1 //! Script to check the validity of `href` links in our HTML documentation.
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.
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.
14 //! A few exceptions are allowed as there's known bugs in rustdoc, but this
15 //! should catch the majority of "broken link" cases.
17 use std::cell::RefCell;
18 use std::collections::{HashMap, HashSet};
21 use std::io::ErrorKind;
22 use std::path::{Component, Path, PathBuf};
24 use std::time::Instant;
26 use once_cell::sync::Lazy;
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])]
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"]),
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"]),
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", &[
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>_",
71 ("reference/tokens.html", &[
76 "0</code>-<code>9</code> <code>a</code>-<code>f</code> <code>A</code>-<code>F",
78 ("reference/notation.html", &[
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"]),
91 static BROKEN_INTRA_DOC_LINK: Lazy<Regex> =
92 Lazy::new(|| Regex::new(r#"\[<code>(.*)</code>\]"#).unwrap());
98 Err(e) => panic!("{} failed with {:?}", stringify!($e), e),
104 let docs = env::args_os().nth(1).expect("doc path should be first argument");
105 let docs = env::current_dir().unwrap().join(docs);
106 let mut checker = Checker { root: docs.clone(), cache: HashMap::new() };
107 let mut report = Report {
109 start: Instant::now(),
113 links_ignored_external: 0,
114 links_ignored_exception: 0,
115 intra_doc_exceptions: 0,
117 checker.walk(&docs, &mut report);
119 if report.errors != 0 {
120 println!("found some broken links");
121 std::process::exit(1);
136 links_ignored_external: u32,
137 links_ignored_exception: u32,
138 intra_doc_exceptions: u32,
145 /// This includes the contents of the HTML file, and an optional set of
146 /// HTML IDs. The IDs are used for checking fragments. They are computed
147 /// as-needed. The source is discarded (replaced with an empty string)
148 /// after the file has been checked, to conserve on memory.
149 HtmlFile { source: Rc<String>, ids: RefCell<HashSet<String>> },
150 /// This file is an HTML redirect to the given local path.
151 Redirect { target: PathBuf },
152 /// This is not an HTML file.
154 /// This is a directory.
156 /// The file doesn't exist.
160 /// A cache to speed up file access.
161 type Cache = HashMap<String, FileEntry>;
163 fn small_url_encode(s: &str) -> String {
164 s.replace("<", "%3C")
175 .replace("\"", "%22")
179 /// Primary entry point for walking the filesystem to find HTML files to check.
180 fn walk(&mut self, dir: &Path, report: &mut Report) {
181 for entry in t!(dir.read_dir()).map(|e| t!(e)) {
182 let path = entry.path();
183 let kind = t!(entry.file_type());
185 self.walk(&path, report);
187 self.check(&path, report);
192 /// Checks a single file.
193 fn check(&mut self, file: &Path, report: &mut Report) {
194 let (pretty_path, entry) = self.load_file(file, report);
195 let source = match entry {
196 FileEntry::Missing => panic!("missing file {:?} while walking", file),
197 FileEntry::Dir => unreachable!("never with `check` path"),
198 FileEntry::OtherFile => return,
199 FileEntry::Redirect { .. } => return,
200 FileEntry::HtmlFile { source, ids } => {
201 parse_ids(&mut ids.borrow_mut(), &pretty_path, source, report);
206 // Search for anything that's the regex 'href[ ]*=[ ]*".*?"'
207 with_attrs_in_source(&source, " href", |url, i, base| {
208 // Ignore external URLs
209 if url.starts_with("http:")
210 || url.starts_with("https:")
211 || url.starts_with("javascript:")
212 || url.starts_with("ftp:")
213 || url.starts_with("irc:")
214 || url.starts_with("data:")
216 report.links_ignored_external += 1;
219 report.links_checked += 1;
220 let (url, fragment) = match url.split_once('#') {
222 Some((url, fragment)) => (url, Some(fragment)),
224 // NB: the `splitn` always succeeds, even if the delimiter is not present.
225 let url = url.splitn(2, '?').next().unwrap();
227 // Once we've plucked out the URL, parse it using our base url and
228 // then try to extract a file path.
229 let mut path = file.to_path_buf();
230 if !base.is_empty() || !url.is_empty() {
232 for part in Path::new(base).join(url).components() {
234 Component::Prefix(_) | Component::RootDir => {
235 // Avoid absolute paths as they make the docs not
236 // relocatable by making assumptions on where the docs
237 // are hosted relative to the site root.
240 "{}:{}: absolute path - {}",
243 Path::new(base).join(url).display()
247 Component::CurDir => {}
248 Component::ParentDir => {
251 Component::Normal(s) => {
258 let (target_pretty_path, target_entry) = self.load_file(&path, report);
259 let (target_source, target_ids) = match target_entry {
260 FileEntry::Missing => {
261 if is_exception(file, &target_pretty_path) {
262 report.links_ignored_exception += 1;
266 "{}:{}: broken link - `{}`",
275 // Links to directories show as directory listings when viewing
276 // the docs offline so it's best to avoid them.
279 "{}:{}: directory link to `{}` \
280 (directory links should use index.html instead)",
287 FileEntry::OtherFile => return,
288 FileEntry::Redirect { target } => {
289 let t = target.clone();
291 let (target, redir_entry) = self.load_file(&t, report);
293 FileEntry::Missing => {
296 "{}:{}: broken redirect from `{}` to `{}`",
304 FileEntry::Redirect { target } => {
305 // Redirect to a redirect, this link checker
306 // currently doesn't support this, since it would
307 // require cycle checking, etc.
310 "{}:{}: redirect from `{}` to `{}` \
311 which is also a redirect (not supported)",
322 "{}:{}: redirect from `{}` to `{}` \
323 which is a directory \
324 (directory links should use index.html instead)",
332 FileEntry::OtherFile => return,
333 FileEntry::HtmlFile { source, ids } => (source, ids),
336 FileEntry::HtmlFile { source, ids } => (source, ids),
339 // Alright, if we've found an HTML file for the target link. If
340 // this is a fragment link, also check that the `id` exists.
341 if let Some(ref fragment) = fragment {
342 // Fragments like `#1-6` are most likely line numbers to be
343 // interpreted by javascript, so we're ignoring these
344 if fragment.splitn(2, '-').all(|f| f.chars().all(|c| c.is_numeric())) {
348 // These appear to be broken in mdbook right now?
349 if fragment.starts_with('-') {
353 parse_ids(&mut target_ids.borrow_mut(), &pretty_path, target_source, report);
355 if target_ids.borrow().contains(*fragment) {
359 if is_exception(file, &format!("#{}", fragment)) {
360 report.links_ignored_exception += 1;
363 print!("{}:{}: broken link fragment ", pretty_path, i + 1);
364 println!("`#{}` pointing to `{}`", fragment, target_pretty_path);
369 // Search for intra-doc links that rustdoc didn't warn about
370 // FIXME(#77199, 77200) Rustdoc should just warn about these directly.
371 // NOTE: only looks at one line at a time; in practice this should find most links
372 for (i, line) in source.lines().enumerate() {
373 for broken_link in BROKEN_INTRA_DOC_LINK.captures_iter(line) {
374 if is_intra_doc_exception(file, &broken_link[1]) {
375 report.intra_doc_exceptions += 1;
378 print!("{}:{}: broken intra-doc link - ", pretty_path, i + 1);
379 println!("{}", &broken_link[0]);
383 // we don't need the source anymore,
384 // so drop to reduce memory-usage
385 match self.cache.get_mut(&pretty_path).unwrap() {
386 FileEntry::HtmlFile { source, .. } => *source = Rc::new(String::new()),
387 _ => unreachable!("must be html file"),
391 /// Load a file from disk, or from the cache if available.
392 fn load_file(&mut self, file: &Path, report: &mut Report) -> (String, &FileEntry) {
393 // https://docs.microsoft.com/en-us/windows/win32/debug/system-error-codes--0-499-
395 const ERROR_INVALID_NAME: i32 = 123;
398 file.strip_prefix(&self.root).unwrap_or(&file).to_str().unwrap().to_string();
401 self.cache.entry(pretty_path.clone()).or_insert_with(|| match fs::metadata(file) {
402 Ok(metadata) if metadata.is_dir() => FileEntry::Dir,
404 if file.extension().and_then(|s| s.to_str()) != Some("html") {
407 report.html_files += 1;
408 load_html_file(file, report)
411 Err(e) if e.kind() == ErrorKind::NotFound => FileEntry::Missing,
413 // If a broken intra-doc link contains `::`, on windows, it will cause `ERROR_INVALID_NAME` rather than `NotFound`.
414 // Explicitly check for that so that the broken link can be allowed in `LINKCHECK_EXCEPTIONS`.
416 if e.raw_os_error() == Some(ERROR_INVALID_NAME)
417 && file.as_os_str().to_str().map_or(false, |s| s.contains("::"))
419 return FileEntry::Missing;
421 panic!("unexpected read error for {}: {}", file.display(), e);
430 println!("checked links in: {:.1}s", self.start.elapsed().as_secs_f64());
431 println!("number of HTML files scanned: {}", self.html_files);
432 println!("number of HTML redirects found: {}", self.html_redirects);
433 println!("number of links checked: {}", self.links_checked);
434 println!("number of links ignored due to external: {}", self.links_ignored_external);
435 println!("number of links ignored due to exceptions: {}", self.links_ignored_exception);
436 println!("number of intra doc links ignored: {}", self.intra_doc_exceptions);
437 println!("errors found: {}", self.errors);
441 fn load_html_file(file: &Path, report: &mut Report) -> FileEntry {
442 let source = match fs::read_to_string(file) {
445 // This usually should not fail since `metadata` was already
446 // called successfully on this file.
447 panic!("unexpected read error for {}: {}", file.display(), err);
450 match maybe_redirect(&source) {
452 report.html_redirects += 1;
453 let target = file.parent().unwrap().join(target);
454 FileEntry::Redirect { target }
456 None => FileEntry::HtmlFile { source: source.clone(), ids: RefCell::new(HashSet::new()) },
460 fn is_intra_doc_exception(file: &Path, link: &str) -> bool {
461 if let Some(entry) = INTRA_DOC_LINK_EXCEPTIONS.iter().find(|&(f, _)| file.ends_with(f)) {
462 entry.1.is_empty() || entry.1.contains(&link)
468 fn is_exception(file: &Path, link: &str) -> bool {
469 if let Some(entry) = LINKCHECK_EXCEPTIONS.iter().find(|&(f, _)| file.ends_with(f)) {
470 entry.1.contains(&link)
472 // FIXME(#63351): Concat trait in alloc/slice reexported in primitive page
474 // NOTE: This cannot be added to `LINKCHECK_EXCEPTIONS` because the resolved path
475 // calculated in `check` function is outside `build/<triple>/doc` dir.
476 // So the `strip_prefix` method just returns the old absolute broken path.
477 if file.ends_with("std/primitive.slice.html") {
478 if link.ends_with("primitive.slice.html") {
486 /// If the given HTML file contents is an HTML redirect, this returns the
487 /// destination path given in the redirect.
488 fn maybe_redirect(source: &str) -> Option<String> {
489 const REDIRECT: &str = "<p>Redirecting to <a href=";
491 let mut lines = source.lines();
492 let redirect_line = lines.nth(7)?;
494 redirect_line.find(REDIRECT).map(|i| {
495 let rest = &redirect_line[(i + REDIRECT.len() + 1)..];
496 let pos_quote = rest.find('"').unwrap();
497 rest[..pos_quote].to_owned()
501 fn with_attrs_in_source<F: FnMut(&str, usize, &str)>(source: &str, attr: &str, mut f: F) {
503 for (i, mut line) in source.lines().enumerate() {
504 while let Some(j) = line.find(attr) {
505 let rest = &line[j + attr.len()..];
506 // The base tag should always be the first link in the document so
507 // we can get away with using one pass.
508 let is_base = line[..j].ends_with("<base");
510 let pos_equals = match rest.find('=') {
514 if rest[..pos_equals].trim_start_matches(' ') != "" {
518 let rest = &rest[pos_equals + 1..];
520 let pos_quote = match rest.find(&['"', '\''][..]) {
524 let quote_delim = rest.as_bytes()[pos_quote] as char;
526 if rest[..pos_quote].trim_start_matches(' ') != "" {
529 let rest = &rest[pos_quote + 1..];
530 let url = match rest.find(quote_delim) {
531 Some(i) => &rest[..i],
543 fn parse_ids(ids: &mut HashSet<String>, file: &str, source: &str, report: &mut Report) {
545 with_attrs_in_source(source, " id", |fragment, i, _| {
546 let frag = fragment.trim_start_matches("#").to_owned();
547 let encoded = small_url_encode(&frag);
548 if !ids.insert(frag) {
550 println!("{}:{}: id is not unique: `{}`", file, i, fragment);
552 // Just in case, we also add the encoded id.