Auto merge of #141437 - matthiaskrgr:rollup-jjagkxd, r=matthiaskrgr

Rollup of 7 pull requests

Successful merges:

 - #136400 (Improve handling of rustdoc lints when used with raw doc fragments.)
 - #140967 (Async drop poll shim for error dropee generates noop body)
 - #141019 (Update std doctests for android)
 - #141109 (discuss deadlocks in the std::io::pipe() example)
 - #141126 (rustdoc JSON: Don't apply `#[repr]` privacy heuristics)
 - #141376 (Rename `kw::Empty` as `sym::empty`.)
 - #141383 (Miri subtree update)

r? `@ghost`
`@rustbot` modify labels: rollup

Author: bors

Cargo.lock

@@ -3116,9 +3116,9 @@ dependencies = [
 
 [[package]]
 name = "rustc-build-sysroot"
-version = "0.5.5"
+version = "0.5.7"
 source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "fb332121f7845c6bd016f9655cf22f03c2999df936694b624a88669a78667d98"
+checksum = "10edc2e4393515193bd766e2f6c050b0536a68e56f2b6d56c07ababfdc114ff0"
 dependencies = [
  "anyhow",
  "rustc_version",

compiler/rustc_ast_lowering/src/format.rs

@@ -6,7 +6,7 @@ use rustc_ast::*;
 use rustc_data_structures::fx::FxIndexMap;
 use rustc_hir as hir;
 use rustc_session::config::FmtDebug;
-use rustc_span::{Ident, Span, Symbol, kw, sym};
+use rustc_span::{Ident, Span, Symbol, sym};
 
 use super::LoweringContext;
 
@@ -418,7 +418,7 @@ fn expand_format_args<'hir>(
                 &FormatArgsPiece::Placeholder(_) => {
                     // Inject empty string before placeholders when not already preceded by a literal piece.
                     if i == 0 || matches!(fmt.template[i - 1], FormatArgsPiece::Placeholder(_)) {
-                        Some(ctx.expr_str(fmt.span, kw::Empty))
+                        Some(ctx.expr_str(fmt.span, sym::empty))
                     } else {
                         None
                     }

compiler/rustc_codegen_ssa/src/mir/debuginfo.rs

@@ -10,7 +10,7 @@ use rustc_middle::ty::layout::{LayoutOf, TyAndLayout};
 use rustc_middle::ty::{Instance, Ty};
 use rustc_middle::{bug, mir, ty};
 use rustc_session::config::DebugInfo;
-use rustc_span::{BytePos, Span, Symbol, hygiene, kw};
+use rustc_span::{BytePos, Span, Symbol, hygiene, sym};
 
 use super::operand::{OperandRef, OperandValue};
 use super::place::{PlaceRef, PlaceValue};
@@ -283,7 +283,7 @@ impl<'a, 'tcx, Bx: BuilderMethods<'a, 'tcx>> FunctionCx<'a, 'tcx, Bx> {
                 // (after #67586 gets fixed).
                 None
             } else {
-                let name = kw::Empty;
+                let name = sym::empty;
                 let decl = &self.mir.local_decls[local];
                 let dbg_var = if full_debug_info {
                     self.adjusted_span_and_dbg_scope(decl.source_info).map(
@@ -318,7 +318,7 @@ impl<'a, 'tcx, Bx: BuilderMethods<'a, 'tcx>> FunctionCx<'a, 'tcx, Bx> {
             None
         } else {
             Some(match whole_local_var.or(fallback_var.clone()) {
-                Some(var) if var.name != kw::Empty => var.name.to_string(),
+                Some(var) if var.name != sym::empty => var.name.to_string(),
                 _ => format!("{local:?}"),
             })
         };

compiler/rustc_hir/src/hir.rs

@@ -85,7 +85,7 @@ impl From<Ident> for LifetimeSyntax {
     fn from(ident: Ident) -> Self {
         let name = ident.name;
 
-        if name == kw::Empty {
+        if name == sym::empty {
             unreachable!("A lifetime name should never be empty");
         } else if name == kw::UnderscoreLifetime {
             LifetimeSyntax::Anonymous

compiler/rustc_mir_transform/src/shim/async_destructor_ctor.rs

@@ -6,7 +6,7 @@ use rustc_middle::mir::{
     BasicBlock, BasicBlockData, Body, Local, LocalDecl, MirSource, Operand, Place, Rvalue,
     SourceInfo, Statement, StatementKind, Terminator, TerminatorKind,
 };
-use rustc_middle::ty::{self, EarlyBinder, Ty, TyCtxt};
+use rustc_middle::ty::{self, EarlyBinder, Ty, TyCtxt, TypeVisitableExt};
 
 use super::*;
 use crate::patch::MirPatch;
@@ -121,9 +121,10 @@ pub(super) fn build_async_drop_shim<'tcx>(
         parent_args.as_coroutine().resume_ty(),
     )));
     body.phase = MirPhase::Runtime(RuntimePhase::Initial);
-    if !needs_async_drop {
+    if !needs_async_drop || drop_ty.references_error() {
         // Returning noop body for types without `need async drop`
-        // (or sync Drop in case of !`need async drop` && `need drop`)
+        // (or sync Drop in case of !`need async drop` && `need drop`).
+        // And also for error types.
         return body;
     }
 

compiler/rustc_passes/src/check_attr.rs

@@ -35,7 +35,7 @@ use rustc_session::lint::builtin::{
     UNKNOWN_OR_MALFORMED_DIAGNOSTIC_ATTRIBUTES, UNUSED_ATTRIBUTES,
 };
 use rustc_session::parse::feature_err;
-use rustc_span::{BytePos, DUMMY_SP, Span, Symbol, edition, kw, sym};
+use rustc_span::{BytePos, DUMMY_SP, Span, Symbol, edition, sym};
 use rustc_trait_selection::error_reporting::InferCtxtErrorExt;
 use rustc_trait_selection::infer::{TyCtxtInferExt, ValuePairs};
 use rustc_trait_selection::traits::ObligationCtxt;
@@ -936,7 +936,7 @@ impl<'tcx> CheckAttrVisitor<'tcx> {
         let span = meta.name_value_literal_span().unwrap_or_else(|| meta.span());
         let attr_str =
             &format!("`#[doc(alias{})]`", if is_list { "(\"...\")" } else { " = \"...\"" });
-        if doc_alias == kw::Empty {
+        if doc_alias == sym::empty {
             tcx.dcx().emit_err(errors::DocAliasEmpty { span, attr_str });
             return;
         }
@@ -1068,7 +1068,7 @@ impl<'tcx> CheckAttrVisitor<'tcx> {
         }
 
         let doc_keyword = match meta.value_str() {
-            Some(value) if value != kw::Empty => value,
+            Some(value) if value != sym::empty => value,
             _ => return self.doc_attr_str_error(meta, "keyword"),
         };
 

compiler/rustc_resolve/src/rustdoc.rs

@@ -12,7 +12,7 @@ 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::{DUMMY_SP, InnerSpan, Span, Symbol, kw, sym};
+use rustc_span::{DUMMY_SP, InnerSpan, Span, Symbol, sym};
 use thin_vec::ThinVec;
 use tracing::{debug, trace};
 
@@ -157,7 +157,7 @@ pub fn unindent_doc_fragments(docs: &mut [DocFragment]) {
     };
 
     for fragment in docs {
-        if fragment.doc == kw::Empty {
+        if fragment.doc == sym::empty {
             continue;
         }
 
@@ -177,7 +177,7 @@ pub fn unindent_doc_fragments(docs: &mut [DocFragment]) {
 ///
 /// Note: remove the trailing newline where appropriate
 pub fn add_doc_fragment(out: &mut String, frag: &DocFragment) {
-    if frag.doc == kw::Empty {
+    if frag.doc == sym::empty {
         out.push('\n');
         return;
     }
@@ -514,20 +514,30 @@ pub fn span_of_fragments(fragments: &[DocFragment]) -> Option<Span> {
 /// 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`.
 ///
-/// This method will return `Some` only if:
+/// `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
+/// - 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`.
 pub fn source_span_for_markdown_range(
     tcx: TyCtxt<'_>,
     markdown: &str,
     md_range: &Range<usize>,
     fragments: &[DocFragment],
 ) -> Option<Span> {
+    use rustc_span::BytePos;
+
+    let map = tcx.sess.source_map();
     if let &[fragment] = &fragments
         && fragment.kind == DocFragmentKind::RawDoc
-        && let Ok(snippet) = tcx.sess.source_map().span_to_snippet(fragment.span)
+        && 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)
@@ -544,10 +554,43 @@ pub fn source_span_for_markdown_range(
     let is_all_sugared_doc = fragments.iter().all(|frag| frag.kind == DocFragmentKind::SugaredDoc);
 
     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.
+                if match_data.is_none() && !snippet[match_start + 1..].contains(pat) {
+                    match_data = Some((i, match_start));
+                } else {
+                    // Heirustic produced ambiguity, return nothing.
+                    return None;
+                }
+            }
+        }
+        if let Some((i, match_start)) = match_data {
+            let sp = fragments[i].span;
+            // we need to calculate the span start,
+            // then use that in our calulations 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)),
+            );
+        }
         return None;
     }
 
-    let snippet = tcx.sess.source_map().span_to_snippet(span_of_fragments(fragments)?).ok()?;
+    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();

compiler/rustc_span/src/symbol.rs

@@ -34,17 +34,8 @@ symbols! {
         // unnamed method parameters, crate root module, error recovery etc.
         // Matching predicates: `is_special`/`is_reserved`
         //
-        // Notes about `kw::Empty`:
-        // - Its use can blur the lines between "empty symbol" and "no symbol".
-        //   Using `Option<Symbol>` is preferable, where possible, because that
-        //   is unambiguous.
-        // - For dummy symbols that are never used and absolutely must be
-        //   present, it's better to use `sym::dummy` than `kw::Empty`, because
-        //   it's clearer that it's intended as a dummy value, and more likely
-        //   to be detected if it accidentally does get used.
         // tidy-alphabetical-start
         DollarCrate:        "$crate",
-        Empty:              "",
         PathRoot:           "{{root}}",
         Underscore:         "_",
         // tidy-alphabetical-end
@@ -863,7 +854,7 @@ symbols! {
         drop_types_in_const,
         dropck_eyepatch,
         dropck_parametricity,
-        dummy: "<!dummy!>", // use this instead of `kw::Empty` for symbols that won't be used
+        dummy: "<!dummy!>", // use this instead of `sym::empty` for symbols that won't be used
         dummy_cgu_name,
         dylib,
         dyn_compatible_for_dispatch,
@@ -882,6 +873,14 @@ symbols! {
         emit_enum_variant_arg,
         emit_struct,
         emit_struct_field,
+        // Notes about `sym::empty`:
+        // - It should only be used when it genuinely means "empty symbol". Use
+        //   `Option<Symbol>` when "no symbol" is a possibility.
+        // - For dummy symbols that are never used and absolutely must be
+        //   present, it's better to use `sym::dummy` than `sym::empty`, because
+        //   it's clearer that it's intended as a dummy value, and more likely
+        //   to be detected if it accidentally does get used.
+        empty: "",
         emscripten_wasm_eh,
         enable,
         encode,
@@ -2361,7 +2360,7 @@ impl Ident {
     #[inline]
     /// Constructs a new identifier from a symbol and a span.
     pub fn new(name: Symbol, span: Span) -> Ident {
-        debug_assert_ne!(name, kw::Empty);
+        debug_assert_ne!(name, sym::empty);
         Ident { name, span }
     }
 
@@ -2583,7 +2582,7 @@ impl Symbol {
     }
 
     pub fn is_empty(self) -> bool {
-        self == kw::Empty
+        self == sym::empty
     }
 
     /// This method is supposed to be used in error messages, so it's expected to be
@@ -2592,7 +2591,7 @@ impl Symbol {
     /// or edition, so we have to guess the rawness using the global edition.
     pub fn to_ident_string(self) -> String {
         // Avoid creating an empty identifier, because that asserts in debug builds.
-        if self == kw::Empty { String::new() } else { Ident::with_dummy_span(self).to_string() }
+        if self == sym::empty { String::new() } else { Ident::with_dummy_span(self).to_string() }
     }
 }
 
@@ -2772,7 +2771,7 @@ impl Symbol {
 
     /// Returns `true` if this symbol can be a raw identifier.
     pub fn can_be_raw(self) -> bool {
-        self != kw::Empty && self != kw::Underscore && !self.is_path_segment_keyword()
+        self != sym::empty && self != kw::Underscore && !self.is_path_segment_keyword()
     }
 
     /// Was this symbol predefined in the compiler's `symbols!` macro

compiler/rustc_symbol_mangling/src/v0.rs

@@ -20,7 +20,7 @@ use rustc_middle::ty::{
     self, FloatTy, GenericArg, GenericArgKind, Instance, IntTy, ReifyReason, Ty, TyCtxt,
     TypeVisitable, TypeVisitableExt, UintTy,
 };
-use rustc_span::kw;
+use rustc_span::sym;
 
 pub(super) fn mangle<'tcx>(
     tcx: TyCtxt<'tcx>,
@@ -902,7 +902,7 @@ impl<'tcx> Printer<'tcx> for SymbolMangler<'tcx> {
             print_prefix,
             ns,
             disambiguated_data.disambiguator as u64,
-            name.unwrap_or(kw::Empty).as_str(),
+            name.unwrap_or(sym::empty).as_str(),
         )
     }
 

library/std/src/io/pipe.rs

@@ -38,30 +38,44 @@ use crate::sys_common::{FromInner, IntoInner};
 /// > not rely on a particular capacity: an application should be designed so that a reading process
 /// > consumes data as soon as it is available, so that a writing process does not remain blocked.
 ///
-/// # Examples
+/// # Example
 ///
 /// ```no_run
 /// # #[cfg(miri)] fn main() {}
 /// # #[cfg(not(miri))]
 /// # fn main() -> std::io::Result<()> {
+/// use std::io::{Read, Write, pipe};
 /// use std::process::Command;
-/// use std::io::{pipe, Read, Write};
-/// let (ping_rx, mut ping_tx) = pipe()?;
-/// let (mut pong_rx, pong_tx) = pipe()?;
+/// let (ping_reader, mut ping_writer) = pipe()?;
+/// let (mut pong_reader, pong_writer) = pipe()?;
 ///
-/// // Spawn a process that echoes its input.
-/// let mut echo_server = Command::new("cat").stdin(ping_rx).stdout(pong_tx).spawn()?;
+/// // Spawn a child process that echoes its input.
+/// let mut echo_command = Command::new("cat");
+/// echo_command.stdin(ping_reader);
+/// echo_command.stdout(pong_writer);
+/// let mut echo_child = echo_command.spawn()?;
 ///
-/// ping_tx.write_all(b"hello")?;
-/// // Close to unblock echo_server's reader.
-/// drop(ping_tx);
+/// // Send input to the child process. Note that because we're writing all the input before we
+/// // read any output, this could deadlock if the child's input and output pipe buffers both
+/// // filled up. Those buffers are usually at least a few KB, so "hello" is fine, but for longer
+/// // inputs we'd need to read and write at the same time, e.g. using threads.
+/// ping_writer.write_all(b"hello")?;
+///
+/// // `cat` exits when it reads EOF from stdin, but that can't happen while any ping writer
+/// // remains open. We need to drop our ping writer, or read_to_string will deadlock below.
+/// drop(ping_writer);
+///
+/// // The pong reader can't report EOF while any pong writer remains open. Our Command object is
+/// // holding a pong writer, and again read_to_string will deadlock if we don't drop it.
+/// drop(echo_command);
 ///
 /// let mut buf = String::new();
-/// // Block until echo_server's writer is closed.
-/// pong_rx.read_to_string(&mut buf)?;
+/// // Block until `cat` closes its stdout (a pong writer).
+/// pong_reader.read_to_string(&mut buf)?;
 /// assert_eq!(&buf, "hello");
 ///
-/// echo_server.wait()?;
+/// // At this point we know `cat` has exited, but we still need to wait to clean up the "zombie".
+/// echo_child.wait()?;
 /// # Ok(())
 /// # }
 /// ```