Auto merge of #3627 - rust-lang:rustup-2024-05-24, r=RalfJung

Automatic Rustup

Author: bors

compiler/rustc_ast/src/tokenstream.rs

@@ -661,11 +661,11 @@ impl TokenStream {
             if attr_style == AttrStyle::Inner {
                 vec![
                     TokenTree::token_joint(token::Pound, span),
-                    TokenTree::token_alone(token::Not, span),
+                    TokenTree::token_joint_hidden(token::Not, span),
                     body,
                 ]
             } else {
-                vec![TokenTree::token_alone(token::Pound, span), body]
+                vec![TokenTree::token_joint_hidden(token::Pound, span), body]
             }
         }
     }

compiler/rustc_ast_pretty/src/pprust/state.rs

@@ -681,22 +681,40 @@ pub trait PrintState<'a>: std::ops::Deref<Target = pp::Printer> + std::ops::Dere
         }
     }
 
+    // The easiest way to implement token stream pretty printing would be to
+    // print each token followed by a single space. But that would produce ugly
+    // output, so we go to some effort to do better.
+    //
+    // First, we track whether each token that appears in source code is
+    // followed by a space, with `Spacing`, and reproduce that in the output.
+    // This works well in a lot of cases. E.g. `stringify!(x + y)` produces
+    // "x + y" and `stringify!(x+y)` produces "x+y".
+    //
+    // But this doesn't work for code produced by proc macros (which have no
+    // original source text representation) nor for code produced by decl
+    // macros (which are tricky because the whitespace after tokens appearing
+    // in macro rules isn't always what you want in the produced output). For
+    // these we mostly use `Spacing::Alone`, which is the conservative choice.
+    //
+    // So we have a backup mechanism for when `Spacing::Alone` occurs between a
+    // pair of tokens: we check if that pair of tokens can obviously go
+    // together without a space between them. E.g. token `x` followed by token
+    // `,` is better printed as `x,` than `x ,`. (Even if the original source
+    // code was `x ,`.)
+    //
+    // Finally, we must be careful about changing the output. Token pretty
+    // printing is used by `stringify!` and `impl Display for
+    // proc_macro::TokenStream`, and some programs rely on the output having a
+    // particular form, even though they shouldn't. In particular, some proc
+    // macros do `format!({stream})` on a token stream and then "parse" the
+    // output with simple string matching that can't handle whitespace changes.
+    // E.g. we have seen cases where a proc macro can handle `a :: b` but not
+    // `a::b`. See #117433 for some examples.
     fn print_tts(&mut self, tts: &TokenStream, convert_dollar_crate: bool) {
         let mut iter = tts.trees().peekable();
         while let Some(tt) = iter.next() {
             let spacing = self.print_tt(tt, convert_dollar_crate);
             if let Some(next) = iter.peek() {
-                // Should we print a space after `tt`? There are two guiding
-                // factors.
-                // - `spacing` is the more important and accurate one. Most
-                //   tokens have good spacing information, and
-                //   `Joint`/`JointHidden` get used a lot.
-                // - `space_between` is the backup. Code produced by proc
-                //   macros has worse spacing information, with no
-                //   `JointHidden` usage and too much `Alone` usage, which
-                //   would result in over-spaced output such as
-                //   `( x () , y . z )`. `space_between` avoids some of the
-                //   excess whitespace.
                 if spacing == Spacing::Alone && space_between(tt, next) {
                     self.space();
                 }

compiler/rustc_builtin_macros/src/assert/context.rs

@@ -153,7 +153,7 @@ impl<'cx, 'a> Context<'cx, 'a> {
     fn build_panic(&self, expr_str: &str, panic_path: Path) -> P<Expr> {
         let escaped_expr_str = escape_to_fmt(expr_str);
         let initial = [
-            TokenTree::token_joint_hidden(
+            TokenTree::token_joint(
                 token::Literal(token::Lit {
                     kind: token::LitKind::Str,
                     symbol: Symbol::intern(&if self.fmt_string.is_empty() {
@@ -172,7 +172,7 @@ impl<'cx, 'a> Context<'cx, 'a> {
         ];
         let captures = self.capture_decls.iter().flat_map(|cap| {
             [
-                TokenTree::token_joint_hidden(
+                TokenTree::token_joint(
                     token::Ident(cap.ident.name, IdentIsRaw::No),
                     cap.ident.span,
                 ),

compiler/rustc_builtin_macros/src/deriving/default.rs

@@ -3,7 +3,7 @@ use crate::deriving::generic::*;
 use crate::errors;
 use core::ops::ControlFlow;
 use rustc_ast as ast;
-use rustc_ast::visit::walk_list;
+use rustc_ast::visit::visit_opt;
 use rustc_ast::{attr, EnumDef, VariantData};
 use rustc_expand::base::{Annotatable, DummyResult, ExtCtxt};
 use rustc_span::symbol::Ident;
@@ -224,7 +224,7 @@ impl<'a, 'b> rustc_ast::visit::Visitor<'a> for DetectNonVariantDefaultAttr<'a, '
         self.visit_ident(v.ident);
         self.visit_vis(&v.vis);
         self.visit_variant_data(&v.data);
-        walk_list!(self, visit_anon_const, &v.disr_expr);
+        visit_opt!(self, visit_anon_const, &v.disr_expr);
         for attr in &v.attrs {
             rustc_ast::visit::walk_attribute(self, attr);
         }

compiler/rustc_codegen_cranelift/src/driver/aot.rs

@@ -200,7 +200,7 @@ fn produce_final_output_artifacts(
     // to get rid of it.
     for output_type in crate_output.outputs.keys() {
         match *output_type {
-            OutputType::Bitcode => {
+            OutputType::Bitcode | OutputType::ThinLinkBitcode => {
                 // Cranelift doesn't have bitcode
                 // user_wants_bitcode = true;
                 // // Copy to .bc, but always keep the .0.bc. There is a later

compiler/rustc_codegen_gcc/src/lib.rs

@@ -335,6 +335,10 @@ impl ThinBufferMethods for ThinBuffer {
     fn data(&self) -> &[u8] {
         unimplemented!();
     }
+
+    fn thin_link_data(&self) -> &[u8] {
+        unimplemented!();
+    }
 }
 
 pub struct GccContext {
@@ -414,7 +418,7 @@ impl WriteBackendMethods for GccCodegenBackend {
         back::write::codegen(cgcx, dcx, module, config)
     }
 
-    fn prepare_thin(_module: ModuleCodegen<Self::Module>) -> (String, Self::ThinBuffer) {
+    fn prepare_thin(_module: ModuleCodegen<Self::Module>, _emit_summary: bool) -> (String, Self::ThinBuffer) {
         unimplemented!();
     }
 

compiler/rustc_codegen_llvm/src/asm.rs

@@ -16,6 +16,7 @@ use rustc_middle::{bug, span_bug, ty::Instance};
 use rustc_span::{Pos, Span};
 use rustc_target::abi::*;
 use rustc_target::asm::*;
+use tracing::debug;
 
 use libc::{c_char, c_uint};
 use smallvec::SmallVec;

compiler/rustc_codegen_llvm/src/back/archive.rs

@@ -18,6 +18,7 @@ use rustc_codegen_ssa::back::archive::{
     get_native_object_symbols, try_extract_macho_fat_archive, ArArchiveBuilder,
     ArchiveBuildFailure, ArchiveBuilder, ArchiveBuilderBuilder, UnknownArchiveKind,
 };
+use tracing::trace;
 
 use rustc_session::cstore::DllImport;
 use rustc_session::Session;

compiler/rustc_codegen_llvm/src/back/lto.rs

@@ -20,6 +20,7 @@ use rustc_middle::bug;
 use rustc_middle::dep_graph::WorkProduct;
 use rustc_middle::middle::exported_symbols::{SymbolExportInfo, SymbolExportLevel};
 use rustc_session::config::{self, CrateType, Lto};
+use tracing::{debug, info};
 
 use std::collections::BTreeMap;
 use std::ffi::{CStr, CString};
@@ -229,9 +230,12 @@ pub(crate) fn run_thin(
     thin_lto(cgcx, &dcx, modules, upstream_modules, cached_modules, &symbols_below_threshold)
 }
 
-pub(crate) fn prepare_thin(module: ModuleCodegen<ModuleLlvm>) -> (String, ThinBuffer) {
+pub(crate) fn prepare_thin(
+    module: ModuleCodegen<ModuleLlvm>,
+    emit_summary: bool,
+) -> (String, ThinBuffer) {
     let name = module.name;
-    let buffer = ThinBuffer::new(module.module_llvm.llmod(), true);
+    let buffer = ThinBuffer::new(module.module_llvm.llmod(), true, emit_summary);
     (name, buffer)
 }
 
@@ -671,9 +675,9 @@ unsafe impl Send for ThinBuffer {}
 unsafe impl Sync for ThinBuffer {}
 
 impl ThinBuffer {
-    pub fn new(m: &llvm::Module, is_thin: bool) -> ThinBuffer {
+    pub fn new(m: &llvm::Module, is_thin: bool, emit_summary: bool) -> ThinBuffer {
         unsafe {
-            let buffer = llvm::LLVMRustThinLTOBufferCreate(m, is_thin);
+            let buffer = llvm::LLVMRustThinLTOBufferCreate(m, is_thin, emit_summary);
             ThinBuffer(buffer)
         }
     }
@@ -687,6 +691,14 @@ impl ThinBufferMethods for ThinBuffer {
             slice::from_raw_parts(ptr, len)
         }
     }
+
+    fn thin_link_data(&self) -> &[u8] {
+        unsafe {
+            let ptr = llvm::LLVMRustThinLTOBufferThinLinkDataPtr(self.0) as *const _;
+            let len = llvm::LLVMRustThinLTOBufferThinLinkDataLen(self.0);
+            slice::from_raw_parts(ptr, len)
+        }
+    }
 }
 
 impl Drop for ThinBuffer {

compiler/rustc_codegen_llvm/src/back/write.rs

@@ -35,6 +35,7 @@ use rustc_session::Session;
 use rustc_span::symbol::sym;
 use rustc_span::InnerSpan;
 use rustc_target::spec::{CodeModel, RelocModel, SanitizerSet, SplitDebuginfo, TlsModel};
+use tracing::debug;
 
 use crate::llvm::diagnostic::OptimizationDiagnosticKind;
 use libc::{c_char, c_int, c_void, size_t};
@@ -708,13 +709,15 @@ pub(crate) unsafe fn codegen(
         //   asm from LLVM and use `gcc` to create the object file.
 
         let bc_out = cgcx.output_filenames.temp_path(OutputType::Bitcode, module_name);
+        let bc_summary_out =
+            cgcx.output_filenames.temp_path(OutputType::ThinLinkBitcode, module_name);
         let obj_out = cgcx.output_filenames.temp_path(OutputType::Object, module_name);
 
         if config.bitcode_needed() {
             let _timer = cgcx
                 .prof
                 .generic_activity_with_arg("LLVM_module_codegen_make_bitcode", &*module.name);
-            let thin = ThinBuffer::new(llmod, config.emit_thin_lto);
+            let thin = ThinBuffer::new(llmod, config.emit_thin_lto, config.emit_thin_lto_summary);
             let data = thin.data();
 
             if let Some(bitcode_filename) = bc_out.file_name() {
@@ -725,6 +728,25 @@ pub(crate) unsafe fn codegen(
                 );
             }
 
+            if config.emit_thin_lto_summary
+                && let Some(thin_link_bitcode_filename) = bc_summary_out.file_name()
+            {
+                let summary_data = thin.thin_link_data();
+                cgcx.prof.artifact_size(
+                    "llvm_bitcode_summary",
+                    thin_link_bitcode_filename.to_string_lossy(),
+                    summary_data.len() as u64,
+                );
+
+                let _timer = cgcx.prof.generic_activity_with_arg(
+                    "LLVM_module_codegen_emit_bitcode_summary",
+                    &*module.name,
+                );
+                if let Err(err) = fs::write(&bc_summary_out, summary_data) {
+                    dcx.emit_err(WriteBytecode { path: &bc_summary_out, err });
+                }
+            }
+
             if config.emit_bc || config.emit_obj == EmitObj::Bitcode {
                 let _timer = cgcx
                     .prof