1 //! Markdown formatting for rustdoc.
3 //! This module implements markdown formatting through the pulldown-cmark library.
6 //! #![feature(rustc_private)]
8 //! extern crate rustc_span;
10 //! use rustc_span::edition::Edition;
11 //! use rustdoc::html::markdown::{HeadingOffset, IdMap, Markdown, ErrorCodes};
13 //! let s = "My *markdown* _text_";
14 //! let mut id_map = IdMap::new();
15 //! let md = Markdown {
19 //! error_codes: ErrorCodes::Yes,
20 //! edition: Edition::Edition2015,
21 //! playground: &None,
22 //! heading_offset: HeadingOffset::H2,
24 //! let html = md.into_string();
25 //! // ... something using html
28 use rustc_data_structures::fx::FxHashMap;
29 use rustc_hir::def_id::DefId;
30 use rustc_middle::ty::TyCtxt;
31 use rustc_span::edition::Edition;
32 use rustc_span::{Span, Symbol};
34 use once_cell::sync::Lazy;
36 use std::cell::RefCell;
37 use std::collections::VecDeque;
38 use std::default::Default;
40 use std::ops::{ControlFlow, Range};
43 use crate::clean::RenderedLink;
45 use crate::html::escape::Escape;
46 use crate::html::format::Buffer;
47 use crate::html::highlight;
48 use crate::html::length_limit::HtmlWithLimit;
49 use crate::html::toc::TocBuilder;
52 html, BrokenLink, CodeBlockKind, CowStr, Event, LinkType, Options, Parser, Tag,
58 const MAX_HEADER_LEVEL: u32 = 6;
60 /// Options for rendering Markdown in the main body of documentation.
61 pub(crate) fn main_body_opts() -> Options {
62 Options::ENABLE_TABLES
63 | Options::ENABLE_FOOTNOTES
64 | Options::ENABLE_STRIKETHROUGH
65 | Options::ENABLE_TASKLISTS
66 | Options::ENABLE_SMART_PUNCTUATION
69 /// Options for rendering Markdown in summaries (e.g., in search results).
70 pub(crate) fn summary_opts() -> Options {
71 Options::ENABLE_TABLES
72 | Options::ENABLE_FOOTNOTES
73 | Options::ENABLE_STRIKETHROUGH
74 | Options::ENABLE_TASKLISTS
75 | Options::ENABLE_SMART_PUNCTUATION
78 #[derive(Debug, Clone, Copy)]
79 pub enum HeadingOffset {
88 /// When `to_string` is called, this struct will emit the HTML corresponding to
89 /// the rendered version of the contained markdown string.
90 pub struct Markdown<'a> {
92 /// A list of link replacements.
93 pub links: &'a [RenderedLink],
94 /// The current list of used header IDs.
95 pub ids: &'a mut IdMap,
96 /// Whether to allow the use of explicit error codes in doctest lang strings.
97 pub error_codes: ErrorCodes,
98 /// Default edition to use when parsing doctests (to add a `fn main`).
100 pub playground: &'a Option<Playground>,
101 /// Offset at which we render headings.
102 /// E.g. if `heading_offset: HeadingOffset::H2`, then `# something` renders an `<h2>`.
103 pub heading_offset: HeadingOffset,
105 /// A tuple struct like `Markdown` that renders the markdown with a table of contents.
106 pub(crate) struct MarkdownWithToc<'a>(
108 pub(crate) &'a mut IdMap,
109 pub(crate) ErrorCodes,
111 pub(crate) &'a Option<Playground>,
113 /// A tuple struct like `Markdown` that renders the markdown escaping HTML tags
114 /// and includes no paragraph tags.
115 pub(crate) struct MarkdownItemInfo<'a>(pub(crate) &'a str, pub(crate) &'a mut IdMap);
116 /// A tuple struct like `Markdown` that renders only the first paragraph.
117 pub(crate) struct MarkdownSummaryLine<'a>(pub &'a str, pub &'a [RenderedLink]);
119 #[derive(Copy, Clone, PartialEq, Debug)]
120 pub enum ErrorCodes {
126 pub(crate) fn from(b: bool) -> Self {
128 true => ErrorCodes::Yes,
129 false => ErrorCodes::No,
133 pub(crate) fn as_bool(self) -> bool {
135 ErrorCodes::Yes => true,
136 ErrorCodes::No => false,
141 /// Controls whether a line will be hidden or shown in HTML output.
143 /// All lines are used in documentation tests.
150 fn for_html(self) -> Option<Cow<'a, str>> {
152 Line::Shown(l) => Some(l),
153 Line::Hidden(_) => None,
157 fn for_code(self) -> Cow<'a, str> {
160 Line::Hidden(l) => Cow::Borrowed(l),
165 // FIXME: There is a minor inconsistency here. For lines that start with ##, we
166 // have no easy way of removing a potential single space after the hashes, which
167 // is done in the single # case. This inconsistency seems okay, if non-ideal. In
168 // order to fix it we'd have to iterate to find the first non-# character, and
169 // then reallocate to remove it; which would make us return a String.
170 fn map_line(s: &str) -> Line<'_> {
171 let trimmed = s.trim();
172 if trimmed.starts_with("##") {
173 Line::Shown(Cow::Owned(s.replacen("##", "#", 1)))
174 } else if let Some(stripped) = trimmed.strip_prefix("# ") {
176 Line::Hidden(stripped)
177 } else if trimmed == "#" {
178 // We cannot handle '#text' because it could be #[attr].
181 Line::Shown(Cow::Borrowed(s))
185 /// Convert chars from a title for an id.
187 /// "Hello, world!" -> "hello-world"
188 fn slugify(c: char) -> Option<char> {
189 if c.is_alphanumeric() || c == '-' || c == '_' {
190 if c.is_ascii() { Some(c.to_ascii_lowercase()) } else { Some(c) }
191 } else if c.is_whitespace() && c.is_ascii() {
198 #[derive(Clone, Debug)]
199 pub struct Playground {
200 pub crate_name: Option<Symbol>,
204 /// Adds syntax highlighting and playground Run buttons to Rust code blocks.
205 struct CodeBlocks<'p, 'a, I: Iterator<Item = Event<'a>>> {
207 check_error_codes: ErrorCodes,
209 // Information about the playground if a URL has been specified, containing an
210 // optional crate name and the URL.
211 playground: &'p Option<Playground>,
214 impl<'p, 'a, I: Iterator<Item = Event<'a>>> CodeBlocks<'p, 'a, I> {
217 error_codes: ErrorCodes,
219 playground: &'p Option<Playground>,
221 CodeBlocks { inner: iter, check_error_codes: error_codes, edition, playground }
225 impl<'a, I: Iterator<Item = Event<'a>>> Iterator for CodeBlocks<'_, 'a, I> {
226 type Item = Event<'a>;
228 fn next(&mut self) -> Option<Self::Item> {
229 let event = self.inner.next();
234 let Some(Event::Start(Tag::CodeBlock(kind))) = event else {
238 let mut original_text = String::new();
239 for event in &mut self.inner {
241 Event::End(Tag::CodeBlock(..)) => break,
242 Event::Text(ref s) => {
243 original_text.push_str(s);
249 let parse_result = match kind {
250 CodeBlockKind::Fenced(ref lang) => {
252 LangString::parse_without_check(lang, self.check_error_codes, false);
253 if !parse_result.rust {
254 return Some(Event::Html(
256 "<div class=\"example-wrap\">\
257 <pre class=\"language-{}\"><code>{}</code></pre>\
260 Escape(&original_text),
267 CodeBlockKind::Indented => Default::default(),
270 let lines = original_text.lines().filter_map(|l| map_line(l).for_html());
271 let text = lines.intersperse("\n".into()).collect::<String>();
273 compile_fail = parse_result.compile_fail;
274 should_panic = parse_result.should_panic;
275 ignore = parse_result.ignore;
276 edition = parse_result.edition;
278 let explicit_edition = edition.is_some();
279 let edition = edition.unwrap_or(self.edition);
281 let playground_button = self.playground.as_ref().and_then(|playground| {
282 let krate = &playground.crate_name;
283 let url = &playground.url;
287 let test = original_text
289 .map(|l| map_line(l).for_code())
290 .intersperse("\n".into())
291 .collect::<String>();
292 let krate = krate.as_ref().map(|s| s.as_str());
294 doctest::make_test(&test, krate, false, &Default::default(), edition, None);
295 let channel = if test.contains("#![feature(") { "&version=nightly" } else { "" };
297 // These characters don't need to be escaped in a URI.
298 // See https://url.spec.whatwg.org/#query-percent-encode-set
299 // and https://url.spec.whatwg.org/#urlencoded-parsing
300 // and https://url.spec.whatwg.org/#url-code-points
301 fn dont_escape(c: u8) -> bool {
302 (b'a' <= c && c <= b'z')
303 || (b'A' <= c && c <= b'Z')
304 || (b'0' <= c && c <= b'9')
319 // As described in urlencoded-parsing, the
320 // first `=` is the one that separates key from
321 // value. Following `=`s are part of the value.
324 let mut test_escaped = String::new();
325 for b in test.bytes() {
327 test_escaped.push(char::from(b));
328 } else if b == b' ' {
329 // URL queries are decoded with + replaced with SP
330 test_escaped.push('+');
331 } else if b == b'%' {
332 test_escaped.push('%');
333 test_escaped.push('%');
335 write!(test_escaped, "%{:02X}", b).unwrap();
339 r#"<a class="test-arrow" target="_blank" href="{}?code={}{}&edition={}">Run</a>"#,
340 url, test_escaped, channel, edition,
344 let tooltip = if ignore != Ignore::None {
345 highlight::Tooltip::Ignore
346 } else if compile_fail {
347 highlight::Tooltip::CompileFail
348 } else if should_panic {
349 highlight::Tooltip::ShouldPanic
350 } else if explicit_edition {
351 highlight::Tooltip::Edition(edition)
353 highlight::Tooltip::None
356 // insert newline to clearly separate it from the
357 // previous block so we can shorten the html output
358 let mut s = Buffer::new();
361 highlight::render_example_with_highlighting(
365 playground_button.as_deref(),
367 Some(Event::Html(s.into_inner().into()))
371 /// Make headings links with anchor IDs and build up TOC.
372 struct LinkReplacer<'a, I: Iterator<Item = Event<'a>>> {
374 links: &'a [RenderedLink],
375 shortcut_link: Option<&'a RenderedLink>,
378 impl<'a, I: Iterator<Item = Event<'a>>> LinkReplacer<'a, I> {
379 fn new(iter: I, links: &'a [RenderedLink]) -> Self {
380 LinkReplacer { inner: iter, links, shortcut_link: None }
384 impl<'a, I: Iterator<Item = Event<'a>>> Iterator for LinkReplacer<'a, I> {
385 type Item = Event<'a>;
387 fn next(&mut self) -> Option<Self::Item> {
388 let mut event = self.inner.next();
390 // Replace intra-doc links and remove disambiguators from shortcut links (`[fn@f]`).
392 // This is a shortcut link that was resolved by the broken_link_callback: `[fn@f]`
393 // Remove any disambiguator.
394 Some(Event::Start(Tag::Link(
395 // [fn@f] or [fn@f][]
396 LinkType::ShortcutUnknown | LinkType::CollapsedUnknown,
400 debug!("saw start of shortcut link to {} with title {}", dest, title);
401 // If this is a shortcut link, it was resolved by the broken_link_callback.
402 // So the URL will already be updated properly.
403 let link = self.links.iter().find(|&link| *link.href == **dest);
404 // Since this is an external iterator, we can't replace the inner text just yet.
405 // Store that we saw a link so we know to replace it later.
406 if let Some(link) = link {
407 trace!("it matched");
408 assert!(self.shortcut_link.is_none(), "shortcut links cannot be nested");
409 self.shortcut_link = Some(link);
412 // Now that we're done with the shortcut link, don't replace any more text.
413 Some(Event::End(Tag::Link(
414 LinkType::ShortcutUnknown | LinkType::CollapsedUnknown,
418 debug!("saw end of shortcut link to {}", dest);
419 if self.links.iter().any(|link| *link.href == **dest) {
420 assert!(self.shortcut_link.is_some(), "saw closing link without opening tag");
421 self.shortcut_link = None;
424 // Handle backticks in inline code blocks, but only if we're in the middle of a shortcut link.
426 Some(Event::Code(text)) => {
427 trace!("saw code {}", text);
428 if let Some(link) = self.shortcut_link {
429 trace!("original text was {}", link.original_text);
430 // NOTE: this only replaces if the code block is the *entire* text.
431 // If only part of the link has code highlighting, the disambiguator will not be removed.
433 // This is a limitation from `collect_intra_doc_links`: it passes a full link,
434 // and does not distinguish at all between code blocks.
435 // So we could never be sure we weren't replacing too much:
436 // [fn@my_`f`unc] is treated the same as [my_func()] in that pass.
438 // NOTE: &[1..len() - 1] is to strip the backticks
439 if **text == link.original_text[1..link.original_text.len() - 1] {
440 debug!("replacing {} with {}", text, link.new_text);
441 *text = CowStr::Borrowed(&link.new_text);
445 // Replace plain text in links, but only in the middle of a shortcut link.
447 Some(Event::Text(text)) => {
448 trace!("saw text {}", text);
449 if let Some(link) = self.shortcut_link {
450 trace!("original text was {}", link.original_text);
451 // NOTE: same limitations as `Event::Code`
452 if **text == *link.original_text {
453 debug!("replacing {} with {}", text, link.new_text);
454 *text = CowStr::Borrowed(&link.new_text);
458 // If this is a link, but not a shortcut link,
459 // replace the URL, since the broken_link_callback was not called.
460 Some(Event::Start(Tag::Link(_, dest, _))) => {
461 if let Some(link) = self.links.iter().find(|&link| *link.original_text == **dest) {
462 *dest = CowStr::Borrowed(link.href.as_ref());
465 // Anything else couldn't have been a valid Rust path, so no need to replace the text.
469 // Yield the modified event
474 /// Wrap HTML tables into `<div>` to prevent having the doc blocks width being too big.
475 struct TableWrapper<'a, I: Iterator<Item = Event<'a>>> {
477 stored_events: VecDeque<Event<'a>>,
480 impl<'a, I: Iterator<Item = Event<'a>>> TableWrapper<'a, I> {
481 fn new(iter: I) -> Self {
482 Self { inner: iter, stored_events: VecDeque::new() }
486 impl<'a, I: Iterator<Item = Event<'a>>> Iterator for TableWrapper<'a, I> {
487 type Item = Event<'a>;
489 fn next(&mut self) -> Option<Self::Item> {
490 if let Some(first) = self.stored_events.pop_front() {
494 let event = self.inner.next()?;
497 Event::Start(Tag::Table(t)) => {
498 self.stored_events.push_back(Event::Start(Tag::Table(t)));
499 Event::Html(CowStr::Borrowed("<div>"))
501 Event::End(Tag::Table(t)) => {
502 self.stored_events.push_back(Event::Html(CowStr::Borrowed("</div>")));
503 Event::End(Tag::Table(t))
510 type SpannedEvent<'a> = (Event<'a>, Range<usize>);
512 /// Make headings links with anchor IDs and build up TOC.
513 struct HeadingLinks<'a, 'b, 'ids, I> {
515 toc: Option<&'b mut TocBuilder>,
516 buf: VecDeque<SpannedEvent<'a>>,
517 id_map: &'ids mut IdMap,
518 heading_offset: HeadingOffset,
521 impl<'a, 'b, 'ids, I> HeadingLinks<'a, 'b, 'ids, I> {
524 toc: Option<&'b mut TocBuilder>,
525 ids: &'ids mut IdMap,
526 heading_offset: HeadingOffset,
528 HeadingLinks { inner: iter, toc, buf: VecDeque::new(), id_map: ids, heading_offset }
532 impl<'a, 'b, 'ids, I: Iterator<Item = SpannedEvent<'a>>> Iterator
533 for HeadingLinks<'a, 'b, 'ids, I>
535 type Item = SpannedEvent<'a>;
537 fn next(&mut self) -> Option<Self::Item> {
538 if let Some(e) = self.buf.pop_front() {
542 let event = self.inner.next();
543 if let Some((Event::Start(Tag::Heading(level, _, _)), _)) = event {
544 let mut id = String::new();
545 for event in &mut self.inner {
547 Event::End(Tag::Heading(..)) => break,
548 Event::Start(Tag::Link(_, _, _)) | Event::End(Tag::Link(..)) => {}
549 Event::Text(text) | Event::Code(text) => {
550 id.extend(text.chars().filter_map(slugify));
551 self.buf.push_back(event);
553 _ => self.buf.push_back(event),
556 let id = self.id_map.derive(id);
558 if let Some(ref mut builder) = self.toc {
559 let mut html_header = String::new();
560 html::push_html(&mut html_header, self.buf.iter().map(|(ev, _)| ev.clone()));
561 let sec = builder.push(level as u32, html_header, id.clone());
562 self.buf.push_front((Event::Html(format!("{} ", sec).into()), 0..0));
566 std::cmp::min(level as u32 + (self.heading_offset as u32), MAX_HEADER_LEVEL);
567 self.buf.push_back((Event::Html(format!("</a></h{}>", level).into()), 0..0));
569 let start_tags = format!(
570 "<h{level} id=\"{id}\">\
575 return Some((Event::Html(start_tags.into()), 0..0));
581 /// Extracts just the first paragraph.
582 struct SummaryLine<'a, I: Iterator<Item = Event<'a>>> {
589 impl<'a, I: Iterator<Item = Event<'a>>> SummaryLine<'a, I> {
590 fn new(iter: I) -> Self {
591 SummaryLine { inner: iter, started: false, depth: 0, skipped_tags: 0 }
595 fn check_if_allowed_tag(t: &Tag<'_>) -> bool {
598 Tag::Paragraph | Tag::Item | Tag::Emphasis | Tag::Strong | Tag::Link(..) | Tag::BlockQuote
602 fn is_forbidden_tag(t: &Tag<'_>) -> bool {
603 matches!(t, Tag::CodeBlock(_) | Tag::Table(_) | Tag::TableHead | Tag::TableRow | Tag::TableCell)
606 impl<'a, I: Iterator<Item = Event<'a>>> Iterator for SummaryLine<'a, I> {
607 type Item = Event<'a>;
609 fn next(&mut self) -> Option<Self::Item> {
610 if self.started && self.depth == 0 {
616 if let Some(event) = self.inner.next() {
617 let mut is_start = true;
618 let is_allowed_tag = match event {
619 Event::Start(ref c) => {
620 if is_forbidden_tag(c) {
621 self.skipped_tags += 1;
625 check_if_allowed_tag(c)
627 Event::End(ref c) => {
628 if is_forbidden_tag(c) {
629 self.skipped_tags += 1;
634 check_if_allowed_tag(c)
639 self.skipped_tags += 1;
641 return if !is_allowed_tag {
643 Some(Event::Start(Tag::Paragraph))
645 Some(Event::End(Tag::Paragraph))
655 /// Moves all footnote definitions to the end and add back links to the
657 struct Footnotes<'a, I> {
659 footnotes: FxHashMap<String, (Vec<Event<'a>>, u16)>,
662 impl<'a, I> Footnotes<'a, I> {
663 fn new(iter: I) -> Self {
664 Footnotes { inner: iter, footnotes: FxHashMap::default() }
667 fn get_entry(&mut self, key: &str) -> &mut (Vec<Event<'a>>, u16) {
668 let new_id = self.footnotes.len() + 1;
669 let key = key.to_owned();
670 self.footnotes.entry(key).or_insert((Vec::new(), new_id as u16))
674 impl<'a, I: Iterator<Item = SpannedEvent<'a>>> Iterator for Footnotes<'a, I> {
675 type Item = SpannedEvent<'a>;
677 fn next(&mut self) -> Option<Self::Item> {
679 match self.inner.next() {
680 Some((Event::FootnoteReference(ref reference), range)) => {
681 let entry = self.get_entry(reference);
682 let reference = format!(
683 "<sup id=\"fnref{0}\"><a href=\"#fn{0}\">{0}</a></sup>",
686 return Some((Event::Html(reference.into()), range));
688 Some((Event::Start(Tag::FootnoteDefinition(def)), _)) => {
689 let mut content = Vec::new();
690 for (event, _) in &mut self.inner {
691 if let Event::End(Tag::FootnoteDefinition(..)) = event {
696 let entry = self.get_entry(&def);
697 (*entry).0 = content;
699 Some(e) => return Some(e),
701 if !self.footnotes.is_empty() {
702 let mut v: Vec<_> = self.footnotes.drain().map(|(_, x)| x).collect();
703 v.sort_by(|a, b| a.1.cmp(&b.1));
704 let mut ret = String::from("<div class=\"footnotes\"><hr><ol>");
705 for (mut content, id) in v {
706 write!(ret, "<li id=\"fn{}\">", id).unwrap();
707 let mut is_paragraph = false;
708 if let Some(&Event::End(Tag::Paragraph)) = content.last() {
712 html::push_html(&mut ret, content.into_iter());
713 write!(ret, " <a href=\"#fnref{}\">↩</a>", id).unwrap();
715 ret.push_str("</p>");
717 ret.push_str("</li>");
719 ret.push_str("</ol></div>");
720 return Some((Event::Html(ret.into()), 0..0));
730 pub(crate) fn find_testable_code<T: doctest::Tester>(
733 error_codes: ErrorCodes,
734 enable_per_target_ignores: bool,
735 extra_info: Option<&ExtraInfo<'_>>,
737 let mut parser = Parser::new(doc).into_offset_iter();
738 let mut prev_offset = 0;
739 let mut nb_lines = 0;
740 let mut register_header = None;
741 while let Some((event, offset)) = parser.next() {
743 Event::Start(Tag::CodeBlock(kind)) => {
744 let block_info = match kind {
745 CodeBlockKind::Fenced(ref lang) => {
752 enable_per_target_ignores,
757 CodeBlockKind::Indented => Default::default(),
759 if !block_info.rust {
763 let mut test_s = String::new();
765 while let Some((Event::Text(s), _)) = parser.next() {
770 .map(|l| map_line(l).for_code())
771 .collect::<Vec<Cow<'_, str>>>()
774 nb_lines += doc[prev_offset..offset.start].lines().count();
775 // If there are characters between the preceding line ending and
776 // this code block, `str::lines` will return an additional line,
777 // which we subtract here.
778 if nb_lines != 0 && !&doc[prev_offset..offset.start].ends_with('\n') {
781 let line = tests.get_line() + nb_lines + 1;
782 tests.add_test(text, block_info, line);
783 prev_offset = offset.start;
785 Event::Start(Tag::Heading(level, _, _)) => {
786 register_header = Some(level as u32);
788 Event::Text(ref s) if register_header.is_some() => {
789 let level = register_header.unwrap();
791 tests.register_header("", level);
793 tests.register_header(s, level);
795 register_header = None;
802 pub(crate) struct ExtraInfo<'tcx> {
808 impl<'tcx> ExtraInfo<'tcx> {
809 pub(crate) fn new(tcx: TyCtxt<'tcx>, def_id: DefId, sp: Span) -> ExtraInfo<'tcx> {
810 ExtraInfo { def_id, sp, tcx }
813 fn error_invalid_codeblock_attr(&self, msg: &str, help: &str) {
814 if let Some(def_id) = self.def_id.as_local() {
815 self.tcx.struct_span_lint_hir(
816 crate::lint::INVALID_CODEBLOCK_ATTRIBUTES,
817 self.tcx.hir().local_def_id_to_hir_id(def_id),
820 |lint| lint.help(help),
826 #[derive(Eq, PartialEq, Clone, Debug)]
827 pub(crate) struct LangString {
829 pub(crate) should_panic: bool,
830 pub(crate) no_run: bool,
831 pub(crate) ignore: Ignore,
832 pub(crate) rust: bool,
833 pub(crate) test_harness: bool,
834 pub(crate) compile_fail: bool,
835 pub(crate) error_codes: Vec<String>,
836 pub(crate) edition: Option<Edition>,
839 #[derive(Eq, PartialEq, Clone, Debug)]
840 pub(crate) enum Ignore {
846 impl Default for LangString {
847 fn default() -> Self {
849 original: String::new(),
852 ignore: Ignore::None,
856 error_codes: Vec::new(),
863 fn parse_without_check(
865 allow_error_code_check: ErrorCodes,
866 enable_per_target_ignores: bool,
868 Self::parse(string, allow_error_code_check, enable_per_target_ignores, None)
871 fn tokens(string: &str) -> impl Iterator<Item = &str> {
872 // Pandoc, which Rust once used for generating documentation,
873 // expects lang strings to be surrounded by `{}` and for each token
874 // to be proceeded by a `.`. Since some of these lang strings are still
875 // loose in the wild, we strip a pair of surrounding `{}` from the lang
876 // string and a leading `.` from each token.
878 let string = string.trim();
880 let first = string.chars().next();
881 let last = string.chars().last();
883 let string = if first == Some('{') && last == Some('}') {
884 &string[1..string.len() - 1]
890 .split(|c| c == ',' || c == ' ' || c == '\t')
892 .map(|token| token.strip_prefix('.').unwrap_or(token))
893 .filter(|token| !token.is_empty())
898 allow_error_code_check: ErrorCodes,
899 enable_per_target_ignores: bool,
900 extra: Option<&ExtraInfo<'_>>,
902 let allow_error_code_check = allow_error_code_check.as_bool();
903 let mut seen_rust_tags = false;
904 let mut seen_other_tags = false;
905 let mut data = LangString::default();
906 let mut ignores = vec![];
908 data.original = string.to_owned();
910 for token in Self::tokens(string) {
913 data.should_panic = true;
914 seen_rust_tags = !seen_other_tags;
918 seen_rust_tags = !seen_other_tags;
921 data.ignore = Ignore::All;
922 seen_rust_tags = !seen_other_tags;
924 x if x.starts_with("ignore-") => {
925 if enable_per_target_ignores {
926 ignores.push(x.trim_start_matches("ignore-").to_owned());
927 seen_rust_tags = !seen_other_tags;
932 seen_rust_tags = true;
935 data.test_harness = true;
936 seen_rust_tags = !seen_other_tags || seen_rust_tags;
939 data.compile_fail = true;
940 seen_rust_tags = !seen_other_tags || seen_rust_tags;
943 x if x.starts_with("edition") => {
944 data.edition = x[7..].parse::<Edition>().ok();
946 x if allow_error_code_check && x.starts_with('E') && x.len() == 5 => {
947 if x[1..].parse::<u32>().is_ok() {
948 data.error_codes.push(x.to_owned());
949 seen_rust_tags = !seen_other_tags || seen_rust_tags;
951 seen_other_tags = true;
954 x if extra.is_some() => {
955 let s = x.to_lowercase();
956 if let Some((flag, help)) = if s == "compile-fail"
957 || s == "compile_fail"
958 || s == "compilefail"
962 "the code block will either not be tested if not marked as a rust one \
963 or won't fail if it compiles successfully",
965 } else if s == "should-panic" || s == "should_panic" || s == "shouldpanic" {
968 "the code block will either not be tested if not marked as a rust one \
969 or won't fail if it doesn't panic when running",
971 } else if s == "no-run" || s == "no_run" || s == "norun" {
974 "the code block will either not be tested if not marked as a rust one \
975 or will be run (which you might not want)",
977 } else if s == "test-harness" || s == "test_harness" || s == "testharness" {
980 "the code block will either not be tested if not marked as a rust one \
981 or the code will be wrapped inside a main function",
986 if let Some(extra) = extra {
987 extra.error_invalid_codeblock_attr(
988 &format!("unknown attribute `{}`. Did you mean `{}`?", x, flag),
993 seen_other_tags = true;
995 _ => seen_other_tags = true,
999 // ignore-foo overrides ignore
1000 if !ignores.is_empty() {
1001 data.ignore = Ignore::Some(ignores);
1004 data.rust &= !seen_other_tags || seen_rust_tags;
1011 pub fn into_string(self) -> String {
1022 // This is actually common enough to special-case
1024 return String::new();
1026 let mut replacer = |broken_link: BrokenLink<'_>| {
1029 .find(|link| link.original_text.as_str() == &*broken_link.reference)
1030 .map(|link| (link.href.as_str().into(), link.new_text.as_str().into()))
1033 let p = Parser::new_with_broken_link_callback(md, main_body_opts(), Some(&mut replacer));
1034 let p = p.into_offset_iter();
1036 let mut s = String::with_capacity(md.len() * 3 / 2);
1038 let p = HeadingLinks::new(p, None, ids, heading_offset);
1039 let p = Footnotes::new(p);
1040 let p = LinkReplacer::new(p.map(|(ev, _)| ev), links);
1041 let p = TableWrapper::new(p);
1042 let p = CodeBlocks::new(p, codes, edition, playground);
1043 html::push_html(&mut s, p);
1049 impl MarkdownWithToc<'_> {
1050 pub(crate) fn into_string(self) -> String {
1051 let MarkdownWithToc(md, ids, codes, edition, playground) = self;
1053 let p = Parser::new_ext(md, main_body_opts()).into_offset_iter();
1055 let mut s = String::with_capacity(md.len() * 3 / 2);
1057 let mut toc = TocBuilder::new();
1060 let p = HeadingLinks::new(p, Some(&mut toc), ids, HeadingOffset::H1);
1061 let p = Footnotes::new(p);
1062 let p = TableWrapper::new(p.map(|(ev, _)| ev));
1063 let p = CodeBlocks::new(p, codes, edition, playground);
1064 html::push_html(&mut s, p);
1067 format!("<nav id=\"TOC\">{}</nav>{}", toc.into_toc().print(), s)
1071 impl MarkdownItemInfo<'_> {
1072 pub(crate) fn into_string(self) -> String {
1073 let MarkdownItemInfo(md, ids) = self;
1075 // This is actually common enough to special-case
1077 return String::new();
1079 let p = Parser::new_ext(md, main_body_opts()).into_offset_iter();
1081 // Treat inline HTML as plain text.
1082 let p = p.map(|event| match event.0 {
1083 Event::Html(text) => (Event::Text(text), event.1),
1087 let mut s = String::with_capacity(md.len() * 3 / 2);
1089 let p = HeadingLinks::new(p, None, ids, HeadingOffset::H1);
1090 let p = Footnotes::new(p);
1091 let p = TableWrapper::new(p.map(|(ev, _)| ev));
1092 let p = p.filter(|event| {
1093 !matches!(event, Event::Start(Tag::Paragraph) | Event::End(Tag::Paragraph))
1095 html::push_html(&mut s, p);
1101 impl MarkdownSummaryLine<'_> {
1102 pub(crate) fn into_string_with_has_more_content(self) -> (String, bool) {
1103 let MarkdownSummaryLine(md, links) = self;
1104 // This is actually common enough to special-case
1106 return (String::new(), false);
1109 let mut replacer = |broken_link: BrokenLink<'_>| {
1112 .find(|link| link.original_text.as_str() == &*broken_link.reference)
1113 .map(|link| (link.href.as_str().into(), link.new_text.as_str().into()))
1116 let p = Parser::new_with_broken_link_callback(md, summary_opts(), Some(&mut replacer))
1118 let mut summary = SummaryLine::new(p);
1120 let mut s = String::new();
1122 let without_paragraphs = LinkReplacer::new(&mut summary, links).filter(|event| {
1123 !matches!(event, Event::Start(Tag::Paragraph) | Event::End(Tag::Paragraph))
1126 html::push_html(&mut s, without_paragraphs);
1128 let has_more_content =
1129 matches!(summary.inner.peek(), Some(Event::Start(_))) || summary.skipped_tags > 0;
1131 (s, has_more_content)
1134 pub(crate) fn into_string(self) -> String {
1135 self.into_string_with_has_more_content().0
1139 /// Renders a subset of Markdown in the first paragraph of the provided Markdown.
1141 /// - *Italics*, **bold**, and `inline code` styles **are** rendered.
1142 /// - Headings and links are stripped (though the text *is* rendered).
1143 /// - HTML, code blocks, and everything else are ignored.
1145 /// Returns a tuple of the rendered HTML string and whether the output was shortened
1146 /// due to the provided `length_limit`.
1147 fn markdown_summary_with_limit(
1149 link_names: &[RenderedLink],
1150 length_limit: usize,
1151 ) -> (String, bool) {
1153 return (String::new(), false);
1156 let mut replacer = |broken_link: BrokenLink<'_>| {
1159 .find(|link| link.original_text.as_str() == &*broken_link.reference)
1160 .map(|link| (link.href.as_str().into(), link.new_text.as_str().into()))
1163 let p = Parser::new_with_broken_link_callback(md, summary_opts(), Some(&mut replacer));
1164 let mut p = LinkReplacer::new(p, link_names);
1166 let mut buf = HtmlWithLimit::new(length_limit);
1167 let mut stopped_early = false;
1168 p.try_for_each(|event| {
1170 Event::Text(text) => {
1172 text.split_inclusive(char::is_whitespace).try_for_each(|word| buf.push(word));
1174 stopped_early = true;
1178 Event::Code(code) => {
1179 buf.open_tag("code");
1180 let r = buf.push(code);
1182 stopped_early = true;
1188 Event::Start(tag) => match tag {
1189 Tag::Emphasis => buf.open_tag("em"),
1190 Tag::Strong => buf.open_tag("strong"),
1191 Tag::CodeBlock(..) => return ControlFlow::Break(()),
1194 Event::End(tag) => match tag {
1195 Tag::Emphasis | Tag::Strong => buf.close_tag(),
1196 Tag::Paragraph | Tag::Heading(..) => return ControlFlow::Break(()),
1199 Event::HardBreak | Event::SoftBreak => buf.push(" ")?,
1202 ControlFlow::Continue(())
1205 (buf.finish(), stopped_early)
1208 /// Renders a shortened first paragraph of the given Markdown as a subset of Markdown,
1209 /// making it suitable for contexts like the search index.
1211 /// Will shorten to 59 or 60 characters, including an ellipsis (…) if it was shortened.
1213 /// See [`markdown_summary_with_limit`] for details about what is rendered and what is not.
1214 pub(crate) fn short_markdown_summary(markdown: &str, link_names: &[RenderedLink]) -> String {
1215 let (mut s, was_shortened) = markdown_summary_with_limit(markdown, link_names, 59);
1224 /// Renders the first paragraph of the provided markdown as plain text.
1225 /// Useful for alt-text.
1227 /// - Headings, links, and formatting are stripped.
1228 /// - Inline code is rendered as-is, surrounded by backticks.
1229 /// - HTML and code blocks are ignored.
1230 pub(crate) fn plain_text_summary(md: &str) -> String {
1232 return String::new();
1235 let mut s = String::with_capacity(md.len() * 3 / 2);
1237 for event in Parser::new_ext(md, summary_opts()) {
1239 Event::Text(text) => s.push_str(text),
1240 Event::Code(code) => {
1245 Event::HardBreak | Event::SoftBreak => s.push(' '),
1246 Event::Start(Tag::CodeBlock(..)) => break,
1247 Event::End(Tag::Paragraph) => break,
1248 Event::End(Tag::Heading(..)) => break,
1257 pub(crate) struct MarkdownLink {
1260 pub range: Range<usize>,
1263 pub(crate) fn markdown_links<R>(
1265 filter_map: impl Fn(MarkdownLink) -> Option<R>,
1271 let links = RefCell::new(vec![]);
1273 // FIXME: remove this function once pulldown_cmark can provide spans for link definitions.
1274 let locate = |s: &str, fallback: Range<usize>| unsafe {
1275 let s_start = s.as_ptr();
1276 let s_end = s_start.add(s.len());
1277 let md_start = md.as_ptr();
1278 let md_end = md_start.add(md.len());
1279 if md_start <= s_start && s_end <= md_end {
1280 let start = s_start.offset_from(md_start) as usize;
1281 let end = s_end.offset_from(md_start) as usize;
1288 let span_for_link = |link: &CowStr<'_>, span: Range<usize>| {
1289 // For diagnostics, we want to underline the link's definition but `span` will point at
1290 // where the link is used. This is a problem for reference-style links, where the definition
1291 // is separate from the usage.
1293 // `Borrowed` variant means the string (the link's destination) may come directly from
1294 // the markdown text and we can locate the original link destination.
1295 // NOTE: LinkReplacer also provides `Borrowed` but possibly from other sources,
1296 // so `locate()` can fall back to use `span`.
1297 CowStr::Borrowed(s) => locate(s, span),
1299 // For anything else, we can only use the provided range.
1300 CowStr::Boxed(_) | CowStr::Inlined(_) => span,
1304 let mut push = |link: BrokenLink<'_>| {
1305 let span = span_for_link(&link.reference, link.span);
1306 filter_map(MarkdownLink {
1307 kind: LinkType::ShortcutUnknown,
1308 link: link.reference.to_string(),
1311 .map(|link| links.borrow_mut().push(link));
1314 let p = Parser::new_with_broken_link_callback(md, main_body_opts(), Some(&mut push))
1315 .into_offset_iter();
1317 // There's no need to thread an IdMap through to here because
1318 // the IDs generated aren't going to be emitted anywhere.
1319 let mut ids = IdMap::new();
1320 let iter = Footnotes::new(HeadingLinks::new(p, None, &mut ids, HeadingOffset::H1));
1323 if let Event::Start(Tag::Link(
1324 // `<>` links cannot be intra-doc links so we skip them.
1325 kind @ (LinkType::Inline
1326 | LinkType::Reference
1327 | LinkType::ReferenceUnknown
1328 | LinkType::Collapsed
1329 | LinkType::CollapsedUnknown
1330 | LinkType::Shortcut
1331 | LinkType::ShortcutUnknown),
1336 debug!("found link: {dest}");
1337 let span = span_for_link(&dest, ev.1);
1338 filter_map(MarkdownLink { kind, link: dest.into_string(), range: span })
1339 .map(|link| links.borrow_mut().push(link));
1347 pub(crate) struct RustCodeBlock {
1348 /// The range in the markdown that the code block occupies. Note that this includes the fences
1349 /// for fenced code blocks.
1350 pub(crate) range: Range<usize>,
1351 /// The range in the markdown that the code within the code block occupies.
1352 pub(crate) code: Range<usize>,
1353 pub(crate) is_fenced: bool,
1354 pub(crate) lang_string: LangString,
1357 /// Returns a range of bytes for each code block in the markdown that is tagged as `rust` or
1358 /// untagged (and assumed to be rust).
1359 pub(crate) fn rust_code_blocks(md: &str, extra_info: &ExtraInfo<'_>) -> Vec<RustCodeBlock> {
1360 let mut code_blocks = vec![];
1366 let mut p = Parser::new_ext(md, main_body_opts()).into_offset_iter();
1368 while let Some((event, offset)) = p.next() {
1369 if let Event::Start(Tag::CodeBlock(syntax)) = event {
1370 let (lang_string, code_start, code_end, range, is_fenced) = match syntax {
1371 CodeBlockKind::Fenced(syntax) => {
1372 let syntax = syntax.as_ref();
1373 let lang_string = if syntax.is_empty() {
1376 LangString::parse(&*syntax, ErrorCodes::Yes, false, Some(extra_info))
1378 if !lang_string.rust {
1381 let (code_start, mut code_end) = match p.next() {
1382 Some((Event::Text(_), offset)) => (offset.start, offset.end),
1383 Some((_, sub_offset)) => {
1384 let code = Range { start: sub_offset.start, end: sub_offset.start };
1385 code_blocks.push(RustCodeBlock {
1394 let code = Range { start: offset.end, end: offset.end };
1395 code_blocks.push(RustCodeBlock {
1404 while let Some((Event::Text(_), offset)) = p.next() {
1405 code_end = offset.end;
1407 (lang_string, code_start, code_end, offset, true)
1409 CodeBlockKind::Indented => {
1410 // The ending of the offset goes too far sometime so we reduce it by one in
1412 if offset.end > offset.start && md.get(offset.end..=offset.end) == Some("\n") {
1414 LangString::default(),
1417 Range { start: offset.start, end: offset.end - 1 },
1421 (LangString::default(), offset.start, offset.end, offset, false)
1426 code_blocks.push(RustCodeBlock {
1429 code: Range { start: code_start, end: code_end },
1438 #[derive(Clone, Default, Debug)]
1440 map: FxHashMap<Cow<'static, str>, usize>,
1443 // The map is pre-initialized and cloned each time to avoid reinitializing it repeatedly.
1444 static DEFAULT_ID_MAP: Lazy<FxHashMap<Cow<'static, str>, usize>> = Lazy::new(|| init_id_map());
1446 fn init_id_map() -> FxHashMap<Cow<'static, str>, usize> {
1447 let mut map = FxHashMap::default();
1448 // This is the list of IDs used in Javascript.
1449 map.insert("help".into(), 1);
1450 map.insert("settings".into(), 1);
1451 map.insert("not-displayed".into(), 1);
1452 map.insert("alternative-display".into(), 1);
1453 map.insert("search".into(), 1);
1454 map.insert("crate-search".into(), 1);
1455 map.insert("crate-search-div".into(), 1);
1456 // This is the list of IDs used in HTML generated in Rust (including the ones
1457 // used in tera template files).
1458 map.insert("mainThemeStyle".into(), 1);
1459 map.insert("themeStyle".into(), 1);
1460 map.insert("settings-menu".into(), 1);
1461 map.insert("help-button".into(), 1);
1462 map.insert("main-content".into(), 1);
1463 map.insert("toggle-all-docs".into(), 1);
1464 map.insert("all-types".into(), 1);
1465 map.insert("default-settings".into(), 1);
1466 map.insert("rustdoc-vars".into(), 1);
1467 map.insert("sidebar-vars".into(), 1);
1468 map.insert("copy-path".into(), 1);
1469 map.insert("TOC".into(), 1);
1470 // This is the list of IDs used by rustdoc sections (but still generated by
1472 map.insert("fields".into(), 1);
1473 map.insert("variants".into(), 1);
1474 map.insert("implementors-list".into(), 1);
1475 map.insert("synthetic-implementors-list".into(), 1);
1476 map.insert("foreign-impls".into(), 1);
1477 map.insert("implementations".into(), 1);
1478 map.insert("trait-implementations".into(), 1);
1479 map.insert("synthetic-implementations".into(), 1);
1480 map.insert("blanket-implementations".into(), 1);
1481 map.insert("required-associated-types".into(), 1);
1482 map.insert("provided-associated-types".into(), 1);
1483 map.insert("provided-associated-consts".into(), 1);
1484 map.insert("required-associated-consts".into(), 1);
1485 map.insert("required-methods".into(), 1);
1486 map.insert("provided-methods".into(), 1);
1487 map.insert("implementors".into(), 1);
1488 map.insert("synthetic-implementors".into(), 1);
1489 map.insert("implementations-list".into(), 1);
1490 map.insert("trait-implementations-list".into(), 1);
1491 map.insert("synthetic-implementations-list".into(), 1);
1492 map.insert("blanket-implementations-list".into(), 1);
1493 map.insert("deref-methods".into(), 1);
1494 map.insert("layout".into(), 1);
1499 pub fn new() -> Self {
1500 IdMap { map: DEFAULT_ID_MAP.clone() }
1503 pub(crate) fn derive<S: AsRef<str> + ToString>(&mut self, candidate: S) -> String {
1504 let id = match self.map.get_mut(candidate.as_ref()) {
1505 None => candidate.to_string(),
1507 let id = format!("{}-{}", candidate.as_ref(), *a);
1513 self.map.insert(id.clone().into(), 1);