]> git.lizzy.rs Git - rust.git/blob - src/tools/linkchecker/main.rs
Rollup merge of #98200 - ouz-a:issue-98177, r=oli-obk
[rust.git] / src / tools / linkchecker / main.rs
1 //! Script to check the validity of `href` links in our HTML documentation.
2 //!
3 //! In the past we've been quite error prone to writing in broken links as most
4 //! of them are manually rather than automatically added. As files move over
5 //! time or apis change old links become stale or broken. The purpose of this
6 //! script is to check all relative links in our documentation to make sure they
7 //! actually point to a valid place.
8 //!
9 //! Currently this doesn't actually do any HTML parsing or anything fancy like
10 //! that, it just has a simple "regex" to search for `href` and `id` tags.
11 //! These values are then translated to file URLs if possible and then the
12 //! destination is asserted to exist.
13 //!
14 //! A few exceptions are allowed as there's known bugs in rustdoc, but this
15 //! should catch the majority of "broken link" cases.
16
17 use std::cell::RefCell;
18 use std::collections::{HashMap, HashSet};
19 use std::env;
20 use std::fs;
21 use std::io::ErrorKind;
22 use std::path::{Component, Path, PathBuf};
23 use std::rc::Rc;
24 use std::time::Instant;
25
26 use once_cell::sync::Lazy;
27 use regex::Regex;
28
29 // Add linkcheck exceptions here
30 // If at all possible you should use intra-doc links to avoid linkcheck issues. These
31 // are cases where that does not work
32 // [(generated_documentation_page, &[broken_links])]
33 #[rustfmt::skip]
34 const LINKCHECK_EXCEPTIONS: &[(&str, &[&str])] = &[
35     // These try to link to std::collections, but are defined in alloc
36     // https://github.com/rust-lang/rust/issues/74481
37     ("std/collections/btree_map/struct.BTreeMap.html", &["#insert-and-complex-keys"]),
38     ("std/collections/btree_set/struct.BTreeSet.html", &["#insert-and-complex-keys"]),
39     ("alloc/collections/btree_map/struct.BTreeMap.html", &["#insert-and-complex-keys"]),
40     ("alloc/collections/btree_set/struct.BTreeSet.html", &["#insert-and-complex-keys"]),
41
42     // These try to link to various things in std, but are defined in core.
43     // The docs in std::primitive use proper intra-doc links, so these seem fine to special-case.
44     // Most these are broken because liballoc uses `#[lang_item]` magic to define things on
45     // primitives that aren't available in core.
46     ("alloc/slice/trait.Join.html", &["#method.join"]),
47     ("alloc/slice/trait.Concat.html", &["#method.concat"]),
48     ("alloc/slice/index.html", &["#method.concat", "#method.join"]),
49     ("alloc/vec/struct.Vec.html", &["#method.sort_by_key", "#method.sort_by_cached_key"]),
50     ("core/primitive.str.html", &["#method.to_ascii_uppercase", "#method.to_ascii_lowercase"]),
51     ("core/primitive.slice.html", &["#method.to_ascii_uppercase", "#method.to_ascii_lowercase",
52                                     "core/slice::sort_by_key", "core\\slice::sort_by_key",
53                                     "#method.sort_by_cached_key"]),
54 ];
55
56 #[rustfmt::skip]
57 const INTRA_DOC_LINK_EXCEPTIONS: &[(&str, &[&str])] = &[
58     // This will never have links that are not in other pages.
59     // To avoid repeating the exceptions twice, an empty list means all broken links are allowed.
60     ("reference/print.html", &[]),
61     // All the reference 'links' are actually ENBF highlighted as code
62     ("reference/comments.html", &[
63          "/</code> <code>!",
64          "*</code> <code>!",
65     ]),
66     ("reference/identifiers.html", &[
67          "a</code>-<code>z</code> <code>A</code>-<code>Z",
68          "a</code>-<code>z</code> <code>A</code>-<code>Z</code> <code>0</code>-<code>9</code> <code>_",
69          "a</code>-<code>z</code> <code>A</code>-<code>Z</code>] [<code>a</code>-<code>z</code> <code>A</code>-<code>Z</code> <code>0</code>-<code>9</code> <code>_",
70     ]),
71     ("reference/tokens.html", &[
72          "0</code>-<code>1",
73          "0</code>-<code>7",
74          "0</code>-<code>9",
75          "0</code>-<code>9",
76          "0</code>-<code>9</code> <code>a</code>-<code>f</code> <code>A</code>-<code>F",
77     ]),
78     ("reference/notation.html", &[
79          "b</code> <code>B",
80          "a</code>-<code>z",
81     ]),
82     // This is being used in the sense of 'inclusive range', not a markdown link
83     ("core/ops/struct.RangeInclusive.html", &["begin</code>, <code>end"]),
84     ("std/ops/struct.RangeInclusive.html", &["begin</code>, <code>end"]),
85     ("core/slice/trait.SliceIndex.html", &["begin</code>, <code>end"]),
86     ("alloc/slice/trait.SliceIndex.html", &["begin</code>, <code>end"]),
87     ("std/slice/trait.SliceIndex.html", &["begin</code>, <code>end"]),
88     ("core/primitive.str.html", &["begin</code>, <code>end"]),
89     ("std/primitive.str.html", &["begin</code>, <code>end"]),
90
91 ];
92
93 static BROKEN_INTRA_DOC_LINK: Lazy<Regex> =
94     Lazy::new(|| Regex::new(r#"\[<code>(.*)</code>\]"#).unwrap());
95
96 macro_rules! t {
97     ($e:expr) => {
98         match $e {
99             Ok(e) => e,
100             Err(e) => panic!("{} failed with {:?}", stringify!($e), e),
101         }
102     };
103 }
104
105 fn main() {
106     let docs = env::args_os().nth(1).expect("doc path should be first argument");
107     let docs = env::current_dir().unwrap().join(docs);
108     let mut checker = Checker { root: docs.clone(), cache: HashMap::new() };
109     let mut report = Report {
110         errors: 0,
111         start: Instant::now(),
112         html_files: 0,
113         html_redirects: 0,
114         links_checked: 0,
115         links_ignored_external: 0,
116         links_ignored_exception: 0,
117         intra_doc_exceptions: 0,
118     };
119     checker.walk(&docs, &mut report);
120     report.report();
121     if report.errors != 0 {
122         println!("found some broken links");
123         std::process::exit(1);
124     }
125 }
126
127 struct Checker {
128     root: PathBuf,
129     cache: Cache,
130 }
131
132 struct Report {
133     errors: u32,
134     start: Instant,
135     html_files: u32,
136     html_redirects: u32,
137     links_checked: u32,
138     links_ignored_external: u32,
139     links_ignored_exception: u32,
140     intra_doc_exceptions: u32,
141 }
142
143 /// A cache entry.
144 enum FileEntry {
145     /// An HTML file.
146     ///
147     /// This includes the contents of the HTML file, and an optional set of
148     /// HTML IDs. The IDs are used for checking fragments. They are computed
149     /// as-needed. The source is discarded (replaced with an empty string)
150     /// after the file has been checked, to conserve on memory.
151     HtmlFile { source: Rc<String>, ids: RefCell<HashSet<String>> },
152     /// This file is an HTML redirect to the given local path.
153     Redirect { target: PathBuf },
154     /// This is not an HTML file.
155     OtherFile,
156     /// This is a directory.
157     Dir,
158     /// The file doesn't exist.
159     Missing,
160 }
161
162 /// A cache to speed up file access.
163 type Cache = HashMap<String, FileEntry>;
164
165 fn small_url_encode(s: &str) -> String {
166     s.replace("<", "%3C")
167         .replace(">", "%3E")
168         .replace(" ", "%20")
169         .replace("?", "%3F")
170         .replace("'", "%27")
171         .replace("&", "%26")
172         .replace(",", "%2C")
173         .replace(":", "%3A")
174         .replace(";", "%3B")
175         .replace("[", "%5B")
176         .replace("]", "%5D")
177         .replace("\"", "%22")
178 }
179
180 impl Checker {
181     /// Primary entry point for walking the filesystem to find HTML files to check.
182     fn walk(&mut self, dir: &Path, report: &mut Report) {
183         for entry in t!(dir.read_dir()).map(|e| t!(e)) {
184             let path = entry.path();
185             // Goes through symlinks
186             let metadata = t!(fs::metadata(&path));
187             if metadata.is_dir() {
188                 self.walk(&path, report);
189             } else {
190                 self.check(&path, report);
191             }
192         }
193     }
194
195     /// Checks a single file.
196     fn check(&mut self, file: &Path, report: &mut Report) {
197         let (pretty_path, entry) = self.load_file(file, report);
198         let source = match entry {
199             FileEntry::Missing => panic!("missing file {:?} while walking", file),
200             FileEntry::Dir => unreachable!("never with `check` path"),
201             FileEntry::OtherFile => return,
202             FileEntry::Redirect { .. } => return,
203             FileEntry::HtmlFile { source, ids } => {
204                 parse_ids(&mut ids.borrow_mut(), &pretty_path, source, report);
205                 source.clone()
206             }
207         };
208
209         // Search for anything that's the regex 'href[ ]*=[ ]*".*?"'
210         with_attrs_in_source(&source, " href", |url, i, base| {
211             // Ignore external URLs
212             if url.starts_with("http:")
213                 || url.starts_with("https:")
214                 || url.starts_with("javascript:")
215                 || url.starts_with("ftp:")
216                 || url.starts_with("irc:")
217                 || url.starts_with("data:")
218             {
219                 report.links_ignored_external += 1;
220                 return;
221             }
222             report.links_checked += 1;
223             let (url, fragment) = match url.split_once('#') {
224                 None => (url, None),
225                 Some((url, fragment)) => (url, Some(fragment)),
226             };
227             // NB: the `splitn` always succeeds, even if the delimiter is not present.
228             let url = url.splitn(2, '?').next().unwrap();
229
230             // Once we've plucked out the URL, parse it using our base url and
231             // then try to extract a file path.
232             let mut path = file.to_path_buf();
233             if !base.is_empty() || !url.is_empty() {
234                 path.pop();
235                 for part in Path::new(base).join(url).components() {
236                     match part {
237                         Component::Prefix(_) | Component::RootDir => {
238                             // Avoid absolute paths as they make the docs not
239                             // relocatable by making assumptions on where the docs
240                             // are hosted relative to the site root.
241                             report.errors += 1;
242                             println!(
243                                 "{}:{}: absolute path - {}",
244                                 pretty_path,
245                                 i + 1,
246                                 Path::new(base).join(url).display()
247                             );
248                             return;
249                         }
250                         Component::CurDir => {}
251                         Component::ParentDir => {
252                             path.pop();
253                         }
254                         Component::Normal(s) => {
255                             path.push(s);
256                         }
257                     }
258                 }
259             }
260
261             let (target_pretty_path, target_entry) = self.load_file(&path, report);
262             let (target_source, target_ids) = match target_entry {
263                 FileEntry::Missing => {
264                     if is_exception(file, &target_pretty_path) {
265                         report.links_ignored_exception += 1;
266                     } else {
267                         report.errors += 1;
268                         println!(
269                             "{}:{}: broken link - `{}`",
270                             pretty_path,
271                             i + 1,
272                             target_pretty_path
273                         );
274                     }
275                     return;
276                 }
277                 FileEntry::Dir => {
278                     // Links to directories show as directory listings when viewing
279                     // the docs offline so it's best to avoid them.
280                     report.errors += 1;
281                     println!(
282                         "{}:{}: directory link to `{}` \
283                          (directory links should use index.html instead)",
284                         pretty_path,
285                         i + 1,
286                         target_pretty_path
287                     );
288                     return;
289                 }
290                 FileEntry::OtherFile => return,
291                 FileEntry::Redirect { target } => {
292                     let t = target.clone();
293                     drop(target);
294                     let (target, redir_entry) = self.load_file(&t, report);
295                     match redir_entry {
296                         FileEntry::Missing => {
297                             report.errors += 1;
298                             println!(
299                                 "{}:{}: broken redirect from `{}` to `{}`",
300                                 pretty_path,
301                                 i + 1,
302                                 target_pretty_path,
303                                 target
304                             );
305                             return;
306                         }
307                         FileEntry::Redirect { target } => {
308                             // Redirect to a redirect, this link checker
309                             // currently doesn't support this, since it would
310                             // require cycle checking, etc.
311                             report.errors += 1;
312                             println!(
313                                 "{}:{}: redirect from `{}` to `{}` \
314                                  which is also a redirect (not supported)",
315                                 pretty_path,
316                                 i + 1,
317                                 target_pretty_path,
318                                 target.display()
319                             );
320                             return;
321                         }
322                         FileEntry::Dir => {
323                             report.errors += 1;
324                             println!(
325                                 "{}:{}: redirect from `{}` to `{}` \
326                                  which is a directory \
327                                  (directory links should use index.html instead)",
328                                 pretty_path,
329                                 i + 1,
330                                 target_pretty_path,
331                                 target
332                             );
333                             return;
334                         }
335                         FileEntry::OtherFile => return,
336                         FileEntry::HtmlFile { source, ids } => (source, ids),
337                     }
338                 }
339                 FileEntry::HtmlFile { source, ids } => (source, ids),
340             };
341
342             // Alright, if we've found an HTML file for the target link. If
343             // this is a fragment link, also check that the `id` exists.
344             if let Some(ref fragment) = fragment {
345                 // Fragments like `#1-6` are most likely line numbers to be
346                 // interpreted by javascript, so we're ignoring these
347                 if fragment.splitn(2, '-').all(|f| f.chars().all(|c| c.is_numeric())) {
348                     return;
349                 }
350
351                 parse_ids(&mut target_ids.borrow_mut(), &pretty_path, target_source, report);
352
353                 if target_ids.borrow().contains(*fragment) {
354                     return;
355                 }
356
357                 if is_exception(file, &format!("#{}", fragment)) {
358                     report.links_ignored_exception += 1;
359                 } else {
360                     report.errors += 1;
361                     print!("{}:{}: broken link fragment ", pretty_path, i + 1);
362                     println!("`#{}` pointing to `{}`", fragment, target_pretty_path);
363                 };
364             }
365         });
366
367         // Search for intra-doc links that rustdoc didn't warn about
368         // FIXME(#77199, 77200) Rustdoc should just warn about these directly.
369         // NOTE: only looks at one line at a time; in practice this should find most links
370         for (i, line) in source.lines().enumerate() {
371             for broken_link in BROKEN_INTRA_DOC_LINK.captures_iter(line) {
372                 if is_intra_doc_exception(file, &broken_link[1]) {
373                     report.intra_doc_exceptions += 1;
374                 } else {
375                     report.errors += 1;
376                     print!("{}:{}: broken intra-doc link - ", pretty_path, i + 1);
377                     println!("{}", &broken_link[0]);
378                 }
379             }
380         }
381         // we don't need the source anymore,
382         // so drop to reduce memory-usage
383         match self.cache.get_mut(&pretty_path).unwrap() {
384             FileEntry::HtmlFile { source, .. } => *source = Rc::new(String::new()),
385             _ => unreachable!("must be html file"),
386         }
387     }
388
389     /// Load a file from disk, or from the cache if available.
390     fn load_file(&mut self, file: &Path, report: &mut Report) -> (String, &FileEntry) {
391         // https://docs.microsoft.com/en-us/windows/win32/debug/system-error-codes--0-499-
392         #[cfg(windows)]
393         const ERROR_INVALID_NAME: i32 = 123;
394
395         let pretty_path =
396             file.strip_prefix(&self.root).unwrap_or(&file).to_str().unwrap().to_string();
397
398         let entry =
399             self.cache.entry(pretty_path.clone()).or_insert_with(|| match fs::metadata(file) {
400                 Ok(metadata) if metadata.is_dir() => FileEntry::Dir,
401                 Ok(_) => {
402                     if file.extension().and_then(|s| s.to_str()) != Some("html") {
403                         FileEntry::OtherFile
404                     } else {
405                         report.html_files += 1;
406                         load_html_file(file, report)
407                     }
408                 }
409                 Err(e) if e.kind() == ErrorKind::NotFound => FileEntry::Missing,
410                 Err(e) => {
411                     // If a broken intra-doc link contains `::`, on windows, it will cause `ERROR_INVALID_NAME` rather than `NotFound`.
412                     // Explicitly check for that so that the broken link can be allowed in `LINKCHECK_EXCEPTIONS`.
413                     #[cfg(windows)]
414                     if e.raw_os_error() == Some(ERROR_INVALID_NAME)
415                         && file.as_os_str().to_str().map_or(false, |s| s.contains("::"))
416                     {
417                         return FileEntry::Missing;
418                     }
419                     panic!("unexpected read error for {}: {}", file.display(), e);
420                 }
421             });
422         (pretty_path, entry)
423     }
424 }
425
426 impl Report {
427     fn report(&self) {
428         println!("checked links in: {:.1}s", self.start.elapsed().as_secs_f64());
429         println!("number of HTML files scanned: {}", self.html_files);
430         println!("number of HTML redirects found: {}", self.html_redirects);
431         println!("number of links checked: {}", self.links_checked);
432         println!("number of links ignored due to external: {}", self.links_ignored_external);
433         println!("number of links ignored due to exceptions: {}", self.links_ignored_exception);
434         println!("number of intra doc links ignored: {}", self.intra_doc_exceptions);
435         println!("errors found: {}", self.errors);
436     }
437 }
438
439 fn load_html_file(file: &Path, report: &mut Report) -> FileEntry {
440     let source = match fs::read_to_string(file) {
441         Ok(s) => Rc::new(s),
442         Err(err) => {
443             // This usually should not fail since `metadata` was already
444             // called successfully on this file.
445             panic!("unexpected read error for {}: {}", file.display(), err);
446         }
447     };
448     match maybe_redirect(&source) {
449         Some(target) => {
450             report.html_redirects += 1;
451             let target = file.parent().unwrap().join(target);
452             FileEntry::Redirect { target }
453         }
454         None => FileEntry::HtmlFile { source: source.clone(), ids: RefCell::new(HashSet::new()) },
455     }
456 }
457
458 fn is_intra_doc_exception(file: &Path, link: &str) -> bool {
459     if let Some(entry) = INTRA_DOC_LINK_EXCEPTIONS.iter().find(|&(f, _)| file.ends_with(f)) {
460         entry.1.is_empty() || entry.1.contains(&link)
461     } else {
462         false
463     }
464 }
465
466 fn is_exception(file: &Path, link: &str) -> bool {
467     if let Some(entry) = LINKCHECK_EXCEPTIONS.iter().find(|&(f, _)| file.ends_with(f)) {
468         entry.1.contains(&link)
469     } else {
470         // FIXME(#63351): Concat trait in alloc/slice reexported in primitive page
471         //
472         // NOTE: This cannot be added to `LINKCHECK_EXCEPTIONS` because the resolved path
473         // calculated in `check` function is outside `build/<triple>/doc` dir.
474         // So the `strip_prefix` method just returns the old absolute broken path.
475         if file.ends_with("std/primitive.slice.html") {
476             if link.ends_with("primitive.slice.html") {
477                 return true;
478             }
479         }
480         false
481     }
482 }
483
484 /// If the given HTML file contents is an HTML redirect, this returns the
485 /// destination path given in the redirect.
486 fn maybe_redirect(source: &str) -> Option<String> {
487     const REDIRECT_RUSTDOC: (usize, &str) = (7, "<p>Redirecting to <a href=");
488     const REDIRECT_MDBOOK: (usize, &str) = (8 - 7, "<p>Redirecting to... <a href=");
489
490     let mut lines = source.lines();
491
492     let mut find_redirect = |(line_rel, redirect_pattern): (usize, &str)| {
493         let redirect_line = lines.nth(line_rel)?;
494
495         redirect_line.find(redirect_pattern).map(|i| {
496             let rest = &redirect_line[(i + redirect_pattern.len() + 1)..];
497             let pos_quote = rest.find('"').unwrap();
498             rest[..pos_quote].to_owned()
499         })
500     };
501
502     find_redirect(REDIRECT_RUSTDOC).or_else(|| find_redirect(REDIRECT_MDBOOK))
503 }
504
505 fn with_attrs_in_source<F: FnMut(&str, usize, &str)>(source: &str, attr: &str, mut f: F) {
506     let mut base = "";
507     for (i, mut line) in source.lines().enumerate() {
508         while let Some(j) = line.find(attr) {
509             let rest = &line[j + attr.len()..];
510             // The base tag should always be the first link in the document so
511             // we can get away with using one pass.
512             let is_base = line[..j].ends_with("<base");
513             line = rest;
514             let pos_equals = match rest.find('=') {
515                 Some(i) => i,
516                 None => continue,
517             };
518             if rest[..pos_equals].trim_start_matches(' ') != "" {
519                 continue;
520             }
521
522             let rest = &rest[pos_equals + 1..];
523
524             let pos_quote = match rest.find(&['"', '\''][..]) {
525                 Some(i) => i,
526                 None => continue,
527             };
528             let quote_delim = rest.as_bytes()[pos_quote] as char;
529
530             if rest[..pos_quote].trim_start_matches(' ') != "" {
531                 continue;
532             }
533             let rest = &rest[pos_quote + 1..];
534             let url = match rest.find(quote_delim) {
535                 Some(i) => &rest[..i],
536                 None => continue,
537             };
538             if is_base {
539                 base = url;
540                 continue;
541             }
542             f(url, i, base)
543         }
544     }
545 }
546
547 fn parse_ids(ids: &mut HashSet<String>, file: &str, source: &str, report: &mut Report) {
548     if ids.is_empty() {
549         with_attrs_in_source(source, " id", |fragment, i, _| {
550             let frag = fragment.trim_start_matches("#").to_owned();
551             let encoded = small_url_encode(&frag);
552             if !ids.insert(frag) {
553                 report.errors += 1;
554                 println!("{}:{}: id is not unique: `{}`", file, i, fragment);
555             }
556             // Just in case, we also add the encoded id.
557             ids.insert(encoded);
558         });
559     }
560 }