|
| 1 | +# RFC: Vite+ Cache Fingerprint Ignore Patterns |
| 2 | + |
| 3 | +## Summary |
| 4 | + |
| 5 | +Add support for glob-based ignore patterns to the cache fingerprint calculation, allowing tasks to exclude specific files/directories from triggering cache invalidation while still including important files within ignored directories. |
| 6 | + |
| 7 | +## Motivation |
| 8 | + |
| 9 | +Current cache fingerprint behavior tracks all files accessed during task execution. This causes unnecessary cache invalidation in scenarios like: |
| 10 | + |
| 11 | +1. **Package installation tasks**: The `node_modules` directory changes frequently, but only `package.json` files within it are relevant for cache validation |
| 12 | +2. **Build output directories**: Generated files in `dist/` or `.next/` that should not invalidate the cache |
| 13 | +3. **Large dependency directories**: When only specific files within large directories matter for reproducibility |
| 14 | + |
| 15 | +### Example Use Case |
| 16 | + |
| 17 | +For an `install` task that runs `pnpm install`: |
| 18 | + |
| 19 | +- Changes to `node_modules/**/*/index.js` should NOT invalidate the cache |
| 20 | +- Changes to `node_modules/**/*/package.json` SHOULD invalidate the cache |
| 21 | +- This allows cache hits when dependencies remain the same, even if their internal implementation files have different timestamps or minor variations |
| 22 | + |
| 23 | +## Proposed Solution |
| 24 | + |
| 25 | +### Configuration Schema |
| 26 | + |
| 27 | +Extend `TaskConfig` in `vite-task.json` to support a new optional field `fingerprintIgnores`: |
| 28 | + |
| 29 | +```json |
| 30 | +{ |
| 31 | + "tasks": { |
| 32 | + "my-task": { |
| 33 | + "command": "echo bar", |
| 34 | + "cacheable": true, |
| 35 | + "fingerprintIgnores": [ |
| 36 | + "node_modules/**/*", |
| 37 | + "!node_modules/**/*/package.json" |
| 38 | + ] |
| 39 | + } |
| 40 | + } |
| 41 | +} |
| 42 | +``` |
| 43 | + |
| 44 | +### Ignore Pattern Syntax |
| 45 | + |
| 46 | +The ignore patterns follow standard glob syntax with gitignore-style semantics: |
| 47 | + |
| 48 | +1. **Basic patterns**: |
| 49 | + - `node_modules/**/*` - ignore all files under node_modules |
| 50 | + - `dist/` - ignore the dist directory |
| 51 | + - `*.log` - ignore all log files |
| 52 | + |
| 53 | +2. **Negation patterns** (prefixed with `!`): |
| 54 | + - `!node_modules/**/*/package.json` - include package.json files even though node_modules is ignored |
| 55 | + - `!important.log` - include important.log even though *.log is ignored |
| 56 | + |
| 57 | +3. **Pattern evaluation order**: |
| 58 | + - Patterns are evaluated in order |
| 59 | + - Later patterns override earlier ones |
| 60 | + - Negation patterns can "un-ignore" files matched by earlier patterns |
| 61 | + - Last match wins semantics |
| 62 | + |
| 63 | +### Implementation Details |
| 64 | + |
| 65 | +#### 1. Configuration Schema Changes |
| 66 | + |
| 67 | +**File**: `crates/vite_task/src/config/mod.rs` |
| 68 | + |
| 69 | +```rust |
| 70 | +pub struct TaskConfig { |
| 71 | + // ... |
| 72 | + |
| 73 | + // New field |
| 74 | + #[serde(default)] |
| 75 | + pub(crate) fingerprint_ignores: Option<Vec<Str>>, |
| 76 | +} |
| 77 | +``` |
| 78 | + |
| 79 | +#### 2. Fingerprint Validation Changes |
| 80 | + |
| 81 | +**File**: `crates/vite_task/src/fingerprint.rs` |
| 82 | + |
| 83 | +Modify `PostRunFingerprint::create()` to filter paths based on ignore patterns: |
| 84 | + |
| 85 | +```rust |
| 86 | +impl PostRunFingerprint { |
| 87 | + pub fn create( |
| 88 | + executed_task: &ExecutedTask, |
| 89 | + fs: &impl FileSystem, |
| 90 | + base_dir: &AbsolutePath, |
| 91 | + fingerprint_ignores: Option<&[Str]>, // New parameter |
| 92 | + ) -> Result<Self, Error> { |
| 93 | + let ignore_matcher = fingerprint_ignores |
| 94 | + .filter(|patterns| !patterns.is_empty()) |
| 95 | + .map(GlobPatternSet::new) |
| 96 | + .transpose()?; |
| 97 | + |
| 98 | + let inputs = executed_task |
| 99 | + .path_reads |
| 100 | + .par_iter() |
| 101 | + .filter(|(path, _)| { |
| 102 | + if let Some(ref matcher) = ignore_matcher { |
| 103 | + !matcher.is_match(path) |
| 104 | + } else { |
| 105 | + true |
| 106 | + } |
| 107 | + }) |
| 108 | + .flat_map(|(path, path_read)| { |
| 109 | + Some((|| { |
| 110 | + let path_fingerprint = |
| 111 | + fs.fingerprint_path(&base_dir.join(path).into(), *path_read)?; |
| 112 | + Ok((path.clone(), path_fingerprint)) |
| 113 | + })()) |
| 114 | + }) |
| 115 | + .collect::<Result<HashMap<RelativePathBuf, PathFingerprint>, Error>>()?; |
| 116 | + Ok(Self { inputs }) |
| 117 | + } |
| 118 | +} |
| 119 | +``` |
| 120 | + |
| 121 | +#### 3. Cache Update Integration |
| 122 | + |
| 123 | +**File**: `crates/vite_task/src/cache.rs` |
| 124 | + |
| 125 | +Update `CommandCacheValue::create()` to pass ignore patterns: |
| 126 | + |
| 127 | +```rust |
| 128 | +impl CommandCacheValue { |
| 129 | + pub fn create( |
| 130 | + executed_task: ExecutedTask, |
| 131 | + fs: &impl FileSystem, |
| 132 | + base_dir: &AbsolutePath, |
| 133 | + fingerprint_ignores: Option<&[Str]>, // New parameter |
| 134 | + ) -> Result<Self, Error> { |
| 135 | + let post_run_fingerprint = PostRunFingerprint::create( |
| 136 | + &executed_task, |
| 137 | + fs, |
| 138 | + base_dir, |
| 139 | + fingerprint_ignores, |
| 140 | + )?; |
| 141 | + Ok(Self { |
| 142 | + post_run_fingerprint, |
| 143 | + std_outputs: executed_task.std_outputs, |
| 144 | + duration: executed_task.duration, |
| 145 | + }) |
| 146 | + } |
| 147 | +} |
| 148 | +``` |
| 149 | + |
| 150 | +### Performance Considerations |
| 151 | + |
| 152 | +1. **Pattern compilation**: Glob patterns are compiled once when loading the task configuration |
| 153 | +2. **Filtering overhead**: Path filtering happens during fingerprint creation (only when caching) |
| 154 | +3. **Memory impact**: Minimal - only stores compiled glob patterns per task |
| 155 | +4. **Parallel processing**: Existing parallel iteration over paths is preserved |
| 156 | + |
| 157 | +### Edge Cases |
| 158 | + |
| 159 | +1. **Empty ignore list**: No filtering applied (backward compatible) |
| 160 | +2. **Conflicting patterns**: Later patterns take precedence |
| 161 | +3. **Invalid glob syntax**: Return error during workspace loading |
| 162 | +4. **Absolute paths in patterns**: Treated as relative to package directory |
| 163 | +5. **Directory vs file patterns**: Both supported via glob syntax |
| 164 | + |
| 165 | +## Alternative Designs Considered |
| 166 | + |
| 167 | +### Alternative 1: `inputs` field extension |
| 168 | + |
| 169 | +Extend the existing `inputs` field to support ignore patterns: |
| 170 | + |
| 171 | +```json |
| 172 | +{ |
| 173 | + "inputs": { |
| 174 | + "include": ["src/**/*"], |
| 175 | + "exclude": ["src/**/*.test.js"] |
| 176 | + } |
| 177 | +} |
| 178 | +``` |
| 179 | + |
| 180 | +**Rejected because**: |
| 181 | + |
| 182 | +- The `inputs` field currently uses a different mechanism (pre-execution declaration) |
| 183 | +- This feature is about post-execution fingerprint filtering |
| 184 | +- Mixing the two concepts would be confusing |
| 185 | + |
| 186 | +### Alternative 2: Separate `fingerprintExcludes` field |
| 187 | + |
| 188 | +Only support exclude patterns (no negation): |
| 189 | + |
| 190 | +```json |
| 191 | +{ |
| 192 | + "fingerprintExcludes": ["node_modules/**/*"] |
| 193 | +} |
| 194 | +``` |
| 195 | + |
| 196 | +**Rejected because**: |
| 197 | + |
| 198 | +- Cannot express "ignore everything except X" patterns |
| 199 | +- Less flexible for complex scenarios |
| 200 | +- Gitignore-style syntax is more familiar to developers |
| 201 | + |
| 202 | +### Alternative 3: Include/Exclude separate fields |
| 203 | + |
| 204 | +```json |
| 205 | +{ |
| 206 | + "fingerprintExcludes": ["node_modules/**/*"], |
| 207 | + "fingerprintIncludes": ["node_modules/**/*/package.json"] |
| 208 | +} |
| 209 | +``` |
| 210 | + |
| 211 | +**Rejected because**: |
| 212 | + |
| 213 | +- More verbose |
| 214 | +- Less clear precedence rules |
| 215 | +- Gitignore-style is a proven pattern |
| 216 | + |
| 217 | +## Migration Path |
| 218 | + |
| 219 | +### Backward Compatibility |
| 220 | + |
| 221 | +This feature is fully backward compatible: |
| 222 | + |
| 223 | +- Existing task configurations work unchanged |
| 224 | +- Default value for `fingerprintIgnores` is `None` (when omitted) |
| 225 | +- No behavior changes when field is absent or `null` |
| 226 | +- Empty array `[]` is treated the same as `None` (no filtering) |
| 227 | + |
| 228 | +## Testing Strategy |
| 229 | + |
| 230 | +### Unit Tests |
| 231 | + |
| 232 | +1. **Pattern matching**: |
| 233 | + - Test glob pattern compilation |
| 234 | + - Test negation pattern precedence |
| 235 | + - Test edge cases (empty patterns, invalid syntax) |
| 236 | + |
| 237 | +2. **Fingerprint filtering**: |
| 238 | + - Test path filtering with various patterns |
| 239 | + - Test no filtering when patterns are empty |
| 240 | + - Test complex pattern combinations |
| 241 | + |
| 242 | +3. **Cache behavior**: |
| 243 | + - Test cache hit when ignored files change |
| 244 | + - Test cache miss when non-ignored files change |
| 245 | + - Test negation patterns work correctly |
| 246 | + |
| 247 | +### Integration Tests |
| 248 | + |
| 249 | +Create fixtures with realistic scenarios: |
| 250 | + |
| 251 | +``` |
| 252 | +fixtures/fingerprint-ignore-test/ |
| 253 | + package.json |
| 254 | + vite-task.json # with fingerprintIgnores config |
| 255 | + node_modules/ |
| 256 | + pkg-a/ |
| 257 | + package.json |
| 258 | + index.js |
| 259 | + pkg-b/ |
| 260 | + package.json |
| 261 | + index.js |
| 262 | +``` |
| 263 | + |
| 264 | +Test cases: |
| 265 | + |
| 266 | +1. Cache hits when `index.js` files change |
| 267 | +2. Cache misses when `package.json` files change |
| 268 | +3. Negation patterns correctly include files |
| 269 | + |
| 270 | +## Documentation Requirements |
| 271 | + |
| 272 | +### User Documentation |
| 273 | + |
| 274 | +Add to task configuration docs: |
| 275 | + |
| 276 | +````markdown |
| 277 | +### fingerprintIgnores |
| 278 | + |
| 279 | +Type: `string[]` |
| 280 | +Default: `[]` |
| 281 | + |
| 282 | +Glob patterns to exclude files from cache fingerprint calculation. |
| 283 | +Patterns starting with `!` are negation patterns that override earlier excludes. |
| 284 | + |
| 285 | +Example: |
| 286 | + |
| 287 | +```json |
| 288 | +{ |
| 289 | + "tasks": { |
| 290 | + "install": { |
| 291 | + "command": "pnpm install", |
| 292 | + "cacheable": true, |
| 293 | + "fingerprintIgnores": [ |
| 294 | + "node_modules/**/*", |
| 295 | + "!node_modules/**/*/package.json" |
| 296 | + ] |
| 297 | + } |
| 298 | + } |
| 299 | +} |
| 300 | +``` |
| 301 | +```` |
| 302 | + |
| 303 | +This configuration ignores all files in `node_modules` except `package.json` |
| 304 | +files, which are still tracked for cache validation. |
| 305 | + |
| 306 | +```` |
| 307 | +### Examples Documentation |
| 308 | +
|
| 309 | +Add common patterns: |
| 310 | +
|
| 311 | +1. **Package installation**: |
| 312 | + ```json |
| 313 | + "fingerprintIgnores": [ |
| 314 | + "node_modules/**/*", |
| 315 | + "!node_modules/**/*/package.json", |
| 316 | + "!node_modules/.pnpm/lock.yaml" |
| 317 | + ] |
| 318 | +```` |
| 319 | + |
| 320 | +2. **Build outputs**: |
| 321 | + ```json |
| 322 | + "fingerprintIgnores": [ |
| 323 | + "dist/**/*", |
| 324 | + ".next/**/*", |
| 325 | + "build/**/*" |
| 326 | + ] |
| 327 | + ``` |
| 328 | + |
| 329 | +3. **Temporary files**: |
| 330 | + ```json |
| 331 | + "fingerprintIgnores": [ |
| 332 | + "**/*.log", |
| 333 | + "**/.DS_Store", |
| 334 | + "**/tmp/**" |
| 335 | + ] |
| 336 | + ``` |
| 337 | + |
| 338 | +## Conclusion |
| 339 | + |
| 340 | +This RFC proposes adding glob-based ignore patterns to cache fingerprint calculation. The feature: |
| 341 | + |
| 342 | +- Solves real caching problems (especially for install tasks) |
| 343 | +- Uses familiar gitignore-style syntax |
| 344 | +- Is fully backward compatible |
| 345 | +- Has minimal performance impact |
| 346 | +- Provides clear migration and documentation path |
| 347 | + |
| 348 | +The implementation is straightforward, leveraging the proven `vite_glob` crate, and integrates cleanly with existing fingerprint and cache systems. |
0 commit comments