1 files changed, 216 insertions, 45 deletions

diff --git a/crates/ide/src/syntax_highlighting/inject.rs b/crates/ide/src/syntax_highlighting/inject.rs
index 4f825523c..947cc974c 100644
--- a/crates/ide/src/syntax_highlighting/inject.rs
+++ b/crates/ide/src/syntax_highlighting/inject.rs
@@ -1,10 +1,18 @@
 //! "Recursive" Syntax highlighting for code in doctests and fixtures.
 
-use hir::Semantics;
-use ide_db::call_info::ActiveParameter;
-use syntax::{ast, AstToken, SyntaxNode, SyntaxToken, TextRange, TextSize};
+use std::{mem, ops::Range};
 
-use crate::{Analysis, HlMod, HlRange, HlTag, RootDatabase};
+use either::Either;
+use hir::{HasAttrs, Semantics};
+use ide_db::{call_info::ActiveParameter, defs::Definition};
+use syntax::{
+    ast::{self, AstNode, AttrsOwner, DocCommentsOwner},
+    match_ast, AstToken, NodeOrToken, SyntaxNode, SyntaxToken, TextRange, TextSize,
+};
+
+use crate::{
+    doc_links::extract_definitions_from_markdown, Analysis, HlMod, HlRange, HlTag, RootDatabase,
+};
 
 use super::{highlights::Highlights, injector::Injector};
 
@@ -81,70 +89,181 @@ const RUSTDOC_FENCE_TOKENS: &[&'static str] = &[
     "edition2021",
 ];
 
-/// Injection of syntax highlighting of doctests.
-pub(super) fn doc_comment(hl: &mut Highlights, node: &SyntaxNode) {
-    let doc_comments = node
-        .children_with_tokens()
-        .filter_map(|it| it.into_token().and_then(ast::Comment::cast))
-        .filter(|it| it.kind().doc.is_some());
+// Basically an owned dyn AttrsOwner without extra Boxing
+struct AttrsOwnerNode {
+    node: SyntaxNode,
+}
+
+impl AttrsOwnerNode {
+    fn new<N: DocCommentsOwner>(node: N) -> Self {
+        AttrsOwnerNode { node: node.syntax().clone() }
+    }
+}
+
+impl AttrsOwner for AttrsOwnerNode {}
+impl AstNode for AttrsOwnerNode {
+    fn can_cast(_: syntax::SyntaxKind) -> bool
+    where
+        Self: Sized,
+    {
+        false
+    }
+    fn cast(_: SyntaxNode) -> Option<Self>
+    where
+        Self: Sized,
+    {
+        None
+    }
+    fn syntax(&self) -> &SyntaxNode {
+        &self.node
+    }
+}
 
-    if !doc_comments.clone().any(|it| it.text().contains(RUSTDOC_FENCE)) {
-        return;
+fn doc_attributes<'node>(
+    sema: &Semantics<RootDatabase>,
+    node: &'node SyntaxNode,
+) -> Option<(AttrsOwnerNode, hir::Attrs, Definition)> {
+    match_ast! {
+        match node {
+            ast::SourceFile(it)  => sema.to_def(&it).map(|def| (AttrsOwnerNode::new(it), def.attrs(sema.db), Definition::ModuleDef(hir::ModuleDef::Module(def)))),
+            ast::Module(it)      => sema.to_def(&it).map(|def| (AttrsOwnerNode::new(it), def.attrs(sema.db), Definition::ModuleDef(hir::ModuleDef::Module(def)))),
+            ast::Fn(it)          => sema.to_def(&it).map(|def| (AttrsOwnerNode::new(it), def.attrs(sema.db), Definition::ModuleDef(hir::ModuleDef::Function(def)))),
+            ast::Struct(it)      => sema.to_def(&it).map(|def| (AttrsOwnerNode::new(it), def.attrs(sema.db), Definition::ModuleDef(hir::ModuleDef::Adt(hir::Adt::Struct(def))))),
+            ast::Union(it)       => sema.to_def(&it).map(|def| (AttrsOwnerNode::new(it), def.attrs(sema.db), Definition::ModuleDef(hir::ModuleDef::Adt(hir::Adt::Union(def))))),
+            ast::Enum(it)        => sema.to_def(&it).map(|def| (AttrsOwnerNode::new(it), def.attrs(sema.db), Definition::ModuleDef(hir::ModuleDef::Adt(hir::Adt::Enum(def))))),
+            ast::Variant(it)     => sema.to_def(&it).map(|def| (AttrsOwnerNode::new(it), def.attrs(sema.db), Definition::ModuleDef(hir::ModuleDef::Variant(def)))),
+            ast::Trait(it)       => sema.to_def(&it).map(|def| (AttrsOwnerNode::new(it), def.attrs(sema.db), Definition::ModuleDef(hir::ModuleDef::Trait(def)))),
+            ast::Static(it)      => sema.to_def(&it).map(|def| (AttrsOwnerNode::new(it), def.attrs(sema.db), Definition::ModuleDef(hir::ModuleDef::Static(def)))),
+            ast::Const(it)       => sema.to_def(&it).map(|def| (AttrsOwnerNode::new(it), def.attrs(sema.db), Definition::ModuleDef(hir::ModuleDef::Const(def)))),
+            ast::TypeAlias(it)   => sema.to_def(&it).map(|def| (AttrsOwnerNode::new(it), def.attrs(sema.db), Definition::ModuleDef(hir::ModuleDef::TypeAlias(def)))),
+            ast::Impl(it)        => sema.to_def(&it).map(|def| (AttrsOwnerNode::new(it), def.attrs(sema.db), Definition::SelfType(def))),
+            ast::RecordField(it) => sema.to_def(&it).map(|def| (AttrsOwnerNode::new(it), def.attrs(sema.db), Definition::Field(def))),
+            ast::TupleField(it)  => sema.to_def(&it).map(|def| (AttrsOwnerNode::new(it), def.attrs(sema.db), Definition::Field(def))),
+            ast::MacroRules(it)  => sema.to_def(&it).map(|def| (AttrsOwnerNode::new(it), def.attrs(sema.db), Definition::Macro(def))),
+            // ast::MacroDef(it) => sema.to_def(&it).map(|def| (Box::new(it) as _, def.attrs(sema.db))),
+            // ast::Use(it) => sema.to_def(&it).map(|def| (Box::new(it) as _, def.attrs(sema.db))),
+            _ => return None
+        }
     }
+}
+
+/// Injection of syntax highlighting of doctests.
+pub(super) fn doc_comment(hl: &mut Highlights, sema: &Semantics<RootDatabase>, node: &SyntaxNode) {
+    let (owner, attributes, def) = match doc_attributes(sema, node) {
+        Some(it) => it,
+        None => return,
+    };
 
     let mut inj = Injector::default();
     inj.add_unmapped("fn doctest() {\n");
 
+    let attrs_source_map = attributes.source_map(&owner);
+
     let mut is_codeblock = false;
     let mut is_doctest = false;
 
     // Replace the original, line-spanning comment ranges by new, only comment-prefix
     // spanning comment ranges.
     let mut new_comments = Vec::new();
-    for comment in doc_comments {
-        match comment.text().find(RUSTDOC_FENCE) {
-            Some(idx) => {
-                is_codeblock = !is_codeblock;
-                // Check whether code is rust by inspecting fence guards
-                let guards = &comment.text()[idx + RUSTDOC_FENCE.len()..];
-                let is_rust =
-                    guards.split(',').all(|sub| RUSTDOC_FENCE_TOKENS.contains(&sub.trim()));
-                is_doctest = is_codeblock && is_rust;
-                continue;
+    let mut intra_doc_links = Vec::new();
+    let mut string;
+    for attr in attributes.by_key("doc").attrs() {
+        let src = attrs_source_map.source_of(&attr);
+        let (line, range, prefix) = match &src {
+            Either::Left(it) => {
+                string = match find_doc_string_in_attr(attr, it) {
+                    Some(it) => it,
+                    None => continue,
+                };
+                let text_range = string.syntax().text_range();
+                let text_range = TextRange::new(
+                    text_range.start() + TextSize::from(1),
+                    text_range.end() - TextSize::from(1),
+                );
+                let text = string.text();
+                (&text[1..text.len() - 1], text_range, "")
             }
-            None if !is_doctest => continue,
-            None => (),
-        }
+            Either::Right(comment) => {
+                (comment.text(), comment.syntax().text_range(), comment.prefix())
+            }
+        };
 
-        let line: &str = comment.text();
-        let range = comment.syntax().text_range();
+        let mut pos = TextSize::from(prefix.len() as u32);
+        let mut range_start = range.start();
+        for line in line.split('\n') {
+            let line_len = TextSize::from(line.len() as u32);
+            let prev_range_start = {
+                let next_range_start = range_start + line_len + TextSize::from(1);
+                mem::replace(&mut range_start, next_range_start)
+            };
+            // only first line has the prefix so take it away for future iterations
+            let mut pos = mem::take(&mut pos);
 
-        let mut pos = TextSize::of(comment.prefix());
-        // whitespace after comment is ignored
-        if let Some(ws) = line[pos.into()..].chars().next().filter(|c| c.is_whitespace()) {
-            pos += TextSize::of(ws);
-        }
-        // lines marked with `#` should be ignored in output, we skip the `#` char
-        if let Some(ws) = line[pos.into()..].chars().next().filter(|&c| c == '#') {
-            pos += TextSize::of(ws);
+            match line.find(RUSTDOC_FENCE) {
+                Some(idx) => {
+                    is_codeblock = !is_codeblock;
+                    // Check whether code is rust by inspecting fence guards
+                    let guards = &line[idx + RUSTDOC_FENCE.len()..];
+                    let is_rust =
+                        guards.split(',').all(|sub| RUSTDOC_FENCE_TOKENS.contains(&sub.trim()));
+                    is_doctest = is_codeblock && is_rust;
+                    continue;
+                }
+                None if !is_doctest => {
+                    intra_doc_links.extend(
+                        extract_definitions_from_markdown(line)
+                            .into_iter()
+                            .filter(|(link, ns, _)| {
+                                validate_intra_doc_link(sema.db, &def, link, *ns)
+                            })
+                            .map(|(.., Range { start, end })| {
+                                TextRange::at(
+                                    prev_range_start + TextSize::from(start as u32),
+                                    TextSize::from((end - start) as u32),
+                                )
+                            }),
+                    );
+                    continue;
+                }
+                None => (),
+            }
+
+            // whitespace after comment is ignored
+            if let Some(ws) = line[pos.into()..].chars().next().filter(|c| c.is_whitespace()) {
+                pos += TextSize::of(ws);
+            }
+            // lines marked with `#` should be ignored in output, we skip the `#` char
+            if line[pos.into()..].starts_with('#') {
+                pos += TextSize::of('#');
+            }
+
+            new_comments.push(TextRange::at(prev_range_start, pos));
+            inj.add(&line[pos.into()..], TextRange::new(pos, line_len) + prev_range_start);
+            inj.add_unmapped("\n");
         }
+    }
 
-        new_comments.push(TextRange::at(range.start(), pos));
+    for range in intra_doc_links {
+        hl.add(HlRange {
+            range,
+            highlight: HlTag::IntraDocLink | HlMod::Documentation,
+            binding_hash: None,
+        });
+    }
 
-        inj.add(&line[pos.into()..], TextRange::new(range.start() + pos, range.end()));
-        inj.add_unmapped("\n");
+    if new_comments.is_empty() {
+        return; // no need to run an analysis on an empty file
     }
+
     inj.add_unmapped("\n}");
 
     let (analysis, tmp_file_id) = Analysis::from_single_file(inj.text().to_string());
 
-    for h in analysis.with_db(|db| super::highlight(db, tmp_file_id, None, true)).unwrap() {
-        for r in inj.map_range_up(h.range) {
-            hl.add(HlRange {
-                range: r,
-                highlight: h.highlight | HlMod::Injected,
-                binding_hash: h.binding_hash,
-            });
+    for HlRange { range, highlight, binding_hash } in
+        analysis.with_db(|db| super::highlight(db, tmp_file_id, None, true)).unwrap()
+    {
+        for range in inj.map_range_up(range) {
+            hl.add(HlRange { range, highlight: highlight | HlMod::Injected, binding_hash });
         }
     }
 
@@ -156,3 +275,55 @@ pub(super) fn doc_comment(hl: &mut Highlights, node: &SyntaxNode) {
         });
     }
 }
+
+fn find_doc_string_in_attr(attr: &hir::Attr, it: &ast::Attr) -> Option<ast::String> {
+    match it.literal() {
+        // #[doc = lit]
+        Some(lit) => match lit.kind() {
+            ast::LiteralKind::String(it) => Some(it),
+            _ => None,
+        },
+        // #[cfg_attr(..., doc = "", ...)]
+        None => {
+            // We gotta hunt the string token manually here
+            let text = attr.string_value()?;
+            // FIXME: We just pick the first string literal that has the same text as the doc attribute
+            // This means technically we might highlight the wrong one
+            it.syntax()
+                .descendants_with_tokens()
+                .filter_map(NodeOrToken::into_token)
+                .filter_map(ast::String::cast)
+                .find(|string| {
+                    string.text().get(1..string.text().len() - 1).map_or(false, |it| it == text)
+                })
+        }
+    }
+}
+
+fn validate_intra_doc_link(
+    db: &RootDatabase,
+    def: &Definition,
+    link: &str,
+    ns: Option<hir::Namespace>,
+) -> bool {
+    match def {
+        Definition::ModuleDef(def) => match def {
+            hir::ModuleDef::Module(it) => it.resolve_doc_path(db, &link, ns),
+            hir::ModuleDef::Function(it) => it.resolve_doc_path(db, &link, ns),
+            hir::ModuleDef::Adt(it) => it.resolve_doc_path(db, &link, ns),
+            hir::ModuleDef::Variant(it) => it.resolve_doc_path(db, &link, ns),
+            hir::ModuleDef::Const(it) => it.resolve_doc_path(db, &link, ns),
+            hir::ModuleDef::Static(it) => it.resolve_doc_path(db, &link, ns),
+            hir::ModuleDef::Trait(it) => it.resolve_doc_path(db, &link, ns),
+            hir::ModuleDef::TypeAlias(it) => it.resolve_doc_path(db, &link, ns),
+            hir::ModuleDef::BuiltinType(_) => None,
+        },
+        Definition::Macro(it) => it.resolve_doc_path(db, &link, ns),
+        Definition::Field(it) => it.resolve_doc_path(db, &link, ns),
+        Definition::SelfType(_)
+        | Definition::Local(_)
+        | Definition::GenericParam(_)
+        | Definition::Label(_) => None,
+    }
+    .is_some()
+}