Rollup merge of #108057 - GuillaumeGomez:fix-reexport-attr-merge, r=notriddle

Prevent some attributes from being merged with others on reexports

Final fix for #59368.

As discussed on zulip [here](https://rust-lang.zulipchat.com/#narrow/stream/182449-t-compiler.2Fhelp/topic/Filtering.20sub.20attributes.20in.20ast.3A.3AAttribute), we need to clone the `Attribute` to be able to filter some parts of it. Then we need to go through the attributes to able to only keep what we want (everything except a few attributes in short).

As for the second commit, when I wrote the test, I realized that the code to traverse all reexports one by one to collect all their attributes was not completely working so I fixed the few issues remaining.

r? `@notriddle`

Author: matthiaskrgr

src/librustdoc/clean/mod.rs

@@ -11,6 +11,8 @@ pub(crate) mod types;
 pub(crate) mod utils;
 
 use rustc_ast as ast;
+use rustc_ast::token::{Token, TokenKind};
+use rustc_ast::tokenstream::{TokenStream, TokenTree};
 use rustc_attr as attr;
 use rustc_data_structures::fx::{FxHashMap, FxHashSet, FxIndexMap, FxIndexSet, IndexEntry};
 use rustc_hir as hir;
@@ -2079,8 +2081,8 @@ impl<'hir> hir::intravisit::Visitor<'hir> for OneLevelVisitor<'hir> {
     fn visit_item(&mut self, item: &'hir hir::Item<'hir>) {
         if self.item.is_none()
             && item.ident == self.looking_for
-            && matches!(item.kind, hir::ItemKind::Use(_, _))
-            || item.owner_id.def_id == self.target_def_id
+            && (matches!(item.kind, hir::ItemKind::Use(_, _))
+                || item.owner_id.def_id == self.target_def_id)
         {
             self.item = Some(item);
         }
@@ -2096,34 +2098,149 @@ fn get_all_import_attributes<'hir>(
     tcx: TyCtxt<'hir>,
     target_def_id: LocalDefId,
     attributes: &mut Vec<ast::Attribute>,
+    is_inline: bool,
 ) {
     let hir_map = tcx.hir();
     let mut visitor = OneLevelVisitor::new(hir_map, target_def_id);
     let mut visited = FxHashSet::default();
     // If the item is an import and has at least a path with two parts, we go into it.
-    while let hir::ItemKind::Use(path, _) = item.kind &&
-        path.segments.len() > 1 &&
-        let hir::def::Res::Def(_, def_id) = path.segments[path.segments.len() - 2].res &&
-        visited.insert(def_id)
-    {
-        if let Some(hir::Node::Item(parent_item)) = hir_map.get_if_local(def_id) {
-            // We add the attributes from this import into the list.
-            attributes.extend_from_slice(hir_map.attrs(item.hir_id()));
-            // We get the `Ident` we will be looking for into `item`.
-            let looking_for = path.segments[path.segments.len() - 1].ident;
-            visitor.reset(looking_for);
-            hir::intravisit::walk_item(&mut visitor, parent_item);
-            if let Some(i) = visitor.item {
-                item = i;
-            } else {
-                break;
+    while let hir::ItemKind::Use(path, _) = item.kind && visited.insert(item.hir_id()) {
+        // We add the attributes from this import into the list.
+        add_without_unwanted_attributes(attributes, hir_map.attrs(item.hir_id()), is_inline);
+
+        let def_id = if path.segments.len() > 1 {
+            match path.segments[path.segments.len() - 2].res {
+                hir::def::Res::Def(_, def_id) => def_id,
+                _ => break,
+            }
+        } else {
+            // If the path doesn't have a parent, then the parent is the current module.
+            tcx.parent(item.owner_id.def_id.to_def_id())
+        };
+
+        let Some(parent) = hir_map.get_if_local(def_id) else { break };
+
+        // We get the `Ident` we will be looking for into `item`.
+        let looking_for = path.segments[path.segments.len() - 1].ident;
+        visitor.reset(looking_for);
+
+        match parent {
+            hir::Node::Item(parent_item) => {
+                hir::intravisit::walk_item(&mut visitor, parent_item);
+            }
+            hir::Node::Crate(m) => {
+                hir::intravisit::walk_mod(
+                    &mut visitor,
+                    m,
+                    tcx.local_def_id_to_hir_id(def_id.as_local().unwrap()),
+                );
             }
+            _ => break,
+        }
+        if let Some(i) = visitor.item {
+            item = i;
         } else {
             break;
         }
     }
 }
 
+fn filter_tokens_from_list(
+    args_tokens: TokenStream,
+    should_retain: impl Fn(&TokenTree) -> bool,
+) -> Vec<TokenTree> {
+    let mut tokens = Vec::with_capacity(args_tokens.len());
+    let mut skip_next_comma = false;
+    for token in args_tokens.into_trees() {
+        match token {
+            TokenTree::Token(Token { kind: TokenKind::Comma, .. }, _) if skip_next_comma => {
+                skip_next_comma = false;
+            }
+            token if should_retain(&token) => {
+                skip_next_comma = false;
+                tokens.push(token);
+            }
+            _ => {
+                skip_next_comma = true;
+            }
+        }
+    }
+    tokens
+}
+
+/// When inlining items, we merge its attributes (and all the reexports attributes too) with the
+/// final reexport. For example:
+///
+/// ```ignore (just an example)
+/// #[doc(hidden, cfg(feature = "foo"))]
+/// pub struct Foo;
+///
+/// #[doc(cfg(feature = "bar"))]
+/// #[doc(hidden, no_inline)]
+/// pub use Foo as Foo1;
+///
+/// #[doc(inline)]
+/// pub use Foo2 as Bar;
+/// ```
+///
+/// So `Bar` at the end will have both `cfg(feature = "...")`. However, we don't want to merge all
+/// attributes so we filter out the following ones:
+/// * `doc(inline)`
+/// * `doc(no_inline)`
+/// * `doc(hidden)`
+fn add_without_unwanted_attributes(
+    attrs: &mut Vec<ast::Attribute>,
+    new_attrs: &[ast::Attribute],
+    is_inline: bool,
+) {
+    // If it's `#[doc(inline)]`, we don't want all attributes, otherwise we keep everything.
+    if !is_inline {
+        attrs.extend_from_slice(new_attrs);
+        return;
+    }
+    for attr in new_attrs {
+        let mut attr = attr.clone();
+        match attr.kind {
+            ast::AttrKind::Normal(ref mut normal) => {
+                if let [ident] = &*normal.item.path.segments &&
+                    let ident = ident.ident.name &&
+                    ident == sym::doc
+                {
+                    match normal.item.args {
+                        ast::AttrArgs::Delimited(ref mut args) => {
+                            let tokens =
+                                filter_tokens_from_list(args.tokens.clone(), |token| {
+                                    !matches!(
+                                        token,
+                                        TokenTree::Token(
+                                            Token {
+                                                kind: TokenKind::Ident(
+                                                    sym::hidden | sym::inline | sym::no_inline,
+                                                    _,
+                                                ),
+                                                ..
+                                            },
+                                            _,
+                                        ),
+                                    )
+                                });
+                            args.tokens = TokenStream::new(tokens);
+                            attrs.push(attr);
+                        }
+                        ast::AttrArgs::Empty | ast::AttrArgs::Eq(..) => {
+                            attrs.push(attr);
+                            continue;
+                        }
+                    }
+                }
+            }
+            ast::AttrKind::DocComment(..) => {
+                attrs.push(attr);
+            }
+        }
+    }
+}
+
 fn clean_maybe_renamed_item<'tcx>(
     cx: &mut DocContext<'tcx>,
     item: &hir::Item<'tcx>,
@@ -2212,19 +2329,20 @@ fn clean_maybe_renamed_item<'tcx>(
         {
             // First, we add the attributes from the current import.
             extra_attrs.extend_from_slice(inline::load_attrs(cx, import_id.to_def_id()));
+            let is_inline = extra_attrs.lists(sym::doc).get_word_attr(sym::inline).is_some();
             // Then we get all the various imports' attributes.
-            get_all_import_attributes(use_node, cx.tcx, item.owner_id.def_id, &mut extra_attrs);
+            get_all_import_attributes(use_node, cx.tcx, item.owner_id.def_id, &mut extra_attrs, is_inline);
+            add_without_unwanted_attributes(&mut extra_attrs, inline::load_attrs(cx, def_id), is_inline);
+        } else {
+            // We only keep the item's attributes.
+            extra_attrs.extend_from_slice(inline::load_attrs(cx, def_id));
         }
 
-        let mut item = if !extra_attrs.is_empty() {
-            extra_attrs.extend_from_slice(inline::load_attrs(cx, def_id));
-            let attrs = Attributes::from_ast(&extra_attrs);
-            let cfg = extra_attrs.cfg(cx.tcx, &cx.cache.hidden_cfg);
+        let attrs = Attributes::from_ast(&extra_attrs);
+        let cfg = extra_attrs.cfg(cx.tcx, &cx.cache.hidden_cfg);
 
-            Item::from_def_id_and_attrs_and_parts(def_id, Some(name), kind, Box::new(attrs), cfg)
-        } else {
-            Item::from_def_id_and_parts(def_id, Some(name), kind, cx)
-        };
+        let mut item =
+            Item::from_def_id_and_attrs_and_parts(def_id, Some(name), kind, Box::new(attrs), cfg);
         item.inline_stmt_id = import_id.map(|def_id| def_id.to_def_id());
         vec![item]
     })

tests/rustdoc/reexport-attr-merge.rs

@@ -0,0 +1,33 @@
+// Regression test for <https://github.com/rust-lang/rust/issues/59368>.
+// The goal is to ensure that `doc(hidden)`, `doc(inline)` and `doc(no_inline)`
+// are not copied from an item when inlined.
+
+#![crate_name = "foo"]
+#![feature(doc_cfg)]
+
+// @has 'foo/index.html'
+
+#[doc(hidden, cfg(feature = "foo"))]
+pub struct Foo;
+
+#[doc(hidden, no_inline, cfg(feature = "bar"))]
+pub use Foo as Foo1;
+
+#[doc(hidden, inline)]
+pub use Foo1 as Foo2;
+
+// First we ensure that only the reexport `Bar2` and the inlined struct `Bar`
+// are inlined.
+// @count - '//a[@class="struct"]' 2
+// Then we check that both `cfg` are displayed.
+// @has - '//*[@class="stab portability"]' 'foo'
+// @has - '//*[@class="stab portability"]' 'bar'
+// And finally we check that the only element displayed is `Bar`.
+// @has - '//a[@class="struct"]' 'Bar'
+#[doc(inline)]
+pub use Foo2 as Bar;
+
+// This one should appear but `Bar2` won't be linked because there is no
+// `#[doc(inline)]`.
+// @has - '//*[@id="reexport.Bar2"]' 'pub use Foo2 as Bar2;'
+pub use Foo2 as Bar2;