1 | use std::cmp; |
2 | use std::ffi::OsStr; |
3 | use std::fmt; |
4 | use std::fs::{self, FileType, Metadata}; |
5 | use std::io; |
6 | use std::path::{Path, PathBuf}; |
7 | use std::sync::atomic::{AtomicBool, AtomicUsize, Ordering}; |
8 | use std::sync::{Arc, Mutex}; |
9 | use std::thread; |
10 | use std::time::Duration; |
11 | use std::vec; |
12 | |
13 | use same_file::Handle; |
14 | use walkdir::{self, WalkDir}; |
15 | |
16 | use crate::dir::{Ignore, IgnoreBuilder}; |
17 | use crate::gitignore::GitignoreBuilder; |
18 | use crate::overrides::Override; |
19 | use crate::types::Types; |
20 | use crate::{Error, PartialErrorBuilder}; |
21 | |
22 | /// A directory entry with a possible error attached. |
23 | /// |
24 | /// The error typically refers to a problem parsing ignore files in a |
25 | /// particular directory. |
26 | #[derive (Clone, Debug)] |
27 | pub struct DirEntry { |
28 | dent: DirEntryInner, |
29 | err: Option<Error>, |
30 | } |
31 | |
32 | impl DirEntry { |
33 | /// The full path that this entry represents. |
34 | pub fn path(&self) -> &Path { |
35 | self.dent.path() |
36 | } |
37 | |
38 | /// The full path that this entry represents. |
39 | /// Analogous to [`path`], but moves ownership of the path. |
40 | /// |
41 | /// [`path`]: struct.DirEntry.html#method.path |
42 | pub fn into_path(self) -> PathBuf { |
43 | self.dent.into_path() |
44 | } |
45 | |
46 | /// Whether this entry corresponds to a symbolic link or not. |
47 | pub fn path_is_symlink(&self) -> bool { |
48 | self.dent.path_is_symlink() |
49 | } |
50 | |
51 | /// Returns true if and only if this entry corresponds to stdin. |
52 | /// |
53 | /// i.e., The entry has depth 0 and its file name is `-`. |
54 | pub fn is_stdin(&self) -> bool { |
55 | self.dent.is_stdin() |
56 | } |
57 | |
58 | /// Return the metadata for the file that this entry points to. |
59 | pub fn metadata(&self) -> Result<Metadata, Error> { |
60 | self.dent.metadata() |
61 | } |
62 | |
63 | /// Return the file type for the file that this entry points to. |
64 | /// |
65 | /// This entry doesn't have a file type if it corresponds to stdin. |
66 | pub fn file_type(&self) -> Option<FileType> { |
67 | self.dent.file_type() |
68 | } |
69 | |
70 | /// Return the file name of this entry. |
71 | /// |
72 | /// If this entry has no file name (e.g., `/`), then the full path is |
73 | /// returned. |
74 | pub fn file_name(&self) -> &OsStr { |
75 | self.dent.file_name() |
76 | } |
77 | |
78 | /// Returns the depth at which this entry was created relative to the root. |
79 | pub fn depth(&self) -> usize { |
80 | self.dent.depth() |
81 | } |
82 | |
83 | /// Returns the underlying inode number if one exists. |
84 | /// |
85 | /// If this entry doesn't have an inode number, then `None` is returned. |
86 | #[cfg (unix)] |
87 | pub fn ino(&self) -> Option<u64> { |
88 | self.dent.ino() |
89 | } |
90 | |
91 | /// Returns an error, if one exists, associated with processing this entry. |
92 | /// |
93 | /// An example of an error is one that occurred while parsing an ignore |
94 | /// file. Errors related to traversing a directory tree itself are reported |
95 | /// as part of yielding the directory entry, and not with this method. |
96 | pub fn error(&self) -> Option<&Error> { |
97 | self.err.as_ref() |
98 | } |
99 | |
100 | /// Returns true if and only if this entry points to a directory. |
101 | pub(crate) fn is_dir(&self) -> bool { |
102 | self.dent.is_dir() |
103 | } |
104 | |
105 | fn new_stdin() -> DirEntry { |
106 | DirEntry { dent: DirEntryInner::Stdin, err: None } |
107 | } |
108 | |
109 | fn new_walkdir(dent: walkdir::DirEntry, err: Option<Error>) -> DirEntry { |
110 | DirEntry { dent: DirEntryInner::Walkdir(dent), err: err } |
111 | } |
112 | |
113 | fn new_raw(dent: DirEntryRaw, err: Option<Error>) -> DirEntry { |
114 | DirEntry { dent: DirEntryInner::Raw(dent), err: err } |
115 | } |
116 | } |
117 | |
118 | /// DirEntryInner is the implementation of DirEntry. |
119 | /// |
120 | /// It specifically represents three distinct sources of directory entries: |
121 | /// |
122 | /// 1. From the walkdir crate. |
123 | /// 2. Special entries that represent things like stdin. |
124 | /// 3. From a path. |
125 | /// |
126 | /// Specifically, (3) has to essentially re-create the DirEntry implementation |
127 | /// from WalkDir. |
128 | #[derive (Clone, Debug)] |
129 | enum DirEntryInner { |
130 | Stdin, |
131 | Walkdir(walkdir::DirEntry), |
132 | Raw(DirEntryRaw), |
133 | } |
134 | |
135 | impl DirEntryInner { |
136 | fn path(&self) -> &Path { |
137 | use self::DirEntryInner::*; |
138 | match *self { |
139 | Stdin => Path::new("<stdin>" ), |
140 | Walkdir(ref x) => x.path(), |
141 | Raw(ref x) => x.path(), |
142 | } |
143 | } |
144 | |
145 | fn into_path(self) -> PathBuf { |
146 | use self::DirEntryInner::*; |
147 | match self { |
148 | Stdin => PathBuf::from("<stdin>" ), |
149 | Walkdir(x) => x.into_path(), |
150 | Raw(x) => x.into_path(), |
151 | } |
152 | } |
153 | |
154 | fn path_is_symlink(&self) -> bool { |
155 | use self::DirEntryInner::*; |
156 | match *self { |
157 | Stdin => false, |
158 | Walkdir(ref x) => x.path_is_symlink(), |
159 | Raw(ref x) => x.path_is_symlink(), |
160 | } |
161 | } |
162 | |
163 | fn is_stdin(&self) -> bool { |
164 | match *self { |
165 | DirEntryInner::Stdin => true, |
166 | _ => false, |
167 | } |
168 | } |
169 | |
170 | fn metadata(&self) -> Result<Metadata, Error> { |
171 | use self::DirEntryInner::*; |
172 | match *self { |
173 | Stdin => { |
174 | let err = Error::Io(io::Error::new( |
175 | io::ErrorKind::Other, |
176 | "<stdin> has no metadata" , |
177 | )); |
178 | Err(err.with_path("<stdin>" )) |
179 | } |
180 | Walkdir(ref x) => x.metadata().map_err(|err| { |
181 | Error::Io(io::Error::from(err)).with_path(x.path()) |
182 | }), |
183 | Raw(ref x) => x.metadata(), |
184 | } |
185 | } |
186 | |
187 | fn file_type(&self) -> Option<FileType> { |
188 | use self::DirEntryInner::*; |
189 | match *self { |
190 | Stdin => None, |
191 | Walkdir(ref x) => Some(x.file_type()), |
192 | Raw(ref x) => Some(x.file_type()), |
193 | } |
194 | } |
195 | |
196 | fn file_name(&self) -> &OsStr { |
197 | use self::DirEntryInner::*; |
198 | match *self { |
199 | Stdin => OsStr::new("<stdin>" ), |
200 | Walkdir(ref x) => x.file_name(), |
201 | Raw(ref x) => x.file_name(), |
202 | } |
203 | } |
204 | |
205 | fn depth(&self) -> usize { |
206 | use self::DirEntryInner::*; |
207 | match *self { |
208 | Stdin => 0, |
209 | Walkdir(ref x) => x.depth(), |
210 | Raw(ref x) => x.depth(), |
211 | } |
212 | } |
213 | |
214 | #[cfg (unix)] |
215 | fn ino(&self) -> Option<u64> { |
216 | use self::DirEntryInner::*; |
217 | use walkdir::DirEntryExt; |
218 | match *self { |
219 | Stdin => None, |
220 | Walkdir(ref x) => Some(x.ino()), |
221 | Raw(ref x) => Some(x.ino()), |
222 | } |
223 | } |
224 | |
225 | /// Returns true if and only if this entry points to a directory. |
226 | fn is_dir(&self) -> bool { |
227 | self.file_type().map(|ft| ft.is_dir()).unwrap_or(false) |
228 | } |
229 | } |
230 | |
231 | /// DirEntryRaw is essentially copied from the walkdir crate so that we can |
232 | /// build `DirEntry`s from whole cloth in the parallel iterator. |
233 | #[derive (Clone)] |
234 | struct DirEntryRaw { |
235 | /// The path as reported by the `fs::ReadDir` iterator (even if it's a |
236 | /// symbolic link). |
237 | path: PathBuf, |
238 | /// The file type. Necessary for recursive iteration, so store it. |
239 | ty: FileType, |
240 | /// Is set when this entry was created from a symbolic link and the user |
241 | /// expects the iterator to follow symbolic links. |
242 | follow_link: bool, |
243 | /// The depth at which this entry was generated relative to the root. |
244 | depth: usize, |
245 | /// The underlying inode number (Unix only). |
246 | #[cfg (unix)] |
247 | ino: u64, |
248 | /// The underlying metadata (Windows only). We store this on Windows |
249 | /// because this comes for free while reading a directory. |
250 | #[cfg (windows)] |
251 | metadata: fs::Metadata, |
252 | } |
253 | |
254 | impl fmt::Debug for DirEntryRaw { |
255 | fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result { |
256 | // Leaving out FileType because it doesn't have a debug impl |
257 | // in Rust 1.9. We could add it if we really wanted to by manually |
258 | // querying each possibly file type. Meh. ---AG |
259 | f&mut DebugStruct<'_, '_>.debug_struct("DirEntryRaw" ) |
260 | .field("path" , &self.path) |
261 | .field("follow_link" , &self.follow_link) |
262 | .field(name:"depth" , &self.depth) |
263 | .finish() |
264 | } |
265 | } |
266 | |
267 | impl DirEntryRaw { |
268 | fn path(&self) -> &Path { |
269 | &self.path |
270 | } |
271 | |
272 | fn into_path(self) -> PathBuf { |
273 | self.path |
274 | } |
275 | |
276 | fn path_is_symlink(&self) -> bool { |
277 | self.ty.is_symlink() || self.follow_link |
278 | } |
279 | |
280 | fn metadata(&self) -> Result<Metadata, Error> { |
281 | self.metadata_internal() |
282 | } |
283 | |
284 | #[cfg (windows)] |
285 | fn metadata_internal(&self) -> Result<fs::Metadata, Error> { |
286 | if self.follow_link { |
287 | fs::metadata(&self.path) |
288 | } else { |
289 | Ok(self.metadata.clone()) |
290 | } |
291 | .map_err(|err| Error::Io(io::Error::from(err)).with_path(&self.path)) |
292 | } |
293 | |
294 | #[cfg (not(windows))] |
295 | fn metadata_internal(&self) -> Result<fs::Metadata, Error> { |
296 | if self.follow_link { |
297 | fs::metadata(&self.path) |
298 | } else { |
299 | fs::symlink_metadata(&self.path) |
300 | } |
301 | .map_err(|err| Error::Io(io::Error::from(err)).with_path(&self.path)) |
302 | } |
303 | |
304 | fn file_type(&self) -> FileType { |
305 | self.ty |
306 | } |
307 | |
308 | fn file_name(&self) -> &OsStr { |
309 | self.path.file_name().unwrap_or_else(|| self.path.as_os_str()) |
310 | } |
311 | |
312 | fn depth(&self) -> usize { |
313 | self.depth |
314 | } |
315 | |
316 | #[cfg (unix)] |
317 | fn ino(&self) -> u64 { |
318 | self.ino |
319 | } |
320 | |
321 | fn from_entry( |
322 | depth: usize, |
323 | ent: &fs::DirEntry, |
324 | ) -> Result<DirEntryRaw, Error> { |
325 | let ty = ent.file_type().map_err(|err| { |
326 | let err = Error::Io(io::Error::from(err)).with_path(ent.path()); |
327 | Error::WithDepth { depth: depth, err: Box::new(err) } |
328 | })?; |
329 | DirEntryRaw::from_entry_os(depth, ent, ty) |
330 | } |
331 | |
332 | #[cfg (windows)] |
333 | fn from_entry_os( |
334 | depth: usize, |
335 | ent: &fs::DirEntry, |
336 | ty: fs::FileType, |
337 | ) -> Result<DirEntryRaw, Error> { |
338 | let md = ent.metadata().map_err(|err| { |
339 | let err = Error::Io(io::Error::from(err)).with_path(ent.path()); |
340 | Error::WithDepth { depth: depth, err: Box::new(err) } |
341 | })?; |
342 | Ok(DirEntryRaw { |
343 | path: ent.path(), |
344 | ty: ty, |
345 | follow_link: false, |
346 | depth: depth, |
347 | metadata: md, |
348 | }) |
349 | } |
350 | |
351 | #[cfg (unix)] |
352 | fn from_entry_os( |
353 | depth: usize, |
354 | ent: &fs::DirEntry, |
355 | ty: fs::FileType, |
356 | ) -> Result<DirEntryRaw, Error> { |
357 | use std::os::unix::fs::DirEntryExt; |
358 | |
359 | Ok(DirEntryRaw { |
360 | path: ent.path(), |
361 | ty: ty, |
362 | follow_link: false, |
363 | depth: depth, |
364 | ino: ent.ino(), |
365 | }) |
366 | } |
367 | |
368 | // Placeholder implementation to allow compiling on non-standard platforms |
369 | // (e.g. wasm32). |
370 | #[cfg (not(any(windows, unix)))] |
371 | fn from_entry_os( |
372 | depth: usize, |
373 | ent: &fs::DirEntry, |
374 | ty: fs::FileType, |
375 | ) -> Result<DirEntryRaw, Error> { |
376 | Err(Error::Io(io::Error::new( |
377 | io::ErrorKind::Other, |
378 | "unsupported platform" , |
379 | ))) |
380 | } |
381 | |
382 | #[cfg (windows)] |
383 | fn from_path( |
384 | depth: usize, |
385 | pb: PathBuf, |
386 | link: bool, |
387 | ) -> Result<DirEntryRaw, Error> { |
388 | let md = |
389 | fs::metadata(&pb).map_err(|err| Error::Io(err).with_path(&pb))?; |
390 | Ok(DirEntryRaw { |
391 | path: pb, |
392 | ty: md.file_type(), |
393 | follow_link: link, |
394 | depth: depth, |
395 | metadata: md, |
396 | }) |
397 | } |
398 | |
399 | #[cfg (unix)] |
400 | fn from_path( |
401 | depth: usize, |
402 | pb: PathBuf, |
403 | link: bool, |
404 | ) -> Result<DirEntryRaw, Error> { |
405 | use std::os::unix::fs::MetadataExt; |
406 | |
407 | let md = |
408 | fs::metadata(&pb).map_err(|err| Error::Io(err).with_path(&pb))?; |
409 | Ok(DirEntryRaw { |
410 | path: pb, |
411 | ty: md.file_type(), |
412 | follow_link: link, |
413 | depth: depth, |
414 | ino: md.ino(), |
415 | }) |
416 | } |
417 | |
418 | // Placeholder implementation to allow compiling on non-standard platforms |
419 | // (e.g. wasm32). |
420 | #[cfg (not(any(windows, unix)))] |
421 | fn from_path( |
422 | depth: usize, |
423 | pb: PathBuf, |
424 | link: bool, |
425 | ) -> Result<DirEntryRaw, Error> { |
426 | Err(Error::Io(io::Error::new( |
427 | io::ErrorKind::Other, |
428 | "unsupported platform" , |
429 | ))) |
430 | } |
431 | } |
432 | |
433 | /// WalkBuilder builds a recursive directory iterator. |
434 | /// |
435 | /// The builder supports a large number of configurable options. This includes |
436 | /// specific glob overrides, file type matching, toggling whether hidden |
437 | /// files are ignored or not, and of course, support for respecting gitignore |
438 | /// files. |
439 | /// |
440 | /// By default, all ignore files found are respected. This includes `.ignore`, |
441 | /// `.gitignore`, `.git/info/exclude` and even your global gitignore |
442 | /// globs, usually found in `$XDG_CONFIG_HOME/git/ignore`. |
443 | /// |
444 | /// Some standard recursive directory options are also supported, such as |
445 | /// limiting the recursive depth or whether to follow symbolic links (disabled |
446 | /// by default). |
447 | /// |
448 | /// # Ignore rules |
449 | /// |
450 | /// There are many rules that influence whether a particular file or directory |
451 | /// is skipped by this iterator. Those rules are documented here. Note that |
452 | /// the rules assume a default configuration. |
453 | /// |
454 | /// * First, glob overrides are checked. If a path matches a glob override, |
455 | /// then matching stops. The path is then only skipped if the glob that matched |
456 | /// the path is an ignore glob. (An override glob is a whitelist glob unless it |
457 | /// starts with a `!`, in which case it is an ignore glob.) |
458 | /// * Second, ignore files are checked. Ignore files currently only come from |
459 | /// git ignore files (`.gitignore`, `.git/info/exclude` and the configured |
460 | /// global gitignore file), plain `.ignore` files, which have the same format |
461 | /// as gitignore files, or explicitly added ignore files. The precedence order |
462 | /// is: `.ignore`, `.gitignore`, `.git/info/exclude`, global gitignore and |
463 | /// finally explicitly added ignore files. Note that precedence between |
464 | /// different types of ignore files is not impacted by the directory hierarchy; |
465 | /// any `.ignore` file overrides all `.gitignore` files. Within each precedence |
466 | /// level, more nested ignore files have a higher precedence than less nested |
467 | /// ignore files. |
468 | /// * Third, if the previous step yields an ignore match, then all matching |
469 | /// is stopped and the path is skipped. If it yields a whitelist match, then |
470 | /// matching continues. A whitelist match can be overridden by a later matcher. |
471 | /// * Fourth, unless the path is a directory, the file type matcher is run on |
472 | /// the path. As above, if it yields an ignore match, then all matching is |
473 | /// stopped and the path is skipped. If it yields a whitelist match, then |
474 | /// matching continues. |
475 | /// * Fifth, if the path hasn't been whitelisted and it is hidden, then the |
476 | /// path is skipped. |
477 | /// * Sixth, unless the path is a directory, the size of the file is compared |
478 | /// against the max filesize limit. If it exceeds the limit, it is skipped. |
479 | /// * Seventh, if the path has made it this far then it is yielded in the |
480 | /// iterator. |
481 | #[derive (Clone)] |
482 | pub struct WalkBuilder { |
483 | paths: Vec<PathBuf>, |
484 | ig_builder: IgnoreBuilder, |
485 | max_depth: Option<usize>, |
486 | max_filesize: Option<u64>, |
487 | follow_links: bool, |
488 | same_file_system: bool, |
489 | sorter: Option<Sorter>, |
490 | threads: usize, |
491 | skip: Option<Arc<Handle>>, |
492 | filter: Option<Filter>, |
493 | } |
494 | |
495 | #[derive (Clone)] |
496 | enum Sorter { |
497 | ByName( |
498 | Arc<dyn Fn(&OsStr, &OsStr) -> cmp::Ordering + Send + Sync + 'static>, |
499 | ), |
500 | ByPath(Arc<dyn Fn(&Path, &Path) -> cmp::Ordering + Send + Sync + 'static>), |
501 | } |
502 | |
503 | #[derive (Clone)] |
504 | struct Filter(Arc<dyn Fn(&DirEntry) -> bool + Send + Sync + 'static>); |
505 | |
506 | impl fmt::Debug for WalkBuilder { |
507 | fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result { |
508 | f&mut DebugStruct<'_, '_>.debug_struct("WalkBuilder" ) |
509 | .field("paths" , &self.paths) |
510 | .field("ig_builder" , &self.ig_builder) |
511 | .field("max_depth" , &self.max_depth) |
512 | .field("max_filesize" , &self.max_filesize) |
513 | .field("follow_links" , &self.follow_links) |
514 | .field("threads" , &self.threads) |
515 | .field(name:"skip" , &self.skip) |
516 | .finish() |
517 | } |
518 | } |
519 | |
520 | impl WalkBuilder { |
521 | /// Create a new builder for a recursive directory iterator for the |
522 | /// directory given. |
523 | /// |
524 | /// Note that if you want to traverse multiple different directories, it |
525 | /// is better to call `add` on this builder than to create multiple |
526 | /// `Walk` values. |
527 | pub fn new<P: AsRef<Path>>(path: P) -> WalkBuilder { |
528 | WalkBuilder { |
529 | paths: vec![path.as_ref().to_path_buf()], |
530 | ig_builder: IgnoreBuilder::new(), |
531 | max_depth: None, |
532 | max_filesize: None, |
533 | follow_links: false, |
534 | same_file_system: false, |
535 | sorter: None, |
536 | threads: 0, |
537 | skip: None, |
538 | filter: None, |
539 | } |
540 | } |
541 | |
542 | /// Build a new `Walk` iterator. |
543 | pub fn build(&self) -> Walk { |
544 | let follow_links = self.follow_links; |
545 | let max_depth = self.max_depth; |
546 | let sorter = self.sorter.clone(); |
547 | let its = self |
548 | .paths |
549 | .iter() |
550 | .map(move |p| { |
551 | if p == Path::new("-" ) { |
552 | (p.to_path_buf(), None) |
553 | } else { |
554 | let mut wd = WalkDir::new(p); |
555 | wd = wd.follow_links(follow_links || p.is_file()); |
556 | wd = wd.same_file_system(self.same_file_system); |
557 | if let Some(max_depth) = max_depth { |
558 | wd = wd.max_depth(max_depth); |
559 | } |
560 | if let Some(ref sorter) = sorter { |
561 | match sorter.clone() { |
562 | Sorter::ByName(cmp) => { |
563 | wd = wd.sort_by(move |a, b| { |
564 | cmp(a.file_name(), b.file_name()) |
565 | }); |
566 | } |
567 | Sorter::ByPath(cmp) => { |
568 | wd = wd.sort_by(move |a, b| { |
569 | cmp(a.path(), b.path()) |
570 | }); |
571 | } |
572 | } |
573 | } |
574 | (p.to_path_buf(), Some(WalkEventIter::from(wd))) |
575 | } |
576 | }) |
577 | .collect::<Vec<_>>() |
578 | .into_iter(); |
579 | let ig_root = self.ig_builder.build(); |
580 | Walk { |
581 | its: its, |
582 | it: None, |
583 | ig_root: ig_root.clone(), |
584 | ig: ig_root.clone(), |
585 | max_filesize: self.max_filesize, |
586 | skip: self.skip.clone(), |
587 | filter: self.filter.clone(), |
588 | } |
589 | } |
590 | |
591 | /// Build a new `WalkParallel` iterator. |
592 | /// |
593 | /// Note that this *doesn't* return something that implements `Iterator`. |
594 | /// Instead, the returned value must be run with a closure. e.g., |
595 | /// `builder.build_parallel().run(|| |path| println!("{:?}", path))`. |
596 | pub fn build_parallel(&self) -> WalkParallel { |
597 | WalkParallel { |
598 | paths: self.paths.clone().into_iter(), |
599 | ig_root: self.ig_builder.build(), |
600 | max_depth: self.max_depth, |
601 | max_filesize: self.max_filesize, |
602 | follow_links: self.follow_links, |
603 | same_file_system: self.same_file_system, |
604 | threads: self.threads, |
605 | skip: self.skip.clone(), |
606 | filter: self.filter.clone(), |
607 | } |
608 | } |
609 | |
610 | /// Add a file path to the iterator. |
611 | /// |
612 | /// Each additional file path added is traversed recursively. This should |
613 | /// be preferred over building multiple `Walk` iterators since this |
614 | /// enables reusing resources across iteration. |
615 | pub fn add<P: AsRef<Path>>(&mut self, path: P) -> &mut WalkBuilder { |
616 | self.paths.push(path.as_ref().to_path_buf()); |
617 | self |
618 | } |
619 | |
620 | /// The maximum depth to recurse. |
621 | /// |
622 | /// The default, `None`, imposes no depth restriction. |
623 | pub fn max_depth(&mut self, depth: Option<usize>) -> &mut WalkBuilder { |
624 | self.max_depth = depth; |
625 | self |
626 | } |
627 | |
628 | /// Whether to follow symbolic links or not. |
629 | pub fn follow_links(&mut self, yes: bool) -> &mut WalkBuilder { |
630 | self.follow_links = yes; |
631 | self |
632 | } |
633 | |
634 | /// Whether to ignore files above the specified limit. |
635 | pub fn max_filesize(&mut self, filesize: Option<u64>) -> &mut WalkBuilder { |
636 | self.max_filesize = filesize; |
637 | self |
638 | } |
639 | |
640 | /// The number of threads to use for traversal. |
641 | /// |
642 | /// Note that this only has an effect when using `build_parallel`. |
643 | /// |
644 | /// The default setting is `0`, which chooses the number of threads |
645 | /// automatically using heuristics. |
646 | pub fn threads(&mut self, n: usize) -> &mut WalkBuilder { |
647 | self.threads = n; |
648 | self |
649 | } |
650 | |
651 | /// Add a global ignore file to the matcher. |
652 | /// |
653 | /// This has lower precedence than all other sources of ignore rules. |
654 | /// |
655 | /// If there was a problem adding the ignore file, then an error is |
656 | /// returned. Note that the error may indicate *partial* failure. For |
657 | /// example, if an ignore file contains an invalid glob, all other globs |
658 | /// are still applied. |
659 | pub fn add_ignore<P: AsRef<Path>>(&mut self, path: P) -> Option<Error> { |
660 | let mut builder = GitignoreBuilder::new("" ); |
661 | let mut errs = PartialErrorBuilder::default(); |
662 | errs.maybe_push(builder.add(path)); |
663 | match builder.build() { |
664 | Ok(gi) => { |
665 | self.ig_builder.add_ignore(gi); |
666 | } |
667 | Err(err) => { |
668 | errs.push(err); |
669 | } |
670 | } |
671 | errs.into_error_option() |
672 | } |
673 | |
674 | /// Add a custom ignore file name |
675 | /// |
676 | /// These ignore files have higher precedence than all other ignore files. |
677 | /// |
678 | /// When specifying multiple names, earlier names have lower precedence than |
679 | /// later names. |
680 | pub fn add_custom_ignore_filename<S: AsRef<OsStr>>( |
681 | &mut self, |
682 | file_name: S, |
683 | ) -> &mut WalkBuilder { |
684 | self.ig_builder.add_custom_ignore_filename(file_name); |
685 | self |
686 | } |
687 | |
688 | /// Add an override matcher. |
689 | /// |
690 | /// By default, no override matcher is used. |
691 | /// |
692 | /// This overrides any previous setting. |
693 | pub fn overrides(&mut self, overrides: Override) -> &mut WalkBuilder { |
694 | self.ig_builder.overrides(overrides); |
695 | self |
696 | } |
697 | |
698 | /// Add a file type matcher. |
699 | /// |
700 | /// By default, no file type matcher is used. |
701 | /// |
702 | /// This overrides any previous setting. |
703 | pub fn types(&mut self, types: Types) -> &mut WalkBuilder { |
704 | self.ig_builder.types(types); |
705 | self |
706 | } |
707 | |
708 | /// Enables all the standard ignore filters. |
709 | /// |
710 | /// This toggles, as a group, all the filters that are enabled by default: |
711 | /// |
712 | /// - [hidden()](#method.hidden) |
713 | /// - [parents()](#method.parents) |
714 | /// - [ignore()](#method.ignore) |
715 | /// - [git_ignore()](#method.git_ignore) |
716 | /// - [git_global()](#method.git_global) |
717 | /// - [git_exclude()](#method.git_exclude) |
718 | /// |
719 | /// They may still be toggled individually after calling this function. |
720 | /// |
721 | /// This is (by definition) enabled by default. |
722 | pub fn standard_filters(&mut self, yes: bool) -> &mut WalkBuilder { |
723 | self.hidden(yes) |
724 | .parents(yes) |
725 | .ignore(yes) |
726 | .git_ignore(yes) |
727 | .git_global(yes) |
728 | .git_exclude(yes) |
729 | } |
730 | |
731 | /// Enables ignoring hidden files. |
732 | /// |
733 | /// This is enabled by default. |
734 | pub fn hidden(&mut self, yes: bool) -> &mut WalkBuilder { |
735 | self.ig_builder.hidden(yes); |
736 | self |
737 | } |
738 | |
739 | /// Enables reading ignore files from parent directories. |
740 | /// |
741 | /// If this is enabled, then .gitignore files in parent directories of each |
742 | /// file path given are respected. Otherwise, they are ignored. |
743 | /// |
744 | /// This is enabled by default. |
745 | pub fn parents(&mut self, yes: bool) -> &mut WalkBuilder { |
746 | self.ig_builder.parents(yes); |
747 | self |
748 | } |
749 | |
750 | /// Enables reading `.ignore` files. |
751 | /// |
752 | /// `.ignore` files have the same semantics as `gitignore` files and are |
753 | /// supported by search tools such as ripgrep and The Silver Searcher. |
754 | /// |
755 | /// This is enabled by default. |
756 | pub fn ignore(&mut self, yes: bool) -> &mut WalkBuilder { |
757 | self.ig_builder.ignore(yes); |
758 | self |
759 | } |
760 | |
761 | /// Enables reading a global gitignore file, whose path is specified in |
762 | /// git's `core.excludesFile` config option. |
763 | /// |
764 | /// Git's config file location is `$HOME/.gitconfig`. If `$HOME/.gitconfig` |
765 | /// does not exist or does not specify `core.excludesFile`, then |
766 | /// `$XDG_CONFIG_HOME/git/ignore` is read. If `$XDG_CONFIG_HOME` is not |
767 | /// set or is empty, then `$HOME/.config/git/ignore` is used instead. |
768 | /// |
769 | /// This is enabled by default. |
770 | pub fn git_global(&mut self, yes: bool) -> &mut WalkBuilder { |
771 | self.ig_builder.git_global(yes); |
772 | self |
773 | } |
774 | |
775 | /// Enables reading `.gitignore` files. |
776 | /// |
777 | /// `.gitignore` files have match semantics as described in the `gitignore` |
778 | /// man page. |
779 | /// |
780 | /// This is enabled by default. |
781 | pub fn git_ignore(&mut self, yes: bool) -> &mut WalkBuilder { |
782 | self.ig_builder.git_ignore(yes); |
783 | self |
784 | } |
785 | |
786 | /// Enables reading `.git/info/exclude` files. |
787 | /// |
788 | /// `.git/info/exclude` files have match semantics as described in the |
789 | /// `gitignore` man page. |
790 | /// |
791 | /// This is enabled by default. |
792 | pub fn git_exclude(&mut self, yes: bool) -> &mut WalkBuilder { |
793 | self.ig_builder.git_exclude(yes); |
794 | self |
795 | } |
796 | |
797 | /// Whether a git repository is required to apply git-related ignore |
798 | /// rules (global rules, .gitignore and local exclude rules). |
799 | /// |
800 | /// When disabled, git-related ignore rules are applied even when searching |
801 | /// outside a git repository. |
802 | pub fn require_git(&mut self, yes: bool) -> &mut WalkBuilder { |
803 | self.ig_builder.require_git(yes); |
804 | self |
805 | } |
806 | |
807 | /// Process ignore files case insensitively |
808 | /// |
809 | /// This is disabled by default. |
810 | pub fn ignore_case_insensitive(&mut self, yes: bool) -> &mut WalkBuilder { |
811 | self.ig_builder.ignore_case_insensitive(yes); |
812 | self |
813 | } |
814 | |
815 | /// Set a function for sorting directory entries by their path. |
816 | /// |
817 | /// If a compare function is set, the resulting iterator will return all |
818 | /// paths in sorted order. The compare function will be called to compare |
819 | /// entries from the same directory. |
820 | /// |
821 | /// This is like `sort_by_file_name`, except the comparator accepts |
822 | /// a `&Path` instead of the base file name, which permits it to sort by |
823 | /// more criteria. |
824 | /// |
825 | /// This method will override any previous sorter set by this method or |
826 | /// by `sort_by_file_name`. |
827 | /// |
828 | /// Note that this is not used in the parallel iterator. |
829 | pub fn sort_by_file_path<F>(&mut self, cmp: F) -> &mut WalkBuilder |
830 | where |
831 | F: Fn(&Path, &Path) -> cmp::Ordering + Send + Sync + 'static, |
832 | { |
833 | self.sorter = Some(Sorter::ByPath(Arc::new(cmp))); |
834 | self |
835 | } |
836 | |
837 | /// Set a function for sorting directory entries by file name. |
838 | /// |
839 | /// If a compare function is set, the resulting iterator will return all |
840 | /// paths in sorted order. The compare function will be called to compare |
841 | /// names from entries from the same directory using only the name of the |
842 | /// entry. |
843 | /// |
844 | /// This method will override any previous sorter set by this method or |
845 | /// by `sort_by_file_path`. |
846 | /// |
847 | /// Note that this is not used in the parallel iterator. |
848 | pub fn sort_by_file_name<F>(&mut self, cmp: F) -> &mut WalkBuilder |
849 | where |
850 | F: Fn(&OsStr, &OsStr) -> cmp::Ordering + Send + Sync + 'static, |
851 | { |
852 | self.sorter = Some(Sorter::ByName(Arc::new(cmp))); |
853 | self |
854 | } |
855 | |
856 | /// Do not cross file system boundaries. |
857 | /// |
858 | /// When this option is enabled, directory traversal will not descend into |
859 | /// directories that are on a different file system from the root path. |
860 | /// |
861 | /// Currently, this option is only supported on Unix and Windows. If this |
862 | /// option is used on an unsupported platform, then directory traversal |
863 | /// will immediately return an error and will not yield any entries. |
864 | pub fn same_file_system(&mut self, yes: bool) -> &mut WalkBuilder { |
865 | self.same_file_system = yes; |
866 | self |
867 | } |
868 | |
869 | /// Do not yield directory entries that are believed to correspond to |
870 | /// stdout. |
871 | /// |
872 | /// This is useful when a command is invoked via shell redirection to a |
873 | /// file that is also being read. For example, `grep -r foo ./ > results` |
874 | /// might end up trying to search `results` even though it is also writing |
875 | /// to it, which could cause an unbounded feedback loop. Setting this |
876 | /// option prevents this from happening by skipping over the `results` |
877 | /// file. |
878 | /// |
879 | /// This is disabled by default. |
880 | pub fn skip_stdout(&mut self, yes: bool) -> &mut WalkBuilder { |
881 | if yes { |
882 | self.skip = stdout_handle().map(Arc::new); |
883 | } else { |
884 | self.skip = None; |
885 | } |
886 | self |
887 | } |
888 | |
889 | /// Yields only entries which satisfy the given predicate and skips |
890 | /// descending into directories that do not satisfy the given predicate. |
891 | /// |
892 | /// The predicate is applied to all entries. If the predicate is |
893 | /// true, iteration carries on as normal. If the predicate is false, the |
894 | /// entry is ignored and if it is a directory, it is not descended into. |
895 | /// |
896 | /// Note that the errors for reading entries that may not satisfy the |
897 | /// predicate will still be yielded. |
898 | pub fn filter_entry<P>(&mut self, filter: P) -> &mut WalkBuilder |
899 | where |
900 | P: Fn(&DirEntry) -> bool + Send + Sync + 'static, |
901 | { |
902 | self.filter = Some(Filter(Arc::new(filter))); |
903 | self |
904 | } |
905 | } |
906 | |
907 | /// Walk is a recursive directory iterator over file paths in one or more |
908 | /// directories. |
909 | /// |
910 | /// Only file and directory paths matching the rules are returned. By default, |
911 | /// ignore files like `.gitignore` are respected. The precise matching rules |
912 | /// and precedence is explained in the documentation for `WalkBuilder`. |
913 | pub struct Walk { |
914 | its: vec::IntoIter<(PathBuf, Option<WalkEventIter>)>, |
915 | it: Option<WalkEventIter>, |
916 | ig_root: Ignore, |
917 | ig: Ignore, |
918 | max_filesize: Option<u64>, |
919 | skip: Option<Arc<Handle>>, |
920 | filter: Option<Filter>, |
921 | } |
922 | |
923 | impl Walk { |
924 | /// Creates a new recursive directory iterator for the file path given. |
925 | /// |
926 | /// Note that this uses default settings, which include respecting |
927 | /// `.gitignore` files. To configure the iterator, use `WalkBuilder` |
928 | /// instead. |
929 | pub fn new<P: AsRef<Path>>(path: P) -> Walk { |
930 | WalkBuilder::new(path).build() |
931 | } |
932 | |
933 | fn skip_entry(&self, ent: &DirEntry) -> Result<bool, Error> { |
934 | if ent.depth() == 0 { |
935 | return Ok(false); |
936 | } |
937 | // We ensure that trivial skipping is done before any other potentially |
938 | // expensive operations (stat, filesystem other) are done. This seems |
939 | // like an obvious optimization but becomes critical when filesystem |
940 | // operations even as simple as stat can result in significant |
941 | // overheads; an example of this was a bespoke filesystem layer in |
942 | // Windows that hosted files remotely and would download them on-demand |
943 | // when particular filesystem operations occurred. Users of this system |
944 | // who ensured correct file-type filters were being used could still |
945 | // get unnecessary file access resulting in large downloads. |
946 | if should_skip_entry(&self.ig, ent) { |
947 | return Ok(true); |
948 | } |
949 | if let Some(ref stdout) = self.skip { |
950 | if path_equals(ent, stdout)? { |
951 | return Ok(true); |
952 | } |
953 | } |
954 | if self.max_filesize.is_some() && !ent.is_dir() { |
955 | return Ok(skip_filesize( |
956 | self.max_filesize.unwrap(), |
957 | ent.path(), |
958 | &ent.metadata().ok(), |
959 | )); |
960 | } |
961 | if let Some(Filter(filter)) = &self.filter { |
962 | if !filter(ent) { |
963 | return Ok(true); |
964 | } |
965 | } |
966 | Ok(false) |
967 | } |
968 | } |
969 | |
970 | impl Iterator for Walk { |
971 | type Item = Result<DirEntry, Error>; |
972 | |
973 | #[inline (always)] |
974 | fn next(&mut self) -> Option<Result<DirEntry, Error>> { |
975 | loop { |
976 | let ev = match self.it.as_mut().and_then(|it| it.next()) { |
977 | Some(ev) => ev, |
978 | None => { |
979 | match self.its.next() { |
980 | None => return None, |
981 | Some((_, None)) => { |
982 | return Some(Ok(DirEntry::new_stdin())); |
983 | } |
984 | Some((path, Some(it))) => { |
985 | self.it = Some(it); |
986 | if path.is_dir() { |
987 | let (ig, err) = self.ig_root.add_parents(path); |
988 | self.ig = ig; |
989 | if let Some(err) = err { |
990 | return Some(Err(err)); |
991 | } |
992 | } else { |
993 | self.ig = self.ig_root.clone(); |
994 | } |
995 | } |
996 | } |
997 | continue; |
998 | } |
999 | }; |
1000 | match ev { |
1001 | Err(err) => { |
1002 | return Some(Err(Error::from_walkdir(err))); |
1003 | } |
1004 | Ok(WalkEvent::Exit) => { |
1005 | self.ig = self.ig.parent().unwrap(); |
1006 | } |
1007 | Ok(WalkEvent::Dir(ent)) => { |
1008 | let mut ent = DirEntry::new_walkdir(ent, None); |
1009 | let should_skip = match self.skip_entry(&ent) { |
1010 | Err(err) => return Some(Err(err)), |
1011 | Ok(should_skip) => should_skip, |
1012 | }; |
1013 | if should_skip { |
1014 | self.it.as_mut().unwrap().it.skip_current_dir(); |
1015 | // Still need to push this on the stack because |
1016 | // we'll get a WalkEvent::Exit event for this dir. |
1017 | // We don't care if it errors though. |
1018 | let (igtmp, _) = self.ig.add_child(ent.path()); |
1019 | self.ig = igtmp; |
1020 | continue; |
1021 | } |
1022 | let (igtmp, err) = self.ig.add_child(ent.path()); |
1023 | self.ig = igtmp; |
1024 | ent.err = err; |
1025 | return Some(Ok(ent)); |
1026 | } |
1027 | Ok(WalkEvent::File(ent)) => { |
1028 | let ent = DirEntry::new_walkdir(ent, None); |
1029 | let should_skip = match self.skip_entry(&ent) { |
1030 | Err(err) => return Some(Err(err)), |
1031 | Ok(should_skip) => should_skip, |
1032 | }; |
1033 | if should_skip { |
1034 | continue; |
1035 | } |
1036 | return Some(Ok(ent)); |
1037 | } |
1038 | } |
1039 | } |
1040 | } |
1041 | } |
1042 | |
1043 | /// WalkEventIter transforms a WalkDir iterator into an iterator that more |
1044 | /// accurately describes the directory tree. Namely, it emits events that are |
1045 | /// one of three types: directory, file or "exit." An "exit" event means that |
1046 | /// the entire contents of a directory have been enumerated. |
1047 | struct WalkEventIter { |
1048 | depth: usize, |
1049 | it: walkdir::IntoIter, |
1050 | next: Option<Result<walkdir::DirEntry, walkdir::Error>>, |
1051 | } |
1052 | |
1053 | #[derive (Debug)] |
1054 | enum WalkEvent { |
1055 | Dir(walkdir::DirEntry), |
1056 | File(walkdir::DirEntry), |
1057 | Exit, |
1058 | } |
1059 | |
1060 | impl From<WalkDir> for WalkEventIter { |
1061 | fn from(it: WalkDir) -> WalkEventIter { |
1062 | WalkEventIter { depth: 0, it: it.into_iter(), next: None } |
1063 | } |
1064 | } |
1065 | |
1066 | impl Iterator for WalkEventIter { |
1067 | type Item = walkdir::Result<WalkEvent>; |
1068 | |
1069 | #[inline (always)] |
1070 | fn next(&mut self) -> Option<walkdir::Result<WalkEvent>> { |
1071 | let dent = self.next.take().or_else(|| self.it.next()); |
1072 | let depth = match dent { |
1073 | None => 0, |
1074 | Some(Ok(ref dent)) => dent.depth(), |
1075 | Some(Err(ref err)) => err.depth(), |
1076 | }; |
1077 | if depth < self.depth { |
1078 | self.depth -= 1; |
1079 | self.next = dent; |
1080 | return Some(Ok(WalkEvent::Exit)); |
1081 | } |
1082 | self.depth = depth; |
1083 | match dent { |
1084 | None => None, |
1085 | Some(Err(err)) => Some(Err(err)), |
1086 | Some(Ok(dent)) => { |
1087 | if walkdir_is_dir(&dent) { |
1088 | self.depth += 1; |
1089 | Some(Ok(WalkEvent::Dir(dent))) |
1090 | } else { |
1091 | Some(Ok(WalkEvent::File(dent))) |
1092 | } |
1093 | } |
1094 | } |
1095 | } |
1096 | } |
1097 | |
1098 | /// WalkState is used in the parallel recursive directory iterator to indicate |
1099 | /// whether walking should continue as normal, skip descending into a |
1100 | /// particular directory or quit the walk entirely. |
1101 | #[derive (Clone, Copy, Debug, Eq, PartialEq)] |
1102 | pub enum WalkState { |
1103 | /// Continue walking as normal. |
1104 | Continue, |
1105 | /// If the directory entry given is a directory, don't descend into it. |
1106 | /// In all other cases, this has no effect. |
1107 | Skip, |
1108 | /// Quit the entire iterator as soon as possible. |
1109 | /// |
1110 | /// Note that this is an inherently asynchronous action. It is possible |
1111 | /// for more entries to be yielded even after instructing the iterator |
1112 | /// to quit. |
1113 | Quit, |
1114 | } |
1115 | |
1116 | impl WalkState { |
1117 | fn is_continue(&self) -> bool { |
1118 | *self == WalkState::Continue |
1119 | } |
1120 | |
1121 | fn is_quit(&self) -> bool { |
1122 | *self == WalkState::Quit |
1123 | } |
1124 | } |
1125 | |
1126 | /// A builder for constructing a visitor when using |
1127 | /// [`WalkParallel::visit`](struct.WalkParallel.html#method.visit). The builder |
1128 | /// will be called for each thread started by `WalkParallel`. The visitor |
1129 | /// returned from each builder is then called for every directory entry. |
1130 | pub trait ParallelVisitorBuilder<'s> { |
1131 | /// Create per-thread `ParallelVisitor`s for `WalkParallel`. |
1132 | fn build(&mut self) -> Box<dyn ParallelVisitor + 's>; |
1133 | } |
1134 | |
1135 | impl<'a, 's, P: ParallelVisitorBuilder<'s>> ParallelVisitorBuilder<'s> |
1136 | for &'a mut P |
1137 | { |
1138 | fn build(&mut self) -> Box<dyn ParallelVisitor + 's> { |
1139 | (**self).build() |
1140 | } |
1141 | } |
1142 | |
1143 | /// Receives files and directories for the current thread. |
1144 | /// |
1145 | /// Setup for the traversal can be implemented as part of |
1146 | /// [`ParallelVisitorBuilder::build`](trait.ParallelVisitorBuilder.html#tymethod.build). |
1147 | /// Teardown when traversal finishes can be implemented by implementing the |
1148 | /// `Drop` trait on your traversal type. |
1149 | pub trait ParallelVisitor: Send { |
1150 | /// Receives files and directories for the current thread. This is called |
1151 | /// once for every directory entry visited by traversal. |
1152 | fn visit(&mut self, entry: Result<DirEntry, Error>) -> WalkState; |
1153 | } |
1154 | |
1155 | struct FnBuilder<F> { |
1156 | builder: F, |
1157 | } |
1158 | |
1159 | impl<'s, F: FnMut() -> FnVisitor<'s>> ParallelVisitorBuilder<'s> |
1160 | for FnBuilder<F> |
1161 | { |
1162 | fn build(&mut self) -> Box<dyn ParallelVisitor + 's> { |
1163 | let visitor: Box) -> … + Send> = (self.builder)(); |
1164 | Box::new(FnVisitorImp { visitor }) |
1165 | } |
1166 | } |
1167 | |
1168 | type FnVisitor<'s> = |
1169 | Box<dyn FnMut(Result<DirEntry, Error>) -> WalkState + Send + 's>; |
1170 | |
1171 | struct FnVisitorImp<'s> { |
1172 | visitor: FnVisitor<'s>, |
1173 | } |
1174 | |
1175 | impl<'s> ParallelVisitor for FnVisitorImp<'s> { |
1176 | fn visit(&mut self, entry: Result<DirEntry, Error>) -> WalkState { |
1177 | (self.visitor)(entry) |
1178 | } |
1179 | } |
1180 | |
1181 | /// WalkParallel is a parallel recursive directory iterator over files paths |
1182 | /// in one or more directories. |
1183 | /// |
1184 | /// Only file and directory paths matching the rules are returned. By default, |
1185 | /// ignore files like `.gitignore` are respected. The precise matching rules |
1186 | /// and precedence is explained in the documentation for `WalkBuilder`. |
1187 | /// |
1188 | /// Unlike `Walk`, this uses multiple threads for traversing a directory. |
1189 | pub struct WalkParallel { |
1190 | paths: vec::IntoIter<PathBuf>, |
1191 | ig_root: Ignore, |
1192 | max_filesize: Option<u64>, |
1193 | max_depth: Option<usize>, |
1194 | follow_links: bool, |
1195 | same_file_system: bool, |
1196 | threads: usize, |
1197 | skip: Option<Arc<Handle>>, |
1198 | filter: Option<Filter>, |
1199 | } |
1200 | |
1201 | impl WalkParallel { |
1202 | /// Execute the parallel recursive directory iterator. `mkf` is called |
1203 | /// for each thread used for iteration. The function produced by `mkf` |
1204 | /// is then in turn called for each visited file path. |
1205 | pub fn run<'s, F>(self, mkf: F) |
1206 | where |
1207 | F: FnMut() -> FnVisitor<'s>, |
1208 | { |
1209 | self.visit(&mut FnBuilder { builder: mkf }) |
1210 | } |
1211 | |
1212 | /// Execute the parallel recursive directory iterator using a custom |
1213 | /// visitor. |
1214 | /// |
1215 | /// The builder given is used to construct a visitor for every thread |
1216 | /// used by this traversal. The visitor returned from each builder is then |
1217 | /// called for every directory entry seen by that thread. |
1218 | /// |
1219 | /// Typically, creating a custom visitor is useful if you need to perform |
1220 | /// some kind of cleanup once traversal is finished. This can be achieved |
1221 | /// by implementing `Drop` for your builder (or for your visitor, if you |
1222 | /// want to execute cleanup for every thread that is launched). |
1223 | /// |
1224 | /// For example, each visitor might build up a data structure of results |
1225 | /// corresponding to the directory entries seen for each thread. Since each |
1226 | /// visitor runs on only one thread, this build-up can be done without |
1227 | /// synchronization. Then, once traversal is complete, all of the results |
1228 | /// can be merged together into a single data structure. |
1229 | pub fn visit(mut self, builder: &mut dyn ParallelVisitorBuilder<'_>) { |
1230 | let threads = self.threads(); |
1231 | let stack = Arc::new(Mutex::new(vec![])); |
1232 | { |
1233 | let mut stack = stack.lock().unwrap(); |
1234 | let mut visitor = builder.build(); |
1235 | let mut paths = Vec::new().into_iter(); |
1236 | std::mem::swap(&mut paths, &mut self.paths); |
1237 | // Send the initial set of root paths to the pool of workers. Note |
1238 | // that we only send directories. For files, we send to them the |
1239 | // callback directly. |
1240 | for path in paths { |
1241 | let (dent, root_device) = if path == Path::new("-" ) { |
1242 | (DirEntry::new_stdin(), None) |
1243 | } else { |
1244 | let root_device = if !self.same_file_system { |
1245 | None |
1246 | } else { |
1247 | match device_num(&path) { |
1248 | Ok(root_device) => Some(root_device), |
1249 | Err(err) => { |
1250 | let err = Error::Io(err).with_path(path); |
1251 | if visitor.visit(Err(err)).is_quit() { |
1252 | return; |
1253 | } |
1254 | continue; |
1255 | } |
1256 | } |
1257 | }; |
1258 | match DirEntryRaw::from_path(0, path, false) { |
1259 | Ok(dent) => { |
1260 | (DirEntry::new_raw(dent, None), root_device) |
1261 | } |
1262 | Err(err) => { |
1263 | if visitor.visit(Err(err)).is_quit() { |
1264 | return; |
1265 | } |
1266 | continue; |
1267 | } |
1268 | } |
1269 | }; |
1270 | stack.push(Message::Work(Work { |
1271 | dent: dent, |
1272 | ignore: self.ig_root.clone(), |
1273 | root_device: root_device, |
1274 | })); |
1275 | } |
1276 | // ... but there's no need to start workers if we don't need them. |
1277 | if stack.is_empty() { |
1278 | return; |
1279 | } |
1280 | } |
1281 | // Create the workers and then wait for them to finish. |
1282 | let quit_now = Arc::new(AtomicBool::new(false)); |
1283 | let num_pending = |
1284 | Arc::new(AtomicUsize::new(stack.lock().unwrap().len())); |
1285 | std::thread::scope(|s| { |
1286 | let mut handles = vec![]; |
1287 | for _ in 0..threads { |
1288 | let worker = Worker { |
1289 | visitor: builder.build(), |
1290 | stack: stack.clone(), |
1291 | quit_now: quit_now.clone(), |
1292 | num_pending: num_pending.clone(), |
1293 | max_depth: self.max_depth, |
1294 | max_filesize: self.max_filesize, |
1295 | follow_links: self.follow_links, |
1296 | skip: self.skip.clone(), |
1297 | filter: self.filter.clone(), |
1298 | }; |
1299 | handles.push(s.spawn(|| worker.run())); |
1300 | } |
1301 | for handle in handles { |
1302 | handle.join().unwrap(); |
1303 | } |
1304 | }); |
1305 | } |
1306 | |
1307 | fn threads(&self) -> usize { |
1308 | if self.threads == 0 { |
1309 | 2 |
1310 | } else { |
1311 | self.threads |
1312 | } |
1313 | } |
1314 | } |
1315 | |
1316 | /// Message is the set of instructions that a worker knows how to process. |
1317 | enum Message { |
1318 | /// A work item corresponds to a directory that should be descended into. |
1319 | /// Work items for entries that should be skipped or ignored should not |
1320 | /// be produced. |
1321 | Work(Work), |
1322 | /// This instruction indicates that the worker should quit. |
1323 | Quit, |
1324 | } |
1325 | |
1326 | /// A unit of work for each worker to process. |
1327 | /// |
1328 | /// Each unit of work corresponds to a directory that should be descended |
1329 | /// into. |
1330 | struct Work { |
1331 | /// The directory entry. |
1332 | dent: DirEntry, |
1333 | /// Any ignore matchers that have been built for this directory's parents. |
1334 | ignore: Ignore, |
1335 | /// The root device number. When present, only files with the same device |
1336 | /// number should be considered. |
1337 | root_device: Option<u64>, |
1338 | } |
1339 | |
1340 | impl Work { |
1341 | /// Returns true if and only if this work item is a directory. |
1342 | fn is_dir(&self) -> bool { |
1343 | self.dent.is_dir() |
1344 | } |
1345 | |
1346 | /// Returns true if and only if this work item is a symlink. |
1347 | fn is_symlink(&self) -> bool { |
1348 | self.dent.file_type().map_or(false, |ft| ft.is_symlink()) |
1349 | } |
1350 | |
1351 | /// Adds ignore rules for parent directories. |
1352 | /// |
1353 | /// Note that this only applies to entries at depth 0. On all other |
1354 | /// entries, this is a no-op. |
1355 | fn add_parents(&mut self) -> Option<Error> { |
1356 | if self.dent.depth() > 0 { |
1357 | return None; |
1358 | } |
1359 | // At depth 0, the path of this entry is a root path, so we can |
1360 | // use it directly to add parent ignore rules. |
1361 | let (ig, err) = self.ignore.add_parents(self.dent.path()); |
1362 | self.ignore = ig; |
1363 | err |
1364 | } |
1365 | |
1366 | /// Reads the directory contents of this work item and adds ignore |
1367 | /// rules for this directory. |
1368 | /// |
1369 | /// If there was a problem with reading the directory contents, then |
1370 | /// an error is returned. If there was a problem reading the ignore |
1371 | /// rules for this directory, then the error is attached to this |
1372 | /// work item's directory entry. |
1373 | fn read_dir(&mut self) -> Result<fs::ReadDir, Error> { |
1374 | let readdir = match fs::read_dir(self.dent.path()) { |
1375 | Ok(readdir) => readdir, |
1376 | Err(err) => { |
1377 | let err = Error::from(err) |
1378 | .with_path(self.dent.path()) |
1379 | .with_depth(self.dent.depth()); |
1380 | return Err(err); |
1381 | } |
1382 | }; |
1383 | let (ig, err) = self.ignore.add_child(self.dent.path()); |
1384 | self.ignore = ig; |
1385 | self.dent.err = err; |
1386 | Ok(readdir) |
1387 | } |
1388 | } |
1389 | |
1390 | /// A worker is responsible for descending into directories, updating the |
1391 | /// ignore matchers, producing new work and invoking the caller's callback. |
1392 | /// |
1393 | /// Note that a worker is *both* a producer and a consumer. |
1394 | struct Worker<'s> { |
1395 | /// The caller's callback. |
1396 | visitor: Box<dyn ParallelVisitor + 's>, |
1397 | /// A stack of work to do. |
1398 | /// |
1399 | /// We use a stack instead of a channel because a stack lets us visit |
1400 | /// directories in depth first order. This can substantially reduce peak |
1401 | /// memory usage by keeping both the number of files path and gitignore |
1402 | /// matchers in memory lower. |
1403 | stack: Arc<Mutex<Vec<Message>>>, |
1404 | /// Whether all workers should terminate at the next opportunity. Note |
1405 | /// that we need this because we don't want other `Work` to be done after |
1406 | /// we quit. We wouldn't need this if have a priority channel. |
1407 | quit_now: Arc<AtomicBool>, |
1408 | /// The number of outstanding work items. |
1409 | num_pending: Arc<AtomicUsize>, |
1410 | /// The maximum depth of directories to descend. A value of `0` means no |
1411 | /// descension at all. |
1412 | max_depth: Option<usize>, |
1413 | /// The maximum size a searched file can be (in bytes). If a file exceeds |
1414 | /// this size it will be skipped. |
1415 | max_filesize: Option<u64>, |
1416 | /// Whether to follow symbolic links or not. When this is enabled, loop |
1417 | /// detection is performed. |
1418 | follow_links: bool, |
1419 | /// A file handle to skip, currently is either `None` or stdout, if it's |
1420 | /// a file and it has been requested to skip files identical to stdout. |
1421 | skip: Option<Arc<Handle>>, |
1422 | /// A predicate applied to dir entries. If true, the entry and all |
1423 | /// children will be skipped. |
1424 | filter: Option<Filter>, |
1425 | } |
1426 | |
1427 | impl<'s> Worker<'s> { |
1428 | /// Runs this worker until there is no more work left to do. |
1429 | /// |
1430 | /// The worker will call the caller's callback for all entries that aren't |
1431 | /// skipped by the ignore matcher. |
1432 | fn run(mut self) { |
1433 | while let Some(work) = self.get_work() { |
1434 | if let WalkState::Quit = self.run_one(work) { |
1435 | self.quit_now(); |
1436 | } |
1437 | self.work_done(); |
1438 | } |
1439 | } |
1440 | |
1441 | fn run_one(&mut self, mut work: Work) -> WalkState { |
1442 | // If the work is not a directory, then we can just execute the |
1443 | // caller's callback immediately and move on. |
1444 | if work.is_symlink() || !work.is_dir() { |
1445 | return self.visitor.visit(Ok(work.dent)); |
1446 | } |
1447 | if let Some(err) = work.add_parents() { |
1448 | let state = self.visitor.visit(Err(err)); |
1449 | if state.is_quit() { |
1450 | return state; |
1451 | } |
1452 | } |
1453 | |
1454 | let descend = if let Some(root_device) = work.root_device { |
1455 | match is_same_file_system(root_device, work.dent.path()) { |
1456 | Ok(true) => true, |
1457 | Ok(false) => false, |
1458 | Err(err) => { |
1459 | let state = self.visitor.visit(Err(err)); |
1460 | if state.is_quit() { |
1461 | return state; |
1462 | } |
1463 | false |
1464 | } |
1465 | } |
1466 | } else { |
1467 | true |
1468 | }; |
1469 | |
1470 | // Try to read the directory first before we transfer ownership |
1471 | // to the provided closure. Do not unwrap it immediately, though, |
1472 | // as we may receive an `Err` value e.g. in the case when we do not |
1473 | // have sufficient read permissions to list the directory. |
1474 | // In that case we still want to provide the closure with a valid |
1475 | // entry before passing the error value. |
1476 | let readdir = work.read_dir(); |
1477 | let depth = work.dent.depth(); |
1478 | let state = self.visitor.visit(Ok(work.dent)); |
1479 | if !state.is_continue() { |
1480 | return state; |
1481 | } |
1482 | if !descend { |
1483 | return WalkState::Skip; |
1484 | } |
1485 | |
1486 | let readdir = match readdir { |
1487 | Ok(readdir) => readdir, |
1488 | Err(err) => { |
1489 | return self.visitor.visit(Err(err)); |
1490 | } |
1491 | }; |
1492 | |
1493 | if self.max_depth.map_or(false, |max| depth >= max) { |
1494 | return WalkState::Skip; |
1495 | } |
1496 | for result in readdir { |
1497 | let state = self.generate_work( |
1498 | &work.ignore, |
1499 | depth + 1, |
1500 | work.root_device, |
1501 | result, |
1502 | ); |
1503 | if state.is_quit() { |
1504 | return state; |
1505 | } |
1506 | } |
1507 | WalkState::Continue |
1508 | } |
1509 | |
1510 | /// Decides whether to submit the given directory entry as a file to |
1511 | /// search. |
1512 | /// |
1513 | /// If the entry is a path that should be ignored, then this is a no-op. |
1514 | /// Otherwise, the entry is pushed on to the queue. (The actual execution |
1515 | /// of the callback happens in `run_one`.) |
1516 | /// |
1517 | /// If an error occurs while reading the entry, then it is sent to the |
1518 | /// caller's callback. |
1519 | /// |
1520 | /// `ig` is the `Ignore` matcher for the parent directory. `depth` should |
1521 | /// be the depth of this entry. `result` should be the item yielded by |
1522 | /// a directory iterator. |
1523 | fn generate_work( |
1524 | &mut self, |
1525 | ig: &Ignore, |
1526 | depth: usize, |
1527 | root_device: Option<u64>, |
1528 | result: Result<fs::DirEntry, io::Error>, |
1529 | ) -> WalkState { |
1530 | let fs_dent = match result { |
1531 | Ok(fs_dent) => fs_dent, |
1532 | Err(err) => { |
1533 | return self |
1534 | .visitor |
1535 | .visit(Err(Error::from(err).with_depth(depth))); |
1536 | } |
1537 | }; |
1538 | let mut dent = match DirEntryRaw::from_entry(depth, &fs_dent) { |
1539 | Ok(dent) => DirEntry::new_raw(dent, None), |
1540 | Err(err) => { |
1541 | return self.visitor.visit(Err(err)); |
1542 | } |
1543 | }; |
1544 | let is_symlink = dent.file_type().map_or(false, |ft| ft.is_symlink()); |
1545 | if self.follow_links && is_symlink { |
1546 | let path = dent.path().to_path_buf(); |
1547 | dent = match DirEntryRaw::from_path(depth, path, true) { |
1548 | Ok(dent) => DirEntry::new_raw(dent, None), |
1549 | Err(err) => { |
1550 | return self.visitor.visit(Err(err)); |
1551 | } |
1552 | }; |
1553 | if dent.is_dir() { |
1554 | if let Err(err) = check_symlink_loop(ig, dent.path(), depth) { |
1555 | return self.visitor.visit(Err(err)); |
1556 | } |
1557 | } |
1558 | } |
1559 | // N.B. See analogous call in the single-threaded implementation about |
1560 | // why it's important for this to come before the checks below. |
1561 | if should_skip_entry(ig, &dent) { |
1562 | return WalkState::Continue; |
1563 | } |
1564 | if let Some(ref stdout) = self.skip { |
1565 | let is_stdout = match path_equals(&dent, stdout) { |
1566 | Ok(is_stdout) => is_stdout, |
1567 | Err(err) => return self.visitor.visit(Err(err)), |
1568 | }; |
1569 | if is_stdout { |
1570 | return WalkState::Continue; |
1571 | } |
1572 | } |
1573 | let should_skip_filesize = |
1574 | if self.max_filesize.is_some() && !dent.is_dir() { |
1575 | skip_filesize( |
1576 | self.max_filesize.unwrap(), |
1577 | dent.path(), |
1578 | &dent.metadata().ok(), |
1579 | ) |
1580 | } else { |
1581 | false |
1582 | }; |
1583 | let should_skip_filtered = |
1584 | if let Some(Filter(predicate)) = &self.filter { |
1585 | !predicate(&dent) |
1586 | } else { |
1587 | false |
1588 | }; |
1589 | if !should_skip_filesize && !should_skip_filtered { |
1590 | self.send(Work { dent, ignore: ig.clone(), root_device }); |
1591 | } |
1592 | WalkState::Continue |
1593 | } |
1594 | |
1595 | /// Returns the next directory to descend into. |
1596 | /// |
1597 | /// If all work has been exhausted, then this returns None. The worker |
1598 | /// should then subsequently quit. |
1599 | fn get_work(&mut self) -> Option<Work> { |
1600 | let mut value = self.recv(); |
1601 | loop { |
1602 | // Simulate a priority channel: If quit_now flag is set, we can |
1603 | // receive only quit messages. |
1604 | if self.is_quit_now() { |
1605 | value = Some(Message::Quit) |
1606 | } |
1607 | match value { |
1608 | Some(Message::Work(work)) => { |
1609 | return Some(work); |
1610 | } |
1611 | Some(Message::Quit) => { |
1612 | // Repeat quit message to wake up sleeping threads, if |
1613 | // any. The domino effect will ensure that every thread |
1614 | // will quit. |
1615 | self.send_quit(); |
1616 | return None; |
1617 | } |
1618 | None => { |
1619 | // Once num_pending reaches 0, it is impossible for it to |
1620 | // ever increase again. Namely, it only reaches 0 once |
1621 | // all jobs have run such that no jobs have produced more |
1622 | // work. We have this guarantee because num_pending is |
1623 | // always incremented before each job is submitted and only |
1624 | // decremented once each job is completely finished. |
1625 | // Therefore, if this reaches zero, then there can be no |
1626 | // other job running. |
1627 | if self.num_pending() == 0 { |
1628 | // Every other thread is blocked at the next recv(). |
1629 | // Send the initial quit message and quit. |
1630 | self.send_quit(); |
1631 | return None; |
1632 | } |
1633 | // Wait for next `Work` or `Quit` message. |
1634 | loop { |
1635 | if let Some(v) = self.recv() { |
1636 | value = Some(v); |
1637 | break; |
1638 | } |
1639 | // Our stack isn't blocking. Instead of burning the |
1640 | // CPU waiting, we let the thread sleep for a bit. In |
1641 | // general, this tends to only occur once the search is |
1642 | // approaching termination. |
1643 | thread::sleep(Duration::from_millis(1)); |
1644 | } |
1645 | } |
1646 | } |
1647 | } |
1648 | } |
1649 | |
1650 | /// Indicates that all workers should quit immediately. |
1651 | fn quit_now(&self) { |
1652 | self.quit_now.store(true, Ordering::SeqCst); |
1653 | } |
1654 | |
1655 | /// Returns true if this worker should quit immediately. |
1656 | fn is_quit_now(&self) -> bool { |
1657 | self.quit_now.load(Ordering::SeqCst) |
1658 | } |
1659 | |
1660 | /// Returns the number of pending jobs. |
1661 | fn num_pending(&self) -> usize { |
1662 | self.num_pending.load(Ordering::SeqCst) |
1663 | } |
1664 | |
1665 | /// Send work. |
1666 | fn send(&self, work: Work) { |
1667 | self.num_pending.fetch_add(1, Ordering::SeqCst); |
1668 | let mut stack = self.stack.lock().unwrap(); |
1669 | stack.push(Message::Work(work)); |
1670 | } |
1671 | |
1672 | /// Send a quit message. |
1673 | fn send_quit(&self) { |
1674 | let mut stack = self.stack.lock().unwrap(); |
1675 | stack.push(Message::Quit); |
1676 | } |
1677 | |
1678 | /// Receive work. |
1679 | fn recv(&self) -> Option<Message> { |
1680 | let mut stack = self.stack.lock().unwrap(); |
1681 | stack.pop() |
1682 | } |
1683 | |
1684 | /// Signal that work has been received. |
1685 | fn work_done(&self) { |
1686 | self.num_pending.fetch_sub(1, Ordering::SeqCst); |
1687 | } |
1688 | } |
1689 | |
1690 | fn check_symlink_loop( |
1691 | ig_parent: &Ignore, |
1692 | child_path: &Path, |
1693 | child_depth: usize, |
1694 | ) -> Result<(), Error> { |
1695 | let hchild: Handle = Handle::from_path(child_path).map_err(|err: Error| { |
1696 | Error::from(err).with_path(child_path).with_depth(child_depth) |
1697 | })?; |
1698 | for ig: &Ignore in ig_parent.parents().take_while(|ig: &&Ignore| !ig.is_absolute_parent()) { |
1699 | let h: Handle = Handle::from_path(ig.path()).map_err(|err: Error| { |
1700 | Error::from(err).with_path(child_path).with_depth(child_depth) |
1701 | })?; |
1702 | if hchild == h { |
1703 | return Err(Error::Loop { |
1704 | ancestor: ig.path().to_path_buf(), |
1705 | child: child_path.to_path_buf(), |
1706 | } |
1707 | .with_depth(child_depth)); |
1708 | } |
1709 | } |
1710 | Ok(()) |
1711 | } |
1712 | |
1713 | // Before calling this function, make sure that you ensure that is really |
1714 | // necessary as the arguments imply a file stat. |
1715 | fn skip_filesize( |
1716 | max_filesize: u64, |
1717 | path: &Path, |
1718 | ent: &Option<Metadata>, |
1719 | ) -> bool { |
1720 | let filesize: Option = match *ent { |
1721 | Some(ref md: &Metadata) => Some(md.len()), |
1722 | None => None, |
1723 | }; |
1724 | |
1725 | if let Some(fs: u64) = filesize { |
1726 | if fs > max_filesize { |
1727 | log::debug!("ignoring {}: {} bytes" , path.display(), fs); |
1728 | true |
1729 | } else { |
1730 | false |
1731 | } |
1732 | } else { |
1733 | false |
1734 | } |
1735 | } |
1736 | |
1737 | fn should_skip_entry(ig: &Ignore, dent: &DirEntry) -> bool { |
1738 | let m: Match> = ig.matched_dir_entry(dent); |
1739 | if m.is_ignore() { |
1740 | log::debug!("ignoring {}: {:?}" , dent.path().display(), m); |
1741 | true |
1742 | } else if m.is_whitelist() { |
1743 | log::debug!("whitelisting {}: {:?}" , dent.path().display(), m); |
1744 | false |
1745 | } else { |
1746 | false |
1747 | } |
1748 | } |
1749 | |
1750 | /// Returns a handle to stdout for filtering search. |
1751 | /// |
1752 | /// A handle is returned if and only if stdout is being redirected to a file. |
1753 | /// The handle returned corresponds to that file. |
1754 | /// |
1755 | /// This can be used to ensure that we do not attempt to search a file that we |
1756 | /// may also be writing to. |
1757 | fn stdout_handle() -> Option<Handle> { |
1758 | let h: Handle = match Handle::stdout() { |
1759 | Err(_) => return None, |
1760 | Ok(h: Handle) => h, |
1761 | }; |
1762 | let md: Metadata = match h.as_file().metadata() { |
1763 | Err(_) => return None, |
1764 | Ok(md: Metadata) => md, |
1765 | }; |
1766 | if !md.is_file() { |
1767 | return None; |
1768 | } |
1769 | Some(h) |
1770 | } |
1771 | |
1772 | /// Returns true if and only if the given directory entry is believed to be |
1773 | /// equivalent to the given handle. If there was a problem querying the path |
1774 | /// for information to determine equality, then that error is returned. |
1775 | fn path_equals(dent: &DirEntry, handle: &Handle) -> Result<bool, Error> { |
1776 | #[cfg (unix)] |
1777 | fn never_equal(dent: &DirEntry, handle: &Handle) -> bool { |
1778 | dent.ino() != Some(handle.ino()) |
1779 | } |
1780 | |
1781 | #[cfg (not(unix))] |
1782 | fn never_equal(_: &DirEntry, _: &Handle) -> bool { |
1783 | false |
1784 | } |
1785 | |
1786 | // If we know for sure that these two things aren't equal, then avoid |
1787 | // the costly extra stat call to determine equality. |
1788 | if dent.is_stdin() || never_equal(dent, handle) { |
1789 | return Ok(false); |
1790 | } |
1791 | Handle::from_path(dent.path()) |
1792 | .map(|h| &h == handle) |
1793 | .map_err(|err: Error| Error::Io(err).with_path(dent.path())) |
1794 | } |
1795 | |
1796 | /// Returns true if the given walkdir entry corresponds to a directory. |
1797 | /// |
1798 | /// This is normally just `dent.file_type().is_dir()`, but when we aren't |
1799 | /// following symlinks, the root directory entry may be a symlink to a |
1800 | /// directory that we *do* follow---by virtue of it being specified by the user |
1801 | /// explicitly. In that case, we need to follow the symlink and query whether |
1802 | /// it's a directory or not. But we only do this for root entries to avoid an |
1803 | /// additional stat check in most cases. |
1804 | fn walkdir_is_dir(dent: &walkdir::DirEntry) -> bool { |
1805 | if dent.file_type().is_dir() { |
1806 | return true; |
1807 | } |
1808 | if !dent.file_type().is_symlink() || dent.depth() > 0 { |
1809 | return false; |
1810 | } |
1811 | dent.path().metadata().ok().map_or(default:false, |md: Metadata| md.file_type().is_dir()) |
1812 | } |
1813 | |
1814 | /// Returns true if and only if the given path is on the same device as the |
1815 | /// given root device. |
1816 | fn is_same_file_system(root_device: u64, path: &Path) -> Result<bool, Error> { |
1817 | let dent_device: u64 = |
1818 | device_num(path).map_err(|err: Error| Error::Io(err).with_path(path))?; |
1819 | Ok(root_device == dent_device) |
1820 | } |
1821 | |
1822 | #[cfg (unix)] |
1823 | fn device_num<P: AsRef<Path>>(path: P) -> io::Result<u64> { |
1824 | use std::os::unix::fs::MetadataExt; |
1825 | |
1826 | path.as_ref().metadata().map(|md: Metadata| md.dev()) |
1827 | } |
1828 | |
1829 | #[cfg (windows)] |
1830 | fn device_num<P: AsRef<Path>>(path: P) -> io::Result<u64> { |
1831 | use winapi_util::{file, Handle}; |
1832 | |
1833 | let h = Handle::from_path_any(path)?; |
1834 | file::information(h).map(|info| info.volume_serial_number()) |
1835 | } |
1836 | |
1837 | #[cfg (not(any(unix, windows)))] |
1838 | fn device_num<P: AsRef<Path>>(_: P) -> io::Result<u64> { |
1839 | Err(io::Error::new( |
1840 | io::ErrorKind::Other, |
1841 | "walkdir: same_file_system option not supported on this platform" , |
1842 | )) |
1843 | } |
1844 | |
1845 | #[cfg (test)] |
1846 | mod tests { |
1847 | use std::ffi::OsStr; |
1848 | use std::fs::{self, File}; |
1849 | use std::io::Write; |
1850 | use std::path::Path; |
1851 | use std::sync::{Arc, Mutex}; |
1852 | |
1853 | use super::{DirEntry, WalkBuilder, WalkState}; |
1854 | use crate::tests::TempDir; |
1855 | |
1856 | fn wfile<P: AsRef<Path>>(path: P, contents: &str) { |
1857 | let mut file = File::create(path).unwrap(); |
1858 | file.write_all(contents.as_bytes()).unwrap(); |
1859 | } |
1860 | |
1861 | fn wfile_size<P: AsRef<Path>>(path: P, size: u64) { |
1862 | let file = File::create(path).unwrap(); |
1863 | file.set_len(size).unwrap(); |
1864 | } |
1865 | |
1866 | #[cfg (unix)] |
1867 | fn symlink<P: AsRef<Path>, Q: AsRef<Path>>(src: P, dst: Q) { |
1868 | use std::os::unix::fs::symlink; |
1869 | symlink(src, dst).unwrap(); |
1870 | } |
1871 | |
1872 | fn mkdirp<P: AsRef<Path>>(path: P) { |
1873 | fs::create_dir_all(path).unwrap(); |
1874 | } |
1875 | |
1876 | fn normal_path(unix: &str) -> String { |
1877 | if cfg!(windows) { |
1878 | unix.replace(" \\" , "/" ) |
1879 | } else { |
1880 | unix.to_string() |
1881 | } |
1882 | } |
1883 | |
1884 | fn walk_collect(prefix: &Path, builder: &WalkBuilder) -> Vec<String> { |
1885 | let mut paths = vec![]; |
1886 | for result in builder.build() { |
1887 | let dent = match result { |
1888 | Err(_) => continue, |
1889 | Ok(dent) => dent, |
1890 | }; |
1891 | let path = dent.path().strip_prefix(prefix).unwrap(); |
1892 | if path.as_os_str().is_empty() { |
1893 | continue; |
1894 | } |
1895 | paths.push(normal_path(path.to_str().unwrap())); |
1896 | } |
1897 | paths.sort(); |
1898 | paths |
1899 | } |
1900 | |
1901 | fn walk_collect_parallel( |
1902 | prefix: &Path, |
1903 | builder: &WalkBuilder, |
1904 | ) -> Vec<String> { |
1905 | let mut paths = vec![]; |
1906 | for dent in walk_collect_entries_parallel(builder) { |
1907 | let path = dent.path().strip_prefix(prefix).unwrap(); |
1908 | if path.as_os_str().is_empty() { |
1909 | continue; |
1910 | } |
1911 | paths.push(normal_path(path.to_str().unwrap())); |
1912 | } |
1913 | paths.sort(); |
1914 | paths |
1915 | } |
1916 | |
1917 | fn walk_collect_entries_parallel(builder: &WalkBuilder) -> Vec<DirEntry> { |
1918 | let dents = Arc::new(Mutex::new(vec![])); |
1919 | builder.build_parallel().run(|| { |
1920 | let dents = dents.clone(); |
1921 | Box::new(move |result| { |
1922 | if let Ok(dent) = result { |
1923 | dents.lock().unwrap().push(dent); |
1924 | } |
1925 | WalkState::Continue |
1926 | }) |
1927 | }); |
1928 | |
1929 | let dents = dents.lock().unwrap(); |
1930 | dents.to_vec() |
1931 | } |
1932 | |
1933 | fn mkpaths(paths: &[&str]) -> Vec<String> { |
1934 | let mut paths: Vec<_> = paths.iter().map(|s| s.to_string()).collect(); |
1935 | paths.sort(); |
1936 | paths |
1937 | } |
1938 | |
1939 | fn tmpdir() -> TempDir { |
1940 | TempDir::new().unwrap() |
1941 | } |
1942 | |
1943 | fn assert_paths(prefix: &Path, builder: &WalkBuilder, expected: &[&str]) { |
1944 | let got = walk_collect(prefix, builder); |
1945 | assert_eq!(got, mkpaths(expected), "single threaded" ); |
1946 | let got = walk_collect_parallel(prefix, builder); |
1947 | assert_eq!(got, mkpaths(expected), "parallel" ); |
1948 | } |
1949 | |
1950 | #[test ] |
1951 | fn no_ignores() { |
1952 | let td = tmpdir(); |
1953 | mkdirp(td.path().join("a/b/c" )); |
1954 | mkdirp(td.path().join("x/y" )); |
1955 | wfile(td.path().join("a/b/foo" ), "" ); |
1956 | wfile(td.path().join("x/y/foo" ), "" ); |
1957 | |
1958 | assert_paths( |
1959 | td.path(), |
1960 | &WalkBuilder::new(td.path()), |
1961 | &["x" , "x/y" , "x/y/foo" , "a" , "a/b" , "a/b/foo" , "a/b/c" ], |
1962 | ); |
1963 | } |
1964 | |
1965 | #[test ] |
1966 | fn custom_ignore() { |
1967 | let td = tmpdir(); |
1968 | let custom_ignore = ".customignore" ; |
1969 | mkdirp(td.path().join("a" )); |
1970 | wfile(td.path().join(custom_ignore), "foo" ); |
1971 | wfile(td.path().join("foo" ), "" ); |
1972 | wfile(td.path().join("a/foo" ), "" ); |
1973 | wfile(td.path().join("bar" ), "" ); |
1974 | wfile(td.path().join("a/bar" ), "" ); |
1975 | |
1976 | let mut builder = WalkBuilder::new(td.path()); |
1977 | builder.add_custom_ignore_filename(&custom_ignore); |
1978 | assert_paths(td.path(), &builder, &["bar" , "a" , "a/bar" ]); |
1979 | } |
1980 | |
1981 | #[test ] |
1982 | fn custom_ignore_exclusive_use() { |
1983 | let td = tmpdir(); |
1984 | let custom_ignore = ".customignore" ; |
1985 | mkdirp(td.path().join("a" )); |
1986 | wfile(td.path().join(custom_ignore), "foo" ); |
1987 | wfile(td.path().join("foo" ), "" ); |
1988 | wfile(td.path().join("a/foo" ), "" ); |
1989 | wfile(td.path().join("bar" ), "" ); |
1990 | wfile(td.path().join("a/bar" ), "" ); |
1991 | |
1992 | let mut builder = WalkBuilder::new(td.path()); |
1993 | builder.ignore(false); |
1994 | builder.git_ignore(false); |
1995 | builder.git_global(false); |
1996 | builder.git_exclude(false); |
1997 | builder.add_custom_ignore_filename(&custom_ignore); |
1998 | assert_paths(td.path(), &builder, &["bar" , "a" , "a/bar" ]); |
1999 | } |
2000 | |
2001 | #[test ] |
2002 | fn gitignore() { |
2003 | let td = tmpdir(); |
2004 | mkdirp(td.path().join(".git" )); |
2005 | mkdirp(td.path().join("a" )); |
2006 | wfile(td.path().join(".gitignore" ), "foo" ); |
2007 | wfile(td.path().join("foo" ), "" ); |
2008 | wfile(td.path().join("a/foo" ), "" ); |
2009 | wfile(td.path().join("bar" ), "" ); |
2010 | wfile(td.path().join("a/bar" ), "" ); |
2011 | |
2012 | assert_paths( |
2013 | td.path(), |
2014 | &WalkBuilder::new(td.path()), |
2015 | &["bar" , "a" , "a/bar" ], |
2016 | ); |
2017 | } |
2018 | |
2019 | #[test ] |
2020 | fn explicit_ignore() { |
2021 | let td = tmpdir(); |
2022 | let igpath = td.path().join(".not-an-ignore" ); |
2023 | mkdirp(td.path().join("a" )); |
2024 | wfile(&igpath, "foo" ); |
2025 | wfile(td.path().join("foo" ), "" ); |
2026 | wfile(td.path().join("a/foo" ), "" ); |
2027 | wfile(td.path().join("bar" ), "" ); |
2028 | wfile(td.path().join("a/bar" ), "" ); |
2029 | |
2030 | let mut builder = WalkBuilder::new(td.path()); |
2031 | assert!(builder.add_ignore(&igpath).is_none()); |
2032 | assert_paths(td.path(), &builder, &["bar" , "a" , "a/bar" ]); |
2033 | } |
2034 | |
2035 | #[test ] |
2036 | fn explicit_ignore_exclusive_use() { |
2037 | let td = tmpdir(); |
2038 | let igpath = td.path().join(".not-an-ignore" ); |
2039 | mkdirp(td.path().join("a" )); |
2040 | wfile(&igpath, "foo" ); |
2041 | wfile(td.path().join("foo" ), "" ); |
2042 | wfile(td.path().join("a/foo" ), "" ); |
2043 | wfile(td.path().join("bar" ), "" ); |
2044 | wfile(td.path().join("a/bar" ), "" ); |
2045 | |
2046 | let mut builder = WalkBuilder::new(td.path()); |
2047 | builder.standard_filters(false); |
2048 | assert!(builder.add_ignore(&igpath).is_none()); |
2049 | assert_paths( |
2050 | td.path(), |
2051 | &builder, |
2052 | &[".not-an-ignore" , "bar" , "a" , "a/bar" ], |
2053 | ); |
2054 | } |
2055 | |
2056 | #[test ] |
2057 | fn gitignore_parent() { |
2058 | let td = tmpdir(); |
2059 | mkdirp(td.path().join(".git" )); |
2060 | mkdirp(td.path().join("a" )); |
2061 | wfile(td.path().join(".gitignore" ), "foo" ); |
2062 | wfile(td.path().join("a/foo" ), "" ); |
2063 | wfile(td.path().join("a/bar" ), "" ); |
2064 | |
2065 | let root = td.path().join("a" ); |
2066 | assert_paths(&root, &WalkBuilder::new(&root), &["bar" ]); |
2067 | } |
2068 | |
2069 | #[test ] |
2070 | fn max_depth() { |
2071 | let td = tmpdir(); |
2072 | mkdirp(td.path().join("a/b/c" )); |
2073 | wfile(td.path().join("foo" ), "" ); |
2074 | wfile(td.path().join("a/foo" ), "" ); |
2075 | wfile(td.path().join("a/b/foo" ), "" ); |
2076 | wfile(td.path().join("a/b/c/foo" ), "" ); |
2077 | |
2078 | let mut builder = WalkBuilder::new(td.path()); |
2079 | assert_paths( |
2080 | td.path(), |
2081 | &builder, |
2082 | &["a" , "a/b" , "a/b/c" , "foo" , "a/foo" , "a/b/foo" , "a/b/c/foo" ], |
2083 | ); |
2084 | assert_paths(td.path(), builder.max_depth(Some(0)), &[]); |
2085 | assert_paths(td.path(), builder.max_depth(Some(1)), &["a" , "foo" ]); |
2086 | assert_paths( |
2087 | td.path(), |
2088 | builder.max_depth(Some(2)), |
2089 | &["a" , "a/b" , "foo" , "a/foo" ], |
2090 | ); |
2091 | } |
2092 | |
2093 | #[test ] |
2094 | fn max_filesize() { |
2095 | let td = tmpdir(); |
2096 | mkdirp(td.path().join("a/b" )); |
2097 | wfile_size(td.path().join("foo" ), 0); |
2098 | wfile_size(td.path().join("bar" ), 400); |
2099 | wfile_size(td.path().join("baz" ), 600); |
2100 | wfile_size(td.path().join("a/foo" ), 600); |
2101 | wfile_size(td.path().join("a/bar" ), 500); |
2102 | wfile_size(td.path().join("a/baz" ), 200); |
2103 | |
2104 | let mut builder = WalkBuilder::new(td.path()); |
2105 | assert_paths( |
2106 | td.path(), |
2107 | &builder, |
2108 | &["a" , "a/b" , "foo" , "bar" , "baz" , "a/foo" , "a/bar" , "a/baz" ], |
2109 | ); |
2110 | assert_paths( |
2111 | td.path(), |
2112 | builder.max_filesize(Some(0)), |
2113 | &["a" , "a/b" , "foo" ], |
2114 | ); |
2115 | assert_paths( |
2116 | td.path(), |
2117 | builder.max_filesize(Some(500)), |
2118 | &["a" , "a/b" , "foo" , "bar" , "a/bar" , "a/baz" ], |
2119 | ); |
2120 | assert_paths( |
2121 | td.path(), |
2122 | builder.max_filesize(Some(50000)), |
2123 | &["a" , "a/b" , "foo" , "bar" , "baz" , "a/foo" , "a/bar" , "a/baz" ], |
2124 | ); |
2125 | } |
2126 | |
2127 | #[cfg (unix)] // because symlinks on windows are weird |
2128 | #[test ] |
2129 | fn symlinks() { |
2130 | let td = tmpdir(); |
2131 | mkdirp(td.path().join("a/b" )); |
2132 | symlink(td.path().join("a/b" ), td.path().join("z" )); |
2133 | wfile(td.path().join("a/b/foo" ), "" ); |
2134 | |
2135 | let mut builder = WalkBuilder::new(td.path()); |
2136 | assert_paths(td.path(), &builder, &["a" , "a/b" , "a/b/foo" , "z" ]); |
2137 | assert_paths( |
2138 | td.path(), |
2139 | &builder.follow_links(true), |
2140 | &["a" , "a/b" , "a/b/foo" , "z" , "z/foo" ], |
2141 | ); |
2142 | } |
2143 | |
2144 | #[cfg (unix)] // because symlinks on windows are weird |
2145 | #[test ] |
2146 | fn first_path_not_symlink() { |
2147 | let td = tmpdir(); |
2148 | mkdirp(td.path().join("foo" )); |
2149 | |
2150 | let dents = WalkBuilder::new(td.path().join("foo" )) |
2151 | .build() |
2152 | .into_iter() |
2153 | .collect::<Result<Vec<_>, _>>() |
2154 | .unwrap(); |
2155 | assert_eq!(1, dents.len()); |
2156 | assert!(!dents[0].path_is_symlink()); |
2157 | |
2158 | let dents = walk_collect_entries_parallel(&WalkBuilder::new( |
2159 | td.path().join("foo" ), |
2160 | )); |
2161 | assert_eq!(1, dents.len()); |
2162 | assert!(!dents[0].path_is_symlink()); |
2163 | } |
2164 | |
2165 | #[cfg (unix)] // because symlinks on windows are weird |
2166 | #[test ] |
2167 | fn symlink_loop() { |
2168 | let td = tmpdir(); |
2169 | mkdirp(td.path().join("a/b" )); |
2170 | symlink(td.path().join("a" ), td.path().join("a/b/c" )); |
2171 | |
2172 | let mut builder = WalkBuilder::new(td.path()); |
2173 | assert_paths(td.path(), &builder, &["a" , "a/b" , "a/b/c" ]); |
2174 | assert_paths(td.path(), &builder.follow_links(true), &["a" , "a/b" ]); |
2175 | } |
2176 | |
2177 | // It's a little tricky to test the 'same_file_system' option since |
2178 | // we need an environment with more than one file system. We adopt a |
2179 | // heuristic where /sys is typically a distinct volume on Linux and roll |
2180 | // with that. |
2181 | #[test ] |
2182 | #[cfg (target_os = "linux" )] |
2183 | fn same_file_system() { |
2184 | use super::device_num; |
2185 | |
2186 | // If for some reason /sys doesn't exist or isn't a directory, just |
2187 | // skip this test. |
2188 | if !Path::new("/sys" ).is_dir() { |
2189 | return; |
2190 | } |
2191 | |
2192 | // If our test directory actually isn't a different volume from /sys, |
2193 | // then this test is meaningless and we shouldn't run it. |
2194 | let td = tmpdir(); |
2195 | if device_num(td.path()).unwrap() == device_num("/sys" ).unwrap() { |
2196 | return; |
2197 | } |
2198 | |
2199 | mkdirp(td.path().join("same_file" )); |
2200 | symlink("/sys" , td.path().join("same_file" ).join("alink" )); |
2201 | |
2202 | // Create a symlink to sys and enable following symlinks. If the |
2203 | // same_file_system option doesn't work, then this probably will hit a |
2204 | // permission error. Otherwise, it should just skip over the symlink |
2205 | // completely. |
2206 | let mut builder = WalkBuilder::new(td.path()); |
2207 | builder.follow_links(true).same_file_system(true); |
2208 | assert_paths(td.path(), &builder, &["same_file" , "same_file/alink" ]); |
2209 | } |
2210 | |
2211 | #[cfg (target_os = "linux" )] |
2212 | #[test ] |
2213 | fn no_read_permissions() { |
2214 | let dir_path = Path::new("/root" ); |
2215 | |
2216 | // There's no /etc/sudoers.d, skip the test. |
2217 | if !dir_path.is_dir() { |
2218 | return; |
2219 | } |
2220 | // We're the root, so the test won't check what we want it to. |
2221 | if fs::read_dir(&dir_path).is_ok() { |
2222 | return; |
2223 | } |
2224 | |
2225 | // Check that we can't descend but get an entry for the parent dir. |
2226 | let builder = WalkBuilder::new(&dir_path); |
2227 | assert_paths(dir_path.parent().unwrap(), &builder, &["root" ]); |
2228 | } |
2229 | |
2230 | #[test ] |
2231 | fn filter() { |
2232 | let td = tmpdir(); |
2233 | mkdirp(td.path().join("a/b/c" )); |
2234 | mkdirp(td.path().join("x/y" )); |
2235 | wfile(td.path().join("a/b/foo" ), "" ); |
2236 | wfile(td.path().join("x/y/foo" ), "" ); |
2237 | |
2238 | assert_paths( |
2239 | td.path(), |
2240 | &WalkBuilder::new(td.path()), |
2241 | &["x" , "x/y" , "x/y/foo" , "a" , "a/b" , "a/b/foo" , "a/b/c" ], |
2242 | ); |
2243 | |
2244 | assert_paths( |
2245 | td.path(), |
2246 | &WalkBuilder::new(td.path()) |
2247 | .filter_entry(|entry| entry.file_name() != OsStr::new("a" )), |
2248 | &["x" , "x/y" , "x/y/foo" ], |
2249 | ); |
2250 | } |
2251 | } |
2252 | |