detail)) |
| 391 | + }); |
| 392 | + |
| 393 | + for (detail, count) in details { |
| 394 | + let slider_only_percentage = count as f32 / likely_slider_only_total as f32 * 100.0; |
| 395 | + let total_percentage = count as f32 / cases.len() as f32 * 100.0; |
| 396 | + |
| 397 | + eprintln!("{detail}: {count} [{slider_only_percentage:.2}% slider-only, {total_percentage:.2}% total]"); |
| 398 | + } |
| 399 | + } |
| 400 | + |
| 401 | + eprintln!(); |
| 402 | + eprintln!("first non-matching cases"); |
| 403 | + eprintln!(); |
| 404 | + |
| 405 | + for case in cases |
| 406 | + .iter() |
| 407 | + .filter(|case| case.gix_postprocess_slider_heuristics != case.git_postprocess_indent_heuristic) |
| 408 | + .take(20) |
| 409 | + { |
| 410 | + eprintln!("{} {:?} {:?}", case.file_name, case.algorithm, case.classify()); |
| 411 | + } |
| 412 | +} |
| 413 | + |
| 93 | 414 | mod baseline { |
| 94 | 415 | use gix_diff::blob::Algorithm; |
| 95 | 416 | use std::ffi::OsStr; |
| @@ -102,17 +423,18 @@ mod baseline { · 6 unmodified lines |
| 102 | 423 | pub new_data: Vec<u8>, |
| 103 | 424 | } |
| 104 | 425 | |
| 105 | | − /// Returns `None` if the file isn't a baseline entry. |
| 426 | + /// Returns `None` if the file isn't a primary baseline entry. |
| 106 | 427 | pub fn parse_dir_entry(asset_dir: &Path, file_name: &OsStr) -> std::io::Result<Option<DirEntry>> { |
| 107 | 428 | let file_name = file_name.to_str().expect("ascii filename").to_owned(); |
| 108 | 429 | |
| 109 | | − if !file_name.ends_with(".baseline") { |
| 430 | + let Some(stem) = file_name.strip_suffix(".baseline") else { |
| 110 | 431 | return Ok(None); |
| 111 | | − } |
| 432 | + }; |
| 112 | 433 | |
| 113 | | − let parts: Vec<_> = file_name.split('.').collect(); |
| 114 | | − let [name, algorithm, ..] = parts[..] else { |
| 115 | | − unreachable!("BUG: Need file named '<name>.<algorithm>'") |
| 434 | + let parts: Vec<_> = stem.split('.').collect(); |
| 435 | + let [name, algorithm] = parts[..] else { |
| 436 | + // Additional baselines like `<name>.<algorithm>.no-indent.baseline` are consumed separately. |
| 437 | + return Ok(None); |
| 116 | 438 | }; |
| 117 | 439 | let algorithm = match algorithm { |
| 118 | 440 | "myers" => Algorithm::Myers, |