1 //! Detects invalid HTML (like an unclosed `<span>`) in doc comments.
4 use crate::core::DocContext;
5 use crate::html::markdown::main_body_opts;
6 use crate::visit::DocVisitor;
8 use pulldown_cmark::{BrokenLink, Event, LinkType, Parser, Tag};
10 use std::iter::Peekable;
12 use std::str::CharIndices;
14 pub(crate) const CHECK_INVALID_HTML_TAGS: Pass = Pass {
15 name: "check-invalid-html-tags",
16 run: check_invalid_html_tags,
17 description: "detects invalid HTML tags in doc comments",
20 struct InvalidHtmlTagsLinter<'a, 'tcx> {
21 cx: &'a mut DocContext<'tcx>,
24 pub(crate) fn check_invalid_html_tags(krate: Crate, cx: &mut DocContext<'_>) -> Crate {
25 let mut coll = InvalidHtmlTagsLinter { cx };
26 coll.visit_crate(&krate);
30 const ALLOWED_UNCLOSED: &[&str] = &[
31 "area", "base", "br", "col", "embed", "hr", "img", "input", "keygen", "link", "meta", "param",
32 "source", "track", "wbr",
36 tags: &mut Vec<(String, Range<usize>)>,
39 f: &impl Fn(&str, &Range<usize>, bool),
41 let tag_name_low = tag_name.to_lowercase();
42 if let Some(pos) = tags.iter().rposition(|(t, _)| t.to_lowercase() == tag_name_low) {
43 // If the tag is nested inside a "<script>" or a "<style>" tag, no warning should
45 let should_not_warn = tags.iter().take(pos + 1).any(|(at, _)| {
46 let at = at.to_lowercase();
47 at == "script" || at == "style"
49 for (last_tag_name, last_tag_span) in tags.drain(pos + 1..) {
53 let last_tag_name_low = last_tag_name.to_lowercase();
54 if ALLOWED_UNCLOSED.contains(&last_tag_name_low.as_str()) {
57 // `tags` is used as a queue, meaning that everything after `pos` is included inside it.
58 // So `<h2><h3></h2>` will look like `["h2", "h3"]`. So when closing `h2`, we will still
59 // have `h3`, meaning the tag wasn't closed as it should have.
60 f(&format!("unclosed HTML tag `{}`", last_tag_name), &last_tag_span, true);
62 // Remove the `tag_name` that was originally closed
65 // It can happen for example in this case: `<h2></script></h2>` (the `h2` tag isn't required
66 // but it helps for the visualization).
67 f(&format!("unopened HTML tag `{}`", tag_name), &range, false);
71 fn extract_path_backwards(text: &str, end_pos: usize) -> Option<usize> {
72 use rustc_lexer::{is_id_continue, is_id_start};
73 let mut current_pos = end_pos;
75 if current_pos >= 2 && text[..current_pos].ends_with("::") {
78 let new_pos = text[..current_pos]
81 .take_while(|(_, c)| is_id_start(*c) || is_id_continue(*c))
82 .reduce(|_accum, item| item)
83 .and_then(|(new_pos, c)| is_id_start(c).then_some(new_pos));
84 if let Some(new_pos) = new_pos {
85 if current_pos != new_pos {
86 current_pos = new_pos;
92 if current_pos == end_pos { None } else { Some(current_pos) }
95 fn extract_path_forward(text: &str, start_pos: usize) -> Option<usize> {
96 use rustc_lexer::{is_id_continue, is_id_start};
97 let mut current_pos = start_pos;
99 if current_pos < text.len() && text[current_pos..].starts_with("::") {
104 let mut chars = text[current_pos..].chars();
105 if let Some(c) = chars.next() {
107 current_pos += c.len_utf8();
112 while let Some(c) = chars.next() {
113 if is_id_continue(c) {
114 current_pos += c.len_utf8();
120 if current_pos == start_pos { None } else { Some(current_pos) }
123 fn is_valid_for_html_tag_name(c: char, is_empty: bool) -> bool {
124 // https://spec.commonmark.org/0.30/#raw-html
126 // > A tag name consists of an ASCII letter followed by zero or more ASCII letters, digits, or
128 c.is_ascii_alphabetic() || !is_empty && (c == '-' || c.is_ascii_digit())
132 tags: &mut Vec<(String, Range<usize>)>,
134 range: &Range<usize>,
136 iter: &mut Peekable<CharIndices<'_>>,
137 f: &impl Fn(&str, &Range<usize>, bool),
139 let mut tag_name = String::new();
140 let mut is_closing = false;
141 let mut prev_pos = start_pos;
144 let (pos, c) = match iter.peek() {
145 Some((pos, c)) => (*pos, *c),
146 // In case we reached the of the doc comment, we want to check that it's an
147 // unclosed HTML tag. For example "/// <h3".
148 None => (prev_pos, '\0'),
151 // Checking if this is a closing tag (like `</a>` for `<a>`).
152 if c == '/' && tag_name.is_empty() {
154 } else if is_valid_for_html_tag_name(c, tag_name.is_empty()) {
157 if !tag_name.is_empty() {
158 let mut r = Range { start: range.start + start_pos, end: range.start + pos };
160 // In case we have a tag without attribute, we can consider the span to
161 // refer to it fully.
165 // In case we have "</div >" or even "</div >".
167 if !c.is_whitespace() {
168 // It seems like it's not a valid HTML tag.
171 let mut found = false;
172 for (new_pos, c) in text[pos..].char_indices() {
173 if !c.is_whitespace() {
175 r.end = range.start + new_pos + 1;
185 drop_tag(tags, tag_name, r, f);
187 tags.push((tag_name, r));
197 tags: &mut Vec<(String, Range<usize>)>,
200 is_in_comment: &mut Option<Range<usize>>,
201 f: &impl Fn(&str, &Range<usize>, bool),
203 let mut iter = text.char_indices().peekable();
205 while let Some((start_pos, c)) = iter.next() {
206 if is_in_comment.is_some() {
207 if text[start_pos..].starts_with("-->") {
208 *is_in_comment = None;
211 if text[start_pos..].starts_with("<!--") {
212 // We skip the "!--" part. (Once `advance_by` is stable, might be nice to use it!)
216 *is_in_comment = Some(Range {
217 start: range.start + start_pos,
218 end: range.start + start_pos + 3,
221 extract_html_tag(tags, text, &range, start_pos, &mut iter, f);
227 impl<'a, 'tcx> DocVisitor for InvalidHtmlTagsLinter<'a, 'tcx> {
228 fn visit_item(&mut self, item: &Item) {
229 let tcx = self.cx.tcx;
230 let Some(hir_id) = DocContext::as_local_hir_id(tcx, item.item_id)
231 // If non-local, no need to check anything.
233 let dox = item.attrs.collapsed_doc_value().unwrap_or_default();
235 let report_diag = |msg: &str, range: &Range<usize>, is_open_tag: bool| {
236 let sp = match super::source_span_for_markdown_range(tcx, &dox, range, &item.attrs)
239 None => item.attr_span(tcx),
241 tcx.struct_span_lint_hir(crate::lint::INVALID_HTML_TAGS, hir_id, sp, msg, |lint| {
242 use rustc_lint_defs::Applicability;
243 // If a tag looks like `<this>`, it might actually be a generic.
244 // We don't try to detect stuff `<like, this>` because that's not valid HTML,
245 // and we don't try to detect stuff `<like this>` because that's not valid Rust.
246 let mut generics_end = range.end;
247 if let Some(Some(mut generics_start)) = (is_open_tag
248 && dox[..generics_end].ends_with('>'))
249 .then(|| extract_path_backwards(&dox, range.start))
251 while generics_start != 0
252 && generics_end < dox.len()
253 && dox.as_bytes()[generics_start - 1] == b'<'
254 && dox.as_bytes()[generics_end] == b'>'
258 if let Some(new_start) = extract_path_backwards(&dox, generics_start) {
259 generics_start = new_start;
261 if let Some(new_end) = extract_path_forward(&dox, generics_end) {
262 generics_end = new_end;
265 if let Some(new_end) = extract_path_forward(&dox, generics_end) {
266 generics_end = new_end;
268 let generics_sp = match super::source_span_for_markdown_range(
271 &(generics_start..generics_end),
275 None => item.attr_span(tcx),
277 // Sometimes, we only extract part of a path. For example, consider this:
279 // <[u32] as IntoIter<u32>>::Item
280 // ^^^^^ unclosed HTML tag `u32`
282 // We don't have any code for parsing fully-qualified trait paths.
283 // In theory, we could add it, but doing it correctly would require
284 // parsing the entire path grammar, which is problematic because of
285 // overlap between the path grammar and Markdown.
287 // The example above shows that ambiguity. Is `[u32]` intended to be an
288 // intra-doc link to the u32 primitive, or is it intended to be a slice?
290 // If the below conditional were removed, we would suggest this, which is
291 // not what the user probably wants.
293 // <[u32] as `IntoIter<u32>`>::Item
295 // We know that the user actually wants to wrap the whole thing in a code
296 // block, but the only reason we know that is because `u32` does not, in
297 // fact, implement IntoIter. If the example looks like this:
299 // <[Vec<i32>] as IntoIter<i32>::Item
301 // The ideal fix would be significantly different.
302 if (generics_start > 0 && dox.as_bytes()[generics_start - 1] == b'<')
303 || (generics_end < dox.len() && dox.as_bytes()[generics_end] == b'>')
307 // multipart form is chosen here because ``Vec<i32>`` would be confusing.
308 lint.multipart_suggestion(
309 "try marking as source code",
311 (generics_sp.shrink_to_lo(), String::from("`")),
312 (generics_sp.shrink_to_hi(), String::from("`")),
314 Applicability::MaybeIncorrect,
322 let mut tags = Vec::new();
323 let mut is_in_comment = None;
324 let mut in_code_block = false;
326 let link_names = item.link_names(&self.cx.cache);
328 let mut replacer = |broken_link: BrokenLink<'_>| {
330 link_names.iter().find(|link| *link.original_text == *broken_link.reference)
332 Some((link.href.as_str().into(), link.new_text.as_str().into()))
334 &broken_link.link_type,
335 LinkType::Reference | LinkType::ReferenceUnknown
337 // If the link is shaped [like][this], suppress any broken HTML in the [this] part.
338 // The `broken_intra_doc_links` will report typos in there anyway.
340 broken_link.reference.to_string().into(),
341 broken_link.reference.to_string().into(),
349 Parser::new_with_broken_link_callback(&dox, main_body_opts(), Some(&mut replacer))
352 for (event, range) in p {
354 Event::Start(Tag::CodeBlock(_)) => in_code_block = true,
355 Event::Html(text) if !in_code_block => {
356 extract_tags(&mut tags, &text, range, &mut is_in_comment, &report_diag)
358 Event::End(Tag::CodeBlock(_)) => in_code_block = false,
363 for (tag, range) in tags.iter().filter(|(t, _)| {
364 let t = t.to_lowercase();
365 !ALLOWED_UNCLOSED.contains(&t.as_str())
367 report_diag(&format!("unclosed HTML tag `{}`", tag), range, true);
370 if let Some(range) = is_in_comment {
371 report_diag("Unclosed HTML comment", &range, false);
375 self.visit_item_recur(item)