rustdoc/passes/lint/
bare_urls.rs

1//! Detects links that are not linkified, e.g., in Markdown such as `Go to https://example.com/.`
2//! Suggests wrapping the link with angle brackets: `Go to <https://example.com/>.` to linkify it.
3
4use core::ops::Range;
5use std::sync::LazyLock;
6
7use regex::Regex;
8use rustc_errors::{Applicability, DiagDecorator};
9use rustc_hir::HirId;
10use rustc_resolve::rustdoc::pulldown_cmark::{
11    DefaultBrokenLinkCallback, Event, Tag, TextMergeWithOffset,
12};
13use rustc_resolve::rustdoc::source_span_for_markdown_range;
14use tracing::trace;
15
16use crate::clean::*;
17use crate::core::DocContext;
18use crate::html::markdown::main_body_opts;
19
20pub(super) fn visit_item(cx: &DocContext<'_>, item: &Item, hir_id: HirId, dox: &str) {
21    let report_diag = |cx: &DocContext<'_>,
22                       msg: &'static str,
23                       range: Range<usize>,
24                       without_brackets: Option<&str>| {
25        let maybe_sp = source_span_for_markdown_range(cx.tcx, dox, &range, &item.attrs.doc_strings)
26            .map(|(sp, _)| sp);
27        let sp = maybe_sp.unwrap_or_else(|| item.attr_span(cx.tcx));
28        cx.tcx.emit_node_span_lint(
29            crate::lint::BARE_URLS,
30            hir_id,
31            sp,
32            DiagDecorator(|lint| {
33                lint.primary_message(msg)
34                    .note("bare URLs are not automatically turned into clickable links");
35                // The fallback of using the attribute span is suitable for
36                // highlighting where the error is, but not for placing the < and >
37                if let Some(sp) = maybe_sp {
38                    if let Some(without_brackets) = without_brackets {
39                        lint.multipart_suggestion(
40                            "use an automatic link instead",
41                            vec![(sp, format!("<{without_brackets}>"))],
42                            Applicability::MachineApplicable,
43                        );
44                    } else {
45                        lint.multipart_suggestion(
46                            "use an automatic link instead",
47                            vec![
48                                (sp.shrink_to_lo(), "<".to_string()),
49                                (sp.shrink_to_hi(), ">".to_string()),
50                            ],
51                            Applicability::MachineApplicable,
52                        );
53                    }
54                }
55            }),
56        );
57    };
58
59    // pulldown-cmark can split a URL into multiple `Text` events while processing
60    // characters such as `_` according to CommonMark's emphasis rules.
61    // `TextMergeWithOffset` merges these events so we can check the complete URL.
62    let mut p = TextMergeWithOffset::<DefaultBrokenLinkCallback>::new_ext(dox, main_body_opts());
63
64    while let Some((event, range)) = p.next() {
65        match event {
66            Event::Text(s) => find_raw_urls(cx, dox, &s, range, &report_diag),
67            // We don't want to check the text inside code blocks or links.
68            Event::Start(tag @ (Tag::CodeBlock(_) | Tag::Link { .. })) => {
69                let end = tag.to_end();
70                for (event, _) in p.by_ref() {
71                    if matches!(event, Event::End(tag) if tag == end) {
72                        break;
73                    }
74                }
75            }
76            _ => {}
77        }
78    }
79}
80
81static URL_REGEX: LazyLock<Regex> = LazyLock::new(|| {
82    Regex::new(concat!(
83        r"https?://",                          // url scheme
84        r"([-a-zA-Z0-9@:%._\+~#=]{2,256}\.)+", // one or more subdomains
85        r"[a-zA-Z]{2,63}",                     // root domain
86        // Match URL characters and balanced parenthesized segments, without
87        // consuming a trailing `)` that belongs to the surrounding prose.
88        r"\b(?:",
89        r"[-a-zA-Z0-9@:%_\+.~#?&/=]",
90        r"|\([-a-zA-Z0-9@:%_\+.~#?&/=]*\)",
91        r")*",
92    ))
93    .expect("failed to build regex")
94});
95
96fn find_raw_urls(
97    cx: &DocContext<'_>,
98    dox: &str,
99    text: &str,
100    range: Range<usize>,
101    f: &impl Fn(&DocContext<'_>, &'static str, Range<usize>, Option<&str>),
102) {
103    trace!("looking for raw urls in {text}");
104    // For now, we only check "full" URLs (meaning, starting with "http://" or "https://").
105    for match_ in URL_REGEX.find_iter(text) {
106        let mut url_range = match_.range();
107        url_range.start += range.start;
108        url_range.end += range.start;
109        let mut without_brackets = None;
110        // If the link is contained inside `[]`, then we need to replace the brackets and
111        // not just add `<>`.
112        if dox[..url_range.start].ends_with('[')
113            && url_range.end <= dox.len()
114            && dox[url_range.end..].starts_with(']')
115        {
116            url_range.start -= 1;
117            url_range.end += 1;
118            without_brackets = Some(match_.as_str());
119        } else {
120            // Periods are valid in URLs, but very uncommon as the last character of one, while
121            // being very common as sentence punctuation right after one. Leave any trailing
122            // period out of the link, so that `Visit https://example.com/docs.` is linkified as
123            // `Visit <https://example.com/docs>.`.
124            let trailing_periods =
125                match_.as_str().len() - match_.as_str().trim_end_matches('.').len();
126            url_range.end -= trailing_periods;
127        }
128        f(cx, "this URL is not a hyperlink", url_range, without_brackets);
129    }
130}
rustdoc/passes/lint/bare_urls.rs

rustdoc/passes/lint/
bare_urls.rs
