]> git.lizzy.rs Git - rust.git/blob - src/tools/linkchecker/main.rs
94e82e3d9f7664f993ea6fcb2db1e0a9a156d1df
[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             let kind = t!(entry.file_type());
186             if kind.is_dir() {
187                 self.walk(&path, report);
188             } else {
189                 self.check(&path, report);
190             }
191         }
192     }
193
194     /// Checks a single file.
195     fn check(&mut self, file: &Path, report: &mut Report) {
196         let (pretty_path, entry) = self.load_file(file, report);
197         let source = match entry {
198             FileEntry::Missing => panic!("missing file {:?} while walking", file),
199             FileEntry::Dir => unreachable!("never with `check` path"),
200             FileEntry::OtherFile => return,
201             FileEntry::Redirect { .. } => return,
202             FileEntry::HtmlFile { source, ids } => {
203                 parse_ids(&mut ids.borrow_mut(), &pretty_path, source, report);
204                 source.clone()
205             }
206         };
207
208         // Search for anything that's the regex 'href[ ]*=[ ]*".*?"'
209         with_attrs_in_source(&source, " href", |url, i, base| {
210             // Ignore external URLs
211             if url.starts_with("http:")
212                 || url.starts_with("https:")
213                 || url.starts_with("javascript:")
214                 || url.starts_with("ftp:")
215                 || url.starts_with("irc:")
216                 || url.starts_with("data:")
217             {
218                 report.links_ignored_external += 1;
219                 return;
220             }
221             report.links_checked += 1;
222             let (url, fragment) = match url.split_once('#') {
223                 None => (url, None),
224                 Some((url, fragment)) => (url, Some(fragment)),
225             };
226             // NB: the `splitn` always succeeds, even if the delimiter is not present.
227             let url = url.splitn(2, '?').next().unwrap();
228
229             // Once we've plucked out the URL, parse it using our base url and
230             // then try to extract a file path.
231             let mut path = file.to_path_buf();
232             if !base.is_empty() || !url.is_empty() {
233                 path.pop();
234                 for part in Path::new(base).join(url).components() {
235                     match part {
236                         Component::Prefix(_) | Component::RootDir => {
237                             // Avoid absolute paths as they make the docs not
238                             // relocatable by making assumptions on where the docs
239                             // are hosted relative to the site root.
240                             report.errors += 1;
241                             println!(
242                                 "{}:{}: absolute path - {}",
243                                 pretty_path,
244                                 i + 1,
245                                 Path::new(base).join(url).display()
246                             );
247                             return;
248                         }
249                         Component::CurDir => {}
250                         Component::ParentDir => {
251                             path.pop();
252                         }
253                         Component::Normal(s) => {
254                             path.push(s);
255                         }
256                     }
257                 }
258             }
259
260             let (target_pretty_path, target_entry) = self.load_file(&path, report);
261             let (target_source, target_ids) = match target_entry {
262                 FileEntry::Missing => {
263                     if is_exception(file, &target_pretty_path) {
264                         report.links_ignored_exception += 1;
265                     } else {
266                         report.errors += 1;
267                         println!(
268                             "{}:{}: broken link - `{}`",
269                             pretty_path,
270                             i + 1,
271                             target_pretty_path
272                         );
273                     }
274                     return;
275                 }
276                 FileEntry::Dir => {
277                     // Links to directories show as directory listings when viewing
278                     // the docs offline so it's best to avoid them.
279                     report.errors += 1;
280                     println!(
281                         "{}:{}: directory link to `{}` \
282                          (directory links should use index.html instead)",
283                         pretty_path,
284                         i + 1,
285                         target_pretty_path
286                     );
287                     return;
288                 }
289                 FileEntry::OtherFile => return,
290                 FileEntry::Redirect { target } => {
291                     let t = target.clone();
292                     drop(target);
293                     let (target, redir_entry) = self.load_file(&t, report);
294                     match redir_entry {
295                         FileEntry::Missing => {
296                             report.errors += 1;
297                             println!(
298                                 "{}:{}: broken redirect from `{}` to `{}`",
299                                 pretty_path,
300                                 i + 1,
301                                 target_pretty_path,
302                                 target
303                             );
304                             return;
305                         }
306                         FileEntry::Redirect { target } => {
307                             // Redirect to a redirect, this link checker
308                             // currently doesn't support this, since it would
309                             // require cycle checking, etc.
310                             report.errors += 1;
311                             println!(
312                                 "{}:{}: redirect from `{}` to `{}` \
313                                  which is also a redirect (not supported)",
314                                 pretty_path,
315                                 i + 1,
316                                 target_pretty_path,
317                                 target.display()
318                             );
319                             return;
320                         }
321                         FileEntry::Dir => {
322                             report.errors += 1;
323                             println!(
324                                 "{}:{}: redirect from `{}` to `{}` \
325                                  which is a directory \
326                                  (directory links should use index.html instead)",
327                                 pretty_path,
328                                 i + 1,
329                                 target_pretty_path,
330                                 target
331                             );
332                             return;
333                         }
334                         FileEntry::OtherFile => return,
335                         FileEntry::HtmlFile { source, ids } => (source, ids),
336                     }
337                 }
338                 FileEntry::HtmlFile { source, ids } => (source, ids),
339             };
340
341             // Alright, if we've found an HTML file for the target link. If
342             // this is a fragment link, also check that the `id` exists.
343             if let Some(ref fragment) = fragment {
344                 // Fragments like `#1-6` are most likely line numbers to be
345                 // interpreted by javascript, so we're ignoring these
346                 if fragment.splitn(2, '-').all(|f| f.chars().all(|c| c.is_numeric())) {
347                     return;
348                 }
349
350                 // These appear to be broken in mdbook right now?
351                 if fragment.starts_with('-') {
352                     return;
353                 }
354
355                 parse_ids(&mut target_ids.borrow_mut(), &pretty_path, target_source, report);
356
357                 if target_ids.borrow().contains(*fragment) {
358                     return;
359                 }
360
361                 if is_exception(file, &format!("#{}", fragment)) {
362                     report.links_ignored_exception += 1;
363                 } else {
364                     report.errors += 1;
365                     print!("{}:{}: broken link fragment ", pretty_path, i + 1);
366                     println!("`#{}` pointing to `{}`", fragment, target_pretty_path);
367                 };
368             }
369         });
370
371         // Search for intra-doc links that rustdoc didn't warn about
372         // FIXME(#77199, 77200) Rustdoc should just warn about these directly.
373         // NOTE: only looks at one line at a time; in practice this should find most links
374         for (i, line) in source.lines().enumerate() {
375             for broken_link in BROKEN_INTRA_DOC_LINK.captures_iter(line) {
376                 if is_intra_doc_exception(file, &broken_link[1]) {
377                     report.intra_doc_exceptions += 1;
378                 } else {
379                     report.errors += 1;
380                     print!("{}:{}: broken intra-doc link - ", pretty_path, i + 1);
381                     println!("{}", &broken_link[0]);
382                 }
383             }
384         }
385         // we don't need the source anymore,
386         // so drop to reduce memory-usage
387         match self.cache.get_mut(&pretty_path).unwrap() {
388             FileEntry::HtmlFile { source, .. } => *source = Rc::new(String::new()),
389             _ => unreachable!("must be html file"),
390         }
391     }
392
393     /// Load a file from disk, or from the cache if available.
394     fn load_file(&mut self, file: &Path, report: &mut Report) -> (String, &FileEntry) {
395         // https://docs.microsoft.com/en-us/windows/win32/debug/system-error-codes--0-499-
396         #[cfg(windows)]
397         const ERROR_INVALID_NAME: i32 = 123;
398
399         let pretty_path =
400             file.strip_prefix(&self.root).unwrap_or(&file).to_str().unwrap().to_string();
401
402         let entry =
403             self.cache.entry(pretty_path.clone()).or_insert_with(|| match fs::metadata(file) {
404                 Ok(metadata) if metadata.is_dir() => FileEntry::Dir,
405                 Ok(_) => {
406                     if file.extension().and_then(|s| s.to_str()) != Some("html") {
407                         FileEntry::OtherFile
408                     } else {
409                         report.html_files += 1;
410                         load_html_file(file, report)
411                     }
412                 }
413                 Err(e) if e.kind() == ErrorKind::NotFound => FileEntry::Missing,
414                 Err(e) => {
415                     // If a broken intra-doc link contains `::`, on windows, it will cause `ERROR_INVALID_NAME` rather than `NotFound`.
416                     // Explicitly check for that so that the broken link can be allowed in `LINKCHECK_EXCEPTIONS`.
417                     #[cfg(windows)]
418                     if e.raw_os_error() == Some(ERROR_INVALID_NAME)
419                         && file.as_os_str().to_str().map_or(false, |s| s.contains("::"))
420                     {
421                         return FileEntry::Missing;
422                     }
423                     panic!("unexpected read error for {}: {}", file.display(), e);
424                 }
425             });
426         (pretty_path, entry)
427     }
428 }
429
430 impl Report {
431     fn report(&self) {
432         println!("checked links in: {:.1}s", self.start.elapsed().as_secs_f64());
433         println!("number of HTML files scanned: {}", self.html_files);
434         println!("number of HTML redirects found: {}", self.html_redirects);
435         println!("number of links checked: {}", self.links_checked);
436         println!("number of links ignored due to external: {}", self.links_ignored_external);
437         println!("number of links ignored due to exceptions: {}", self.links_ignored_exception);
438         println!("number of intra doc links ignored: {}", self.intra_doc_exceptions);
439         println!("errors found: {}", self.errors);
440     }
441 }
442
443 fn load_html_file(file: &Path, report: &mut Report) -> FileEntry {
444     let source = match fs::read_to_string(file) {
445         Ok(s) => Rc::new(s),
446         Err(err) => {
447             // This usually should not fail since `metadata` was already
448             // called successfully on this file.
449             panic!("unexpected read error for {}: {}", file.display(), err);
450         }
451     };
452     match maybe_redirect(&source) {
453         Some(target) => {
454             report.html_redirects += 1;
455             let target = file.parent().unwrap().join(target);
456             FileEntry::Redirect { target }
457         }
458         None => FileEntry::HtmlFile { source: source.clone(), ids: RefCell::new(HashSet::new()) },
459     }
460 }
461
462 fn is_intra_doc_exception(file: &Path, link: &str) -> bool {
463     if let Some(entry) = INTRA_DOC_LINK_EXCEPTIONS.iter().find(|&(f, _)| file.ends_with(f)) {
464         entry.1.is_empty() || entry.1.contains(&link)
465     } else {
466         false
467     }
468 }
469
470 fn is_exception(file: &Path, link: &str) -> bool {
471     if let Some(entry) = LINKCHECK_EXCEPTIONS.iter().find(|&(f, _)| file.ends_with(f)) {
472         entry.1.contains(&link)
473     } else {
474         // FIXME(#63351): Concat trait in alloc/slice reexported in primitive page
475         //
476         // NOTE: This cannot be added to `LINKCHECK_EXCEPTIONS` because the resolved path
477         // calculated in `check` function is outside `build/<triple>/doc` dir.
478         // So the `strip_prefix` method just returns the old absolute broken path.
479         if file.ends_with("std/primitive.slice.html") {
480             if link.ends_with("primitive.slice.html") {
481                 return true;
482             }
483         }
484         false
485     }
486 }
487
488 /// If the given HTML file contents is an HTML redirect, this returns the
489 /// destination path given in the redirect.
490 fn maybe_redirect(source: &str) -> Option<String> {
491     const REDIRECT: &str = "<p>Redirecting to <a href=";
492
493     let mut lines = source.lines();
494     let redirect_line = lines.nth(7)?;
495
496     redirect_line.find(REDIRECT).map(|i| {
497         let rest = &redirect_line[(i + REDIRECT.len() + 1)..];
498         let pos_quote = rest.find('"').unwrap();
499         rest[..pos_quote].to_owned()
500     })
501 }
502
503 fn with_attrs_in_source<F: FnMut(&str, usize, &str)>(source: &str, attr: &str, mut f: F) {
504     let mut base = "";
505     for (i, mut line) in source.lines().enumerate() {
506         while let Some(j) = line.find(attr) {
507             let rest = &line[j + attr.len()..];
508             // The base tag should always be the first link in the document so
509             // we can get away with using one pass.
510             let is_base = line[..j].ends_with("<base");
511             line = rest;
512             let pos_equals = match rest.find('=') {
513                 Some(i) => i,
514                 None => continue,
515             };
516             if rest[..pos_equals].trim_start_matches(' ') != "" {
517                 continue;
518             }
519
520             let rest = &rest[pos_equals + 1..];
521
522             let pos_quote = match rest.find(&['"', '\''][..]) {
523                 Some(i) => i,
524                 None => continue,
525             };
526             let quote_delim = rest.as_bytes()[pos_quote] as char;
527
528             if rest[..pos_quote].trim_start_matches(' ') != "" {
529                 continue;
530             }
531             let rest = &rest[pos_quote + 1..];
532             let url = match rest.find(quote_delim) {
533                 Some(i) => &rest[..i],
534                 None => continue,
535             };
536             if is_base {
537                 base = url;
538                 continue;
539             }
540             f(url, i, base)
541         }
542     }
543 }
544
545 fn parse_ids(ids: &mut HashSet<String>, file: &str, source: &str, report: &mut Report) {
546     if ids.is_empty() {
547         with_attrs_in_source(source, " id", |fragment, i, _| {
548             let frag = fragment.trim_start_matches("#").to_owned();
549             let encoded = small_url_encode(&frag);
550             if !ids.insert(frag) {
551                 report.errors += 1;
552                 println!("{}:{}: id is not unique: `{}`", file, i, fragment);
553             }
554             // Just in case, we also add the encoded id.
555             ids.insert(encoded);
556         });
557     }
558 }