rustc_resolve/

rustdoc.rs

use std::mem;
use std::ops::Range;

use itertools::Itertools;
/// Re-export the markdown parser used by rustdoc.
pub use pulldown_cmark;
use pulldown_cmark::{
    BrokenLink, BrokenLinkCallback, CowStr, Event, LinkType, Options, Parser, Tag,
};
use rustc_ast as ast;
use rustc_ast::attr::AttributeExt;
use rustc_ast::token::DocFragmentKind;
use rustc_ast::util::comments::beautify_doc_string;
use rustc_ast::{AttrStyle, join_path_syms};
use rustc_data_structures::fx::FxIndexMap;
use rustc_data_structures::unord::UnordSet;
use rustc_middle::ty::TyCtxt;
use rustc_span::def_id::DefId;
use rustc_span::source_map::SourceMap;
use rustc_span::{DUMMY_SP, InnerSpan, Span, Symbol, sym};
use thin_vec::ThinVec;
use tracing::{debug, trace};

#[cfg(test)]
mod tests;

/// A portion of documentation, extracted from a `#[doc]` attribute.
///
/// Each variant contains the line number within the complete doc-comment where the fragment
/// starts, as well as the Span where the corresponding doc comment or attribute is located.
///
/// Included files are kept separate from inline doc comments so that proper line-number
/// information can be given when a doctest fails. Sugared doc comments and "raw" doc comments are
/// kept separate because of issue #42760.
#[derive(Clone, PartialEq, Eq, Debug)]
pub struct DocFragment {
    pub span: Span,
    /// The item this doc-comment came from.
    /// Used to determine the scope in which doc links in this fragment are resolved.
    /// Typically filled for reexport docs when they are merged into the docs of the
    /// original reexported item.
    /// If the id is not filled, which happens for the original reexported item, then
    /// it has to be taken from somewhere else during doc link resolution.
    pub item_id: Option<DefId>,
    pub doc: Symbol,
    pub kind: DocFragmentKind,
    pub indent: usize,
    /// Because we tamper with the spans context, this information cannot be correctly retrieved
    /// later on. So instead, we compute it and store it here.
    pub from_expansion: bool,
    pub style: AttrStyle,
}

#[derive(Clone, Copy, Debug)]
pub enum MalformedGenerics {
    /// This link has unbalanced angle brackets.
    ///
    /// For example, `Vec<T` should trigger this, as should `Vec<T>>`.
    UnbalancedAngleBrackets,
    /// The generics are not attached to a type.
    ///
    /// For example, `<T>` should trigger this.
    ///
    /// This is detected by checking if the path is empty after the generics are stripped.
    MissingType,
    /// The link uses fully-qualified syntax, which is currently unsupported.
    ///
    /// For example, `<Vec as IntoIterator>::into_iter` should trigger this.
    ///
    /// This is detected by checking if ` as ` (the keyword `as` with spaces around it) is inside
    /// angle brackets.
    HasFullyQualifiedSyntax,
    /// The link has an invalid path separator.
    ///
    /// For example, `Vec:<T>:new()` should trigger this. Note that `Vec:new()` will **not**
    /// trigger this because it has no generics and thus [`strip_generics_from_path`] will not be
    /// called.
    ///
    /// Note that this will also **not** be triggered if the invalid path separator is inside angle
    /// brackets because rustdoc mostly ignores what's inside angle brackets (except for
    /// [`HasFullyQualifiedSyntax`](MalformedGenerics::HasFullyQualifiedSyntax)).
    ///
    /// This is detected by checking if there is a colon followed by a non-colon in the link.
    InvalidPathSeparator,
    /// The link has too many angle brackets.
    ///
    /// For example, `Vec<<T>>` should trigger this.
    TooManyAngleBrackets,
    /// The link has empty angle brackets.
    ///
    /// For example, `Vec<>` should trigger this.
    EmptyAngleBrackets,
}

/// Removes excess indentation on comments in order for the Markdown
/// to be parsed correctly. This is necessary because the convention for
/// writing documentation is to provide a space between the /// or //! marker
/// and the doc text, but Markdown is whitespace-sensitive. For example,
/// a block of text with four-space indentation is parsed as a code block,
/// so if we didn't unindent comments, these list items
///
/// /// A list:
/// ///
/// ///    - Foo
/// ///    - Bar
///
/// would be parsed as if they were in a code block, which is likely not what the user intended.
pub fn unindent_doc_fragments(docs: &mut [DocFragment]) {
    // `add` is used in case the most common sugared doc syntax is used ("/// "). The other
    // fragments kind's lines are never starting with a whitespace unless they are using some
    // markdown formatting requiring it. Therefore, if the doc block have a mix between the two,
    // we need to take into account the fact that the minimum indent minus one (to take this
    // whitespace into account).
    //
    // For example:
    //
    // /// hello!
    // #[doc = "another"]
    //
    // In this case, you want "hello! another" and not "hello!  another".
    let add = if docs.windows(2).any(|arr| arr[0].kind != arr[1].kind)
        && docs.iter().any(|d| d.kind.is_sugared())
    {
        // In case we have a mix of sugared doc comments and "raw" ones, we want the sugared one to
        // "decide" how much the minimum indent will be.
        1
    } else {
        0
    };

    // `min_indent` is used to know how much whitespaces from the start of each lines must be
    // removed. Example:
    //
    // ///     hello!
    // #[doc = "another"]
    //
    // In here, the `min_indent` is 1 (because non-sugared fragment are always counted with minimum
    // 1 whitespace), meaning that "hello!" will be considered a codeblock because it starts with 4
    // (5 - 1) whitespaces.
    let Some(min_indent) = docs
        .iter()
        .map(|fragment| {
            fragment
                .doc
                .as_str()
                .lines()
                .filter(|line| line.chars().any(|c| !c.is_whitespace()))
                .map(|line| {
                    // Compare against either space or tab, ignoring whether they are
                    // mixed or not.
                    let whitespace = line.chars().take_while(|c| *c == ' ' || *c == '\t').count();
                    whitespace + (if fragment.kind.is_sugared() { 0 } else { add })
                })
                .min()
                .unwrap_or(usize::MAX)
        })
        .min()
    else {
        return;
    };

    for fragment in docs {
        if fragment.doc == sym::empty {
            continue;
        }

        let indent = if !fragment.kind.is_sugared() && min_indent > 0 {
            min_indent - add
        } else {
            min_indent
        };

        fragment.indent = indent;
    }
}

/// The goal of this function is to apply the `DocFragment` transformation that is required when
/// transforming into the final Markdown, which is applying the computed indent to each line in
/// each doc fragment (a `DocFragment` can contain multiple lines in case of `#[doc = ""]`).
///
/// Note: remove the trailing newline where appropriate
pub fn add_doc_fragment(out: &mut String, frag: &DocFragment) {
    if frag.doc == sym::empty {
        out.push('\n');
        return;
    }
    let s = frag.doc.as_str();
    let mut iter = s.lines();

    while let Some(line) = iter.next() {
        if line.chars().any(|c| !c.is_whitespace()) {
            assert!(line.len() >= frag.indent);
            out.push_str(&line[frag.indent..]);
        } else {
            out.push_str(line);
        }
        out.push('\n');
    }
}

pub fn attrs_to_doc_fragments<'a, A: AttributeExt + Clone + 'a>(
    attrs: impl Iterator<Item = (&'a A, Option<DefId>)>,
    doc_only: bool,
) -> (Vec<DocFragment>, ThinVec<A>) {
    let (min_size, max_size) = attrs.size_hint();
    let size_hint = max_size.unwrap_or(min_size);
    let mut doc_fragments = Vec::with_capacity(size_hint);
    let mut other_attrs = ThinVec::<A>::with_capacity(if doc_only { 0 } else { size_hint });
    for (attr, item_id) in attrs {
        if let Some((doc_str, fragment_kind)) = attr.doc_str_and_fragment_kind() {
            let doc = beautify_doc_string(doc_str, fragment_kind.comment_kind());
            let attr_span = attr.span();
            let (span, from_expansion) = match fragment_kind {
                DocFragmentKind::Sugared(_) => (attr_span, attr_span.from_expansion()),
                DocFragmentKind::Raw(value_span) => {
                    (value_span.with_ctxt(attr_span.ctxt()), value_span.from_expansion())
                }
            };
            let fragment = DocFragment {
                span,
                doc,
                kind: fragment_kind,
                item_id,
                indent: 0,
                from_expansion,
                style: attr.doc_resolution_scope().unwrap(),
            };
            doc_fragments.push(fragment);
        } else if !doc_only {
            other_attrs.push(attr.clone());
        }
    }

    doc_fragments.shrink_to_fit();
    other_attrs.shrink_to_fit();

    unindent_doc_fragments(&mut doc_fragments);

    (doc_fragments, other_attrs)
}

/// Represents a doc comment, split in two parts, outer and inner attributes. It's needed for
/// the intra-doc link resolution context.
#[derive(Default)]
pub struct DocStrings {
    pub outer: String,
    pub inner: String,
}

/// Return the doc-comments on this item, grouped by the module they came from.
/// The module can be different if this is a re-export with added documentation.
///
/// The last newline is not trimmed so the produced strings are reusable between
/// early and late doc link resolution regardless of their position.
pub fn prepare_to_doc_link_resolution(
    doc_fragments: &[DocFragment],
) -> FxIndexMap<Option<DefId>, DocStrings> {
    let mut res: FxIndexMap<Option<DefId>, DocStrings> = FxIndexMap::default();
    for fragment in doc_fragments {
        let out_strs = res.entry(fragment.item_id).or_default();
        let out = if fragment.style == AttrStyle::Inner {
            &mut out_strs.inner
        } else {
            &mut out_strs.outer
        };
        add_doc_fragment(out, fragment);
    }
    res
}

/// Options for rendering Markdown in the main body of documentation.
pub fn main_body_opts() -> Options {
    Options::ENABLE_TABLES
        | Options::ENABLE_FOOTNOTES
        | Options::ENABLE_STRIKETHROUGH
        | Options::ENABLE_TASKLISTS
        | Options::ENABLE_SMART_PUNCTUATION
}

fn strip_generics_from_path_segment(segment: Vec<char>) -> Result<Symbol, MalformedGenerics> {
    let mut stripped_segment = String::new();
    let mut param_depth = 0;

    let mut latest_generics_chunk = String::new();

    for c in segment {
        if c == '<' {
            param_depth += 1;
            latest_generics_chunk.clear();
        } else if c == '>' {
            param_depth -= 1;
            if latest_generics_chunk.contains(" as ") {
                // The segment tries to use fully-qualified syntax, which is currently unsupported.
                // Give a helpful error message instead of completely ignoring the angle brackets.
                return Err(MalformedGenerics::HasFullyQualifiedSyntax);
            }
        } else if param_depth == 0 {
            stripped_segment.push(c);
        } else {
            latest_generics_chunk.push(c);
        }
    }

    if param_depth == 0 {
        Ok(Symbol::intern(&stripped_segment))
    } else {
        // The segment has unbalanced angle brackets, e.g. `Vec<T` or `Vec<T>>`
        Err(MalformedGenerics::UnbalancedAngleBrackets)
    }
}

pub fn strip_generics_from_path(path_str: &str) -> Result<Box<str>, MalformedGenerics> {
    if !path_str.contains(['<', '>']) {
        return Ok(path_str.into());
    }
    let mut stripped_segments = vec![];
    let mut path = path_str.chars().peekable();
    let mut segment = Vec::new();

    while let Some(chr) = path.next() {
        match chr {
            ':' => {
                if path.next_if_eq(&':').is_some() {
                    let stripped_segment =
                        strip_generics_from_path_segment(mem::take(&mut segment))?;
                    if !stripped_segment.is_empty() {
                        stripped_segments.push(stripped_segment);
                    }
                } else {
                    return Err(MalformedGenerics::InvalidPathSeparator);
                }
            }
            '<' => {
                segment.push(chr);

                match path.next() {
                    Some('<') => {
                        return Err(MalformedGenerics::TooManyAngleBrackets);
                    }
                    Some('>') => {
                        return Err(MalformedGenerics::EmptyAngleBrackets);
                    }
                    Some(chr) => {
                        segment.push(chr);

                        while let Some(chr) = path.next_if(|c| *c != '>') {
                            segment.push(chr);
                        }
                    }
                    None => break,
                }
            }
            _ => segment.push(chr),
        }
        trace!("raw segment: {:?}", segment);
    }

    if !segment.is_empty() {
        let stripped_segment = strip_generics_from_path_segment(segment)?;
        if !stripped_segment.is_empty() {
            stripped_segments.push(stripped_segment);
        }
    }

    debug!("path_str: {path_str:?}\nstripped segments: {stripped_segments:?}");

    if !stripped_segments.is_empty() {
        let stripped_path = join_path_syms(stripped_segments);
        Ok(stripped_path.into())
    } else {
        Err(MalformedGenerics::MissingType)
    }
}

/// Has `#[rustc_doc_primitive]` or `#[doc(keyword)]` or `#[doc(attribute)]`.
pub fn has_primitive_or_keyword_or_attribute_docs(attrs: &[impl AttributeExt]) -> bool {
    for attr in attrs {
        if attr.is_rustc_doc_primitive() || attr.is_doc_keyword_or_attribute() {
            return true;
        }
    }
    false
}

/// Simplified version of the corresponding function in rustdoc.
fn preprocess_link(link: &str) -> Box<str> {
    // IMPORTANT: To be kept in sync with the corresponding function in rustdoc.
    // Namely, whenever the rustdoc function returns a successful result for a given input,
    // this function *MUST* return a link that's equal to `PreprocessingInfo.path_str`!

    let link = link.replace('`', "");
    let link = link.split('#').next().unwrap();
    let link = link.trim();
    let link = link.split_once('@').map_or(link, |(_, rhs)| rhs);
    let link = link.trim_suffix("()");
    let link = link.trim_suffix("{}");
    let link = link.trim_suffix("[]");
    let link = if link != "!" { link.trim_suffix('!') } else { link };
    let link = link.trim();
    strip_generics_from_path(link).unwrap_or_else(|_| link.into())
}

/// Keep inline and reference links `[]`,
/// but skip autolinks `<>` which we never consider to be intra-doc links.
pub fn may_be_doc_link(link_type: LinkType) -> bool {
    match link_type {
        LinkType::Inline
        | LinkType::Reference
        | LinkType::ReferenceUnknown
        | LinkType::Collapsed
        | LinkType::CollapsedUnknown
        | LinkType::Shortcut
        | LinkType::ShortcutUnknown => true,
        LinkType::Autolink | LinkType::Email => false,
    }
}

/// Simplified version of `preprocessed_markdown_links` from rustdoc.
/// Must return at least the same links as it, but may add some more links on top of that.
pub(crate) fn attrs_to_preprocessed_links(attrs: &[ast::Attribute]) -> Vec<Box<str>> {
    let (doc_fragments, other_attrs) =
        attrs_to_doc_fragments(attrs.iter().map(|attr| (attr, None)), false);
    let doc = prepare_to_doc_link_resolution(&doc_fragments).into_values().next();
    let mut links = doc
        .as_ref()
        .map(|doc| parse_links(&format!("{}{}", doc.outer, doc.inner)))
        .unwrap_or_default();

    for attr in other_attrs {
        if let Some(note) = attr.deprecation_note() {
            links.extend(parse_links(note.as_str()));
        }
    }

    links
}

/// Similar version of `markdown_links` from rustdoc.
/// This will collect destination links and display text if exists.
fn parse_links<'md>(doc: &'md str) -> Vec<Box<str>> {
    let mut broken_link_callback = |link: BrokenLink<'md>| Some((link.reference, "".into()));
    let mut event_iter = Parser::new_with_broken_link_callback(
        doc,
        main_body_opts(),
        Some(&mut broken_link_callback),
    );
    let mut links = Vec::new();

    let mut refids = UnordSet::default();

    while let Some(event) = event_iter.next() {
        match event {
            Event::Start(Tag::Link { link_type, dest_url, title: _, id })
                if may_be_doc_link(link_type) =>
            {
                if matches!(
                    link_type,
                    LinkType::Inline
                        | LinkType::ReferenceUnknown
                        | LinkType::Reference
                        | LinkType::Shortcut
                        | LinkType::ShortcutUnknown
                ) {
                    if let Some(display_text) = collect_link_data(&mut event_iter) {
                        links.push(display_text);
                    }
                }
                if matches!(
                    link_type,
                    LinkType::Reference | LinkType::Shortcut | LinkType::Collapsed
                ) {
                    refids.insert(id);
                }

                links.push(preprocess_link(&dest_url));
            }
            _ => {}
        }
    }

    for (label, refdef) in event_iter.reference_definitions().iter().sorted_by_key(|x| x.0) {
        if !refids.contains(label) {
            links.push(preprocess_link(&refdef.dest));
        }
    }

    links
}

/// Collects additional data of link.
fn collect_link_data<'input, F: BrokenLinkCallback<'input>>(
    event_iter: &mut Parser<'input, F>,
) -> Option<Box<str>> {
    let mut display_text: Option<String> = None;
    let mut append_text = |text: CowStr<'_>| {
        if let Some(display_text) = &mut display_text {
            display_text.push_str(&text);
        } else {
            display_text = Some(text.to_string());
        }
    };

    while let Some(event) = event_iter.next() {
        match event {
            Event::Text(text) => {
                append_text(text);
            }
            Event::Code(code) => {
                append_text(code);
            }
            Event::End(_) => {
                break;
            }
            _ => {}
        }
    }

    display_text.map(String::into_boxed_str)
}

/// Returns a span encompassing all the document fragments.
pub fn span_of_fragments(fragments: &[DocFragment]) -> Option<Span> {
    let (first_fragment, last_fragment) = match fragments {
        [] => return None,
        [first, .., last] => (first, last),
        [first] => (first, first),
    };
    if first_fragment.span == DUMMY_SP {
        return None;
    }
    Some(first_fragment.span.to(last_fragment.span))
}

/// Attempts to match a range of bytes from parsed markdown to a `Span` in the source code.
///
/// This method does not always work, because markdown bytes don't necessarily match source bytes,
/// like if escapes are used in the string. In this case, it returns `None`.
///
/// `markdown` is typically the entire documentation for an item,
/// after combining fragments.
///
/// This method will return `Some` only if one of the following is true:
///
/// - The doc is made entirely from sugared doc comments, which cannot contain escapes
/// - The doc is entirely from a single doc fragment with a string literal exactly equal to
///   `markdown`.
/// - The doc comes from `include_str!`
/// - The doc includes exactly one substring matching `markdown[md_range]` which is contained in a
///   single doc fragment.
///
/// This function is defined in the compiler so it can be used by both `rustdoc` and `clippy`.
///
/// It returns a tuple containing a span encompassing all the document fragments and a boolean that
/// is `true` if any of the *matched* fragments are from a macro expansion.
pub fn source_span_for_markdown_range(
    tcx: TyCtxt<'_>,
    markdown: &str,
    md_range: &Range<usize>,
    fragments: &[DocFragment],
) -> Option<(Span, bool)> {
    let map = tcx.sess.source_map();
    source_span_for_markdown_range_inner(map, markdown, md_range, fragments)
}

// inner function used for unit testing
pub fn source_span_for_markdown_range_inner(
    map: &SourceMap,
    markdown: &str,
    md_range: &Range<usize>,
    fragments: &[DocFragment],
) -> Option<(Span, bool)> {
    use rustc_span::BytePos;

    if let &[fragment] = &fragments
        && !fragment.kind.is_sugared()
        && let Ok(snippet) = map.span_to_snippet(fragment.span)
        && snippet.trim_end() == markdown.trim_end()
        && let Ok(md_range_lo) = u32::try_from(md_range.start)
        && let Ok(md_range_hi) = u32::try_from(md_range.end)
    {
        // Single fragment with string that contains same bytes as doc.
        return Some((
            Span::new(
                fragment.span.lo() + rustc_span::BytePos(md_range_lo),
                fragment.span.lo() + rustc_span::BytePos(md_range_hi),
                fragment.span.ctxt(),
                fragment.span.parent(),
            ),
            fragment.from_expansion,
        ));
    }

    let is_all_sugared_doc = fragments.iter().all(|frag| frag.kind.is_sugared());

    if !is_all_sugared_doc {
        // This case ignores the markdown outside of the range so that it can
        // work in cases where the markdown is made from several different
        // doc fragments, but the target range does not span across multiple
        // fragments.
        let mut match_data = None;
        let pat = &markdown[md_range.clone()];
        // This heirustic doesn't make sense with a zero-sized range.
        if pat.is_empty() {
            return None;
        }
        for (i, fragment) in fragments.iter().enumerate() {
            if let Ok(snippet) = map.span_to_snippet(fragment.span)
                && let Some(match_start) = snippet.find(pat)
            {
                // If there is either a match in a previous fragment, or
                // multiple matches in this fragment, there is ambiguity.
                // the snippet cannot be zero-sized, because it matches
                // the pattern, which is checked to not be zero sized.
                if match_data.is_none()
                    && !snippet.as_bytes()[match_start + 1..]
                        .windows(pat.len())
                        .any(|s| s == pat.as_bytes())
                {
                    match_data = Some((i, match_start));
                } else {
                    // Heuristic produced ambiguity, return nothing.
                    return None;
                }
            }
        }
        if let Some((i, match_start)) = match_data {
            let fragment = &fragments[i];
            let sp = fragment.span;
            // we need to calculate the span start,
            // then use that in our calculations for the span end
            let lo = sp.lo() + BytePos(match_start as u32);
            return Some((
                sp.with_lo(lo).with_hi(lo + BytePos((md_range.end - md_range.start) as u32)),
                fragment.from_expansion,
            ));
        }
        return None;
    }

    let snippet = map.span_to_snippet(span_of_fragments(fragments)?).ok()?;

    let starting_line = markdown[..md_range.start].matches('\n').count();
    let ending_line = starting_line + markdown[md_range.start..md_range.end].matches('\n').count();

    // We use `split_terminator('\n')` instead of `lines()` when counting bytes so that we treat
    // CRLF and LF line endings the same way.
    let mut src_lines = snippet.split_terminator('\n');
    let md_lines = markdown.split_terminator('\n');

    // The number of bytes from the source span to the markdown span that are not part
    // of the markdown, like comment markers.
    let mut start_bytes = 0;
    let mut end_bytes = 0;

    let span_of_all_fragments: Span = span_of_fragments(fragments)?;

    let mut prev_lines_bytes = 0;
    'outer: for (line_no, md_line) in md_lines.enumerate() {
        loop {
            let source_line = src_lines.next()?;
            let source_line_len = u32::try_from(source_line.len()).unwrap();
            let source_line_span =
                span_of_all_fragments.split_at(prev_lines_bytes).1.split_at(source_line_len).0;
            let fragment = fragments
                .iter()
                // `source_line_span` might contain indentation that `fragment.span` doesn't contain
                .find(|fragment| fragment.span.overlaps(source_line_span));
            // Since we're counting bytes, `prev_line_bytes` includes the "\n".
            prev_lines_bytes += source_line_len + 1;
            if let Some(fragment) = fragment
                && let Some(offset) = source_line.find(md_line)
            {
                if fragment.span.lo() > source_line_span.lo()
                    && source_line
                        [..usize::try_from(fragment.span.lo().0 - source_line_span.lo().0).unwrap()]
                        .chars()
                        .any(|c| !c.is_whitespace())
                {
                    // Make sure anything between the start of this line and the fragment itself is just indentation.
                    // Because source_line is built by splitting the span that covers all fragments, this only finds
                    // characters *between* doc comments, not characters before or after doc comments.
                    //
                    //     1| /** doc */
                    //     2| #[inline] /** doc2 */
                    //        ^^^^^^^^^
                    //        | this
                    //
                    //     3| fn foo() {}
                    return None;
                }
                if fragment.span.hi() < source_line_span.hi()
                    && source_line
                        [usize::try_from(fragment.span.hi().0 - source_line_span.lo().0).unwrap()..]
                        .chars()
                        .any(|c| !c.is_whitespace())
                {
                    // Make sure anything between the start of this line and the fragment itself is just indentation.
                    // Because source_line is built by splitting the span that covers all fragments, this only finds
                    // characters *between* doc comments, not characters before or after doc comments.
                    //     1| /** doc */ #[inline]
                    //                   ^^^^^^^^^
                    //                   | this
                    //
                    //     2| /** doc2 */
                    //     3| fn foo() {}
                    return None;
                }
                if line_no == starting_line {
                    start_bytes += offset;

                    if starting_line == ending_line {
                        break 'outer;
                    }
                } else if line_no == ending_line {
                    end_bytes += offset;
                    break 'outer;
                } else if line_no < starting_line {
                    start_bytes += source_line.len() - md_line.len();
                } else {
                    end_bytes += source_line.len() - md_line.len();
                }
                break;
            } else {
                // Since this is a source line that doesn't include a markdown line,
                // we have to count it and its newline as non-markdown bytes.
                if line_no <= starting_line {
                    start_bytes += source_line.len() + 1;
                } else if source_line.chars().any(|c| !c.is_whitespace()) {
                    // We're past the first line, but haven't found the last line,
                    // but we found a non-empty non-markdown line.
                    // This could be an attribute, and we don't want a diagnostic
                    // suggesting to delete that attribute, so we return None to be safe.
                    //     1| /** doc */
                    //     2 | #[inline]
                    //          ^^^^^^^^^
                    //          | this
                    //     3| /** doc2 */
                    //     4| fn foo() {}
                    return None;
                } else {
                    end_bytes += source_line.len() + 1;
                }
            }
        }
    }

    let src_span = span_of_all_fragments.from_inner(InnerSpan::new(
        md_range.start + start_bytes,
        md_range.end + start_bytes + end_bytes,
    ));
    Some((
        src_span,
        fragments.iter().any(|frag| frag.span.overlaps(src_span) && frag.from_expansion),
    ))
}