diff options
Diffstat (limited to 'compiler')
106 files changed, 2925 insertions, 2124 deletions
diff --git a/compiler/rustc_ast/src/token.rs b/compiler/rustc_ast/src/token.rs index ea98bebd305..6dc6d1026f6 100644 --- a/compiler/rustc_ast/src/token.rs +++ b/compiler/rustc_ast/src/token.rs @@ -22,8 +22,7 @@ pub enum CommentKind { Block, } -// This type must not implement `Hash` due to the unusual `PartialEq` impl below. -#[derive(Copy, Clone, Debug, Encodable, Decodable, HashStable_Generic)] +#[derive(Copy, Clone, PartialEq, Debug, Encodable, Decodable, HashStable_Generic)] pub enum InvisibleOrigin { // From the expansion of a metavariable in a declarative macro. MetaVar(MetaVarKind), @@ -45,20 +44,6 @@ impl InvisibleOrigin { } } -impl PartialEq for InvisibleOrigin { - #[inline] - fn eq(&self, _other: &InvisibleOrigin) -> bool { - // When we had AST-based nonterminals we couldn't compare them, and the - // old `Nonterminal` type had an `eq` that always returned false, - // resulting in this restriction: - // https://doc.rust-lang.org/nightly/reference/macros-by-example.html#forwarding-a-matched-fragment - // This `eq` emulates that behaviour. We could consider lifting this - // restriction now but there are still cases involving invisible - // delimiters that make it harder than it first appears. - false - } -} - /// Annoyingly similar to `NonterminalKind`, but the slight differences are important. #[derive(Debug, Copy, Clone, PartialEq, Eq, Encodable, Decodable, Hash, HashStable_Generic)] pub enum MetaVarKind { @@ -142,7 +127,8 @@ impl Delimiter { } } - // This exists because `InvisibleOrigin`s should be compared. It is only used for assertions. + // This exists because `InvisibleOrigin`s should not be compared. It is only used for + // assertions. pub fn eq_ignoring_invisible_origin(&self, other: &Delimiter) -> bool { match (self, other) { (Delimiter::Parenthesis, Delimiter::Parenthesis) => true, diff --git a/compiler/rustc_attr_parsing/messages.ftl b/compiler/rustc_attr_parsing/messages.ftl index b8a748563d5..7fa1293463c 100644 --- a/compiler/rustc_attr_parsing/messages.ftl +++ b/compiler/rustc_attr_parsing/messages.ftl @@ -151,7 +151,7 @@ attr_parsing_unrecognized_repr_hint = attr_parsing_unstable_cfg_target_compact = compact `cfg(target(..))` is experimental and subject to change -attr_parsing_unstable_feature_bound_incompatible_stability = Item annotated with `#[unstable_feature_bound]` should not be stable +attr_parsing_unstable_feature_bound_incompatible_stability = item annotated with `#[unstable_feature_bound]` should not be stable .help = If this item is meant to be stable, do not use any functions annotated with `#[unstable_feature_bound]`. Otherwise, mark this item as unstable with `#[unstable]` attr_parsing_unsupported_literal_cfg_boolean = diff --git a/compiler/rustc_borrowck/src/region_infer/opaque_types/mod.rs b/compiler/rustc_borrowck/src/region_infer/opaque_types/mod.rs index 72615cb33b3..95c5366a6c8 100644 --- a/compiler/rustc_borrowck/src/region_infer/opaque_types/mod.rs +++ b/compiler/rustc_borrowck/src/region_infer/opaque_types/mod.rs @@ -503,7 +503,16 @@ pub(crate) fn apply_computed_concrete_opaque_types<'tcx>( let mut errors = Vec::new(); for &(key, hidden_type) in opaque_types { let Some(expected) = get_concrete_opaque_type(concrete_opaque_types, key.def_id) else { - assert!(tcx.use_typing_mode_borrowck(), "non-defining use in defining scope"); + if !tcx.use_typing_mode_borrowck() { + if let ty::Alias(ty::Opaque, alias_ty) = hidden_type.ty.kind() + && alias_ty.def_id == key.def_id.to_def_id() + && alias_ty.args == key.args + { + continue; + } else { + unreachable!("non-defining use in defining scope"); + } + } errors.push(DeferredOpaqueTypeError::NonDefiningUseInDefiningScope { span: hidden_type.span, opaque_type_key: key, diff --git a/compiler/rustc_codegen_cranelift/scripts/test_rustc_tests.sh b/compiler/rustc_codegen_cranelift/scripts/test_rustc_tests.sh index 52e02c857c7..62f1cc6a893 100755 --- a/compiler/rustc_codegen_cranelift/scripts/test_rustc_tests.sh +++ b/compiler/rustc_codegen_cranelift/scripts/test_rustc_tests.sh @@ -121,7 +121,7 @@ rm tests/ui/abi/large-byval-align.rs # exceeds implementation limit of Cranelift # ============================================================ rm -r tests/run-make/remap-path-prefix-dwarf # requires llvm-dwarfdump rm -r tests/run-make/strip # same -rm -r tests/run-make/compiler-builtins # Expects lib/rustlib/src/rust to contains the standard library source +rm -r tests/run-make-cargo/compiler-builtins # Expects lib/rustlib/src/rust to contains the standard library source rm -r tests/run-make/translation # same rm -r tests/run-make/missing-unstable-trait-bound # This disables support for unstable features, but running cg_clif needs some unstable features rm -r tests/run-make/const-trait-stable-toolchain # same @@ -166,5 +166,5 @@ index 073116933bd..c3e4578204d 100644 EOF echo "[TEST] rustc test suite" -./x.py test --stage 0 --test-args=--no-capture tests/{codegen-units,run-make,ui,incremental} +./x.py test --stage 0 --test-args=--no-capture tests/{codegen-units,run-make,run-make-cargo,ui,incremental} popd diff --git a/compiler/rustc_codegen_gcc/build_system/src/test.rs b/compiler/rustc_codegen_gcc/build_system/src/test.rs index 3dd3fce2eec..1823aa71f40 100644 --- a/compiler/rustc_codegen_gcc/build_system/src/test.rs +++ b/compiler/rustc_codegen_gcc/build_system/src/test.rs @@ -1083,11 +1083,12 @@ where fn test_rustc(env: &Env, args: &TestArg) -> Result<(), String> { test_rustc_inner(env, args, |_| Ok(false), false, "run-make")?; + test_rustc_inner(env, args, |_| Ok(false), false, "run-make-cargo")?; test_rustc_inner(env, args, |_| Ok(false), false, "ui") } fn test_failing_rustc(env: &Env, args: &TestArg) -> Result<(), String> { - let result1 = test_rustc_inner( + let run_make_result = test_rustc_inner( env, args, retain_files_callback("tests/failing-run-make-tests.txt", "run-make"), @@ -1095,7 +1096,15 @@ fn test_failing_rustc(env: &Env, args: &TestArg) -> Result<(), String> { "run-make", ); - let result2 = test_rustc_inner( + let run_make_cargo_result = test_rustc_inner( + env, + args, + retain_files_callback("tests/failing-run-make-tests.txt", "run-make-cargo"), + false, + "run-make", + ); + + let ui_result = test_rustc_inner( env, args, retain_files_callback("tests/failing-ui-tests.txt", "ui"), @@ -1103,7 +1112,7 @@ fn test_failing_rustc(env: &Env, args: &TestArg) -> Result<(), String> { "ui", ); - result1.and(result2) + run_make_result.and(run_make_cargo_result).and(ui_result) } fn test_successful_rustc(env: &Env, args: &TestArg) -> Result<(), String> { @@ -1120,6 +1129,13 @@ fn test_successful_rustc(env: &Env, args: &TestArg) -> Result<(), String> { remove_files_callback("tests/failing-run-make-tests.txt", "run-make"), false, "run-make", + )?; + test_rustc_inner( + env, + args, + remove_files_callback("tests/failing-run-make-tests.txt", "run-make-cargo"), + false, + "run-make-cargo", ) } diff --git a/compiler/rustc_codegen_gcc/src/back/lto.rs b/compiler/rustc_codegen_gcc/src/back/lto.rs index fcee6b6df62..d29bba2570f 100644 --- a/compiler/rustc_codegen_gcc/src/back/lto.rs +++ b/compiler/rustc_codegen_gcc/src/back/lto.rs @@ -204,7 +204,7 @@ fn fat_lto( let path = tmp_path.path().to_path_buf().join(&module.name); let path = path.to_str().expect("path"); let context = &module.module_llvm.context; - let config = cgcx.config(module.kind); + let config = &cgcx.module_config; // NOTE: we need to set the optimization level here in order for LTO to do its job. context.set_optimization_level(to_gcc_opt_level(config.opt_level)); context.add_command_line_option("-flto=auto"); @@ -305,12 +305,9 @@ pub(crate) fn run_thin( ) } -pub(crate) fn prepare_thin( - module: ModuleCodegen<GccContext>, - _emit_summary: bool, -) -> (String, ThinBuffer) { +pub(crate) fn prepare_thin(module: ModuleCodegen<GccContext>) -> (String, ThinBuffer) { let name = module.name; - //let buffer = ThinBuffer::new(module.module_llvm.context, true, emit_summary); + //let buffer = ThinBuffer::new(module.module_llvm.context, true); let buffer = ThinBuffer::new(&module.module_llvm.context); (name, buffer) } @@ -650,10 +647,6 @@ impl ThinBufferMethods for ThinBuffer { fn data(&self) -> &[u8] { &[] } - - fn thin_link_data(&self) -> &[u8] { - unimplemented!(); - } } pub struct ThinData; //(Arc<TempDir>); diff --git a/compiler/rustc_codegen_gcc/src/lib.rs b/compiler/rustc_codegen_gcc/src/lib.rs index 2d7df79ba95..f76f933cad4 100644 --- a/compiler/rustc_codegen_gcc/src/lib.rs +++ b/compiler/rustc_codegen_gcc/src/lib.rs @@ -408,11 +408,8 @@ impl WriteBackendMethods for GccCodegenBackend { back::write::codegen(cgcx, module, config) } - fn prepare_thin( - module: ModuleCodegen<Self::Module>, - emit_summary: bool, - ) -> (String, Self::ThinBuffer) { - back::lto::prepare_thin(module, emit_summary) + fn prepare_thin(module: ModuleCodegen<Self::Module>) -> (String, Self::ThinBuffer) { + back::lto::prepare_thin(module) } fn serialize_module(_module: ModuleCodegen<Self::Module>) -> (String, Self::ModuleBuffer) { diff --git a/compiler/rustc_codegen_llvm/src/attributes.rs b/compiler/rustc_codegen_llvm/src/attributes.rs index 5affb26483a..573c51a9539 100644 --- a/compiler/rustc_codegen_llvm/src/attributes.rs +++ b/compiler/rustc_codegen_llvm/src/attributes.rs @@ -29,8 +29,18 @@ pub(crate) fn apply_to_callsite(callsite: &Value, idx: AttributePlace, attrs: &[ } /// Get LLVM attribute for the provided inline heuristic. -#[inline] -fn inline_attr<'ll>(cx: &CodegenCx<'ll, '_>, inline: InlineAttr) -> Option<&'ll Attribute> { +pub(crate) fn inline_attr<'ll, 'tcx>( + cx: &CodegenCx<'ll, 'tcx>, + instance: ty::Instance<'tcx>, +) -> Option<&'ll Attribute> { + // `optnone` requires `noinline` + let codegen_fn_attrs = cx.tcx.codegen_fn_attrs(instance.def_id()); + let inline = match (codegen_fn_attrs.inline, &codegen_fn_attrs.optimize) { + (_, OptimizeAttr::DoNotOptimize) => InlineAttr::Never, + (InlineAttr::None, _) if instance.def.requires_inline(cx.tcx) => InlineAttr::Hint, + (inline, _) => inline, + }; + if !cx.tcx.sess.opts.unstable_opts.inline_llvm { // disable LLVM inlining return Some(AttributeKind::NoInline.create_attr(cx.llcx)); @@ -286,6 +296,19 @@ pub(crate) fn tune_cpu_attr<'ll>(cx: &CodegenCx<'ll, '_>) -> Option<&'ll Attribu .map(|tune_cpu| llvm::CreateAttrStringValue(cx.llcx, "tune-cpu", tune_cpu)) } +/// Get the `target-features` LLVM attribute. +pub(crate) fn target_features_attr<'ll>( + cx: &CodegenCx<'ll, '_>, + function_features: Vec<String>, +) -> Option<&'ll Attribute> { + let global_features = cx.tcx.global_backend_features(()).iter().map(String::as_str); + let function_features = function_features.iter().map(String::as_str); + let target_features = + global_features.chain(function_features).intersperse(",").collect::<String>(); + (!target_features.is_empty()) + .then(|| llvm::CreateAttrStringValue(cx.llcx, "target-features", &target_features)) +} + /// Get the `NonLazyBind` LLVM attribute, /// if the codegen options allow skipping the PLT. pub(crate) fn non_lazy_bind_attr<'ll>(cx: &CodegenCx<'ll, '_>) -> Option<&'ll Attribute> { @@ -346,14 +369,6 @@ pub(crate) fn llfn_attrs_from_instance<'ll, 'tcx>( OptimizeAttr::Speed => {} } - // `optnone` requires `noinline` - let inline = match (codegen_fn_attrs.inline, &codegen_fn_attrs.optimize) { - (_, OptimizeAttr::DoNotOptimize) => InlineAttr::Never, - (InlineAttr::None, _) if instance.def.requires_inline(cx.tcx) => InlineAttr::Hint, - (inline, _) => inline, - }; - to_add.extend(inline_attr(cx, inline)); - if cx.sess().must_emit_unwind_tables() { to_add.push(uwtable_attr(cx.llcx, cx.sess().opts.unstable_opts.use_sync_unwind)); } @@ -488,6 +503,14 @@ pub(crate) fn llfn_attrs_from_instance<'ll, 'tcx>( let function_features = codegen_fn_attrs.target_features.iter().map(|f| f.name.as_str()).collect::<Vec<&str>>(); + // Apply function attributes as per usual if there are no user defined + // target features otherwise this will get applied at the callsite. + if function_features.is_empty() { + if let Some(inline_attr) = inline_attr(cx, instance) { + to_add.push(inline_attr); + } + } + let function_features = function_features .iter() // Convert to LLVMFeatures and filter out unavailable ones @@ -513,13 +536,7 @@ pub(crate) fn llfn_attrs_from_instance<'ll, 'tcx>( } } - let global_features = cx.tcx.global_backend_features(()).iter().map(|s| s.as_str()); - let function_features = function_features.iter().map(|s| s.as_str()); - let target_features: String = - global_features.chain(function_features).intersperse(",").collect(); - if !target_features.is_empty() { - to_add.push(llvm::CreateAttrStringValue(cx.llcx, "target-features", &target_features)); - } + to_add.extend(target_features_attr(cx, function_features)); attributes::apply_to_llfn(llfn, Function, &to_add); } diff --git a/compiler/rustc_codegen_llvm/src/back/lto.rs b/compiler/rustc_codegen_llvm/src/back/lto.rs index fc38c4f3e51..f571716d9dd 100644 --- a/compiler/rustc_codegen_llvm/src/back/lto.rs +++ b/compiler/rustc_codegen_llvm/src/back/lto.rs @@ -43,9 +43,7 @@ fn prepare_lto( .map(|symbol| CString::new(symbol.to_owned()).unwrap()) .collect::<Vec<CString>>(); - if cgcx.regular_module_config.instrument_coverage - || cgcx.regular_module_config.pgo_gen.enabled() - { + if cgcx.module_config.instrument_coverage || cgcx.module_config.pgo_gen.enabled() { // These are weak symbols that point to the profile version and the // profile name, which need to be treated as exported so LTO doesn't nix // them. @@ -55,15 +53,15 @@ fn prepare_lto( symbols_below_threshold.extend(PROFILER_WEAK_SYMBOLS.iter().map(|&sym| sym.to_owned())); } - if cgcx.regular_module_config.sanitizer.contains(SanitizerSet::MEMORY) { + if cgcx.module_config.sanitizer.contains(SanitizerSet::MEMORY) { let mut msan_weak_symbols = Vec::new(); // Similar to profiling, preserve weak msan symbol during LTO. - if cgcx.regular_module_config.sanitizer_recover.contains(SanitizerSet::MEMORY) { + if cgcx.module_config.sanitizer_recover.contains(SanitizerSet::MEMORY) { msan_weak_symbols.push(c"__msan_keep_going"); } - if cgcx.regular_module_config.sanitizer_memory_track_origins != 0 { + if cgcx.module_config.sanitizer_memory_track_origins != 0 { msan_weak_symbols.push(c"__msan_track_origins"); } @@ -187,12 +185,9 @@ 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>, - emit_summary: bool, -) -> (String, ThinBuffer) { +pub(crate) fn prepare_thin(module: ModuleCodegen<ModuleLlvm>) -> (String, ThinBuffer) { let name = module.name; - let buffer = ThinBuffer::new(module.module_llvm.llmod(), true, emit_summary); + let buffer = ThinBuffer::new(module.module_llvm.llmod(), true); (name, buffer) } @@ -583,7 +578,7 @@ pub(crate) fn run_pass_manager( thin: bool, ) { let _timer = cgcx.prof.generic_activity_with_arg("LLVM_lto_optimize", &*module.name); - let config = cgcx.config(module.kind); + let config = &cgcx.module_config; // Now we have one massive module inside of llmod. Time to run the // LTO-specific optimization passes that LLVM provides. @@ -689,9 +684,9 @@ unsafe impl Send for ThinBuffer {} unsafe impl Sync for ThinBuffer {} impl ThinBuffer { - pub(crate) fn new(m: &llvm::Module, is_thin: bool, emit_summary: bool) -> ThinBuffer { + pub(crate) fn new(m: &llvm::Module, is_thin: bool) -> ThinBuffer { unsafe { - let buffer = llvm::LLVMRustThinLTOBufferCreate(m, is_thin, emit_summary); + let buffer = llvm::LLVMRustThinLTOBufferCreate(m, is_thin); ThinBuffer(buffer) } } @@ -700,21 +695,21 @@ impl ThinBuffer { let mut ptr = NonNull::new(ptr).unwrap(); ThinBuffer(unsafe { ptr.as_mut() }) } -} -impl ThinBufferMethods for ThinBuffer { - fn data(&self) -> &[u8] { + pub(crate) fn thin_link_data(&self) -> &[u8] { unsafe { - let ptr = llvm::LLVMRustThinLTOBufferPtr(self.0) as *const _; - let len = llvm::LLVMRustThinLTOBufferLen(self.0); + let ptr = llvm::LLVMRustThinLTOBufferThinLinkDataPtr(self.0) as *const _; + let len = llvm::LLVMRustThinLTOBufferThinLinkDataLen(self.0); slice::from_raw_parts(ptr, len) } } +} - fn thin_link_data(&self) -> &[u8] { +impl ThinBufferMethods for ThinBuffer { + fn data(&self) -> &[u8] { unsafe { - let ptr = llvm::LLVMRustThinLTOBufferThinLinkDataPtr(self.0) as *const _; - let len = llvm::LLVMRustThinLTOBufferThinLinkDataLen(self.0); + let ptr = llvm::LLVMRustThinLTOBufferPtr(self.0) as *const _; + let len = llvm::LLVMRustThinLTOBufferLen(self.0); slice::from_raw_parts(ptr, len) } } @@ -745,7 +740,7 @@ pub(crate) fn optimize_thin_module( let module_llvm = ModuleLlvm::parse(cgcx, module_name, thin_module.data(), dcx); let mut module = ModuleCodegen::new_regular(thin_module.name(), module_llvm); // Given that the newly created module lacks a thinlto buffer for embedding, we need to re-add it here. - if cgcx.config(ModuleKind::Regular).embed_bitcode() { + if cgcx.module_config.embed_bitcode() { module.thin_lto_buffer = Some(thin_module.data().to_vec()); } { diff --git a/compiler/rustc_codegen_llvm/src/back/write.rs b/compiler/rustc_codegen_llvm/src/back/write.rs index 7ea2ae6673b..423f0da4878 100644 --- a/compiler/rustc_codegen_llvm/src/back/write.rs +++ b/compiler/rustc_codegen_llvm/src/back/write.rs @@ -837,7 +837,7 @@ pub(crate) fn codegen( "LLVM_module_codegen_make_bitcode", &*module.name, ); - ThinBuffer::new(llmod, config.emit_thin_lto, false) + ThinBuffer::new(llmod, config.emit_thin_lto) }; let data = thin.data(); let _timer = cgcx diff --git a/compiler/rustc_codegen_llvm/src/builder.rs b/compiler/rustc_codegen_llvm/src/builder.rs index 37379586d58..7d0691366e6 100644 --- a/compiler/rustc_codegen_llvm/src/builder.rs +++ b/compiler/rustc_codegen_llvm/src/builder.rs @@ -1392,7 +1392,7 @@ impl<'a, 'll, 'tcx> BuilderMethods<'a, 'tcx> for Builder<'a, 'll, 'tcx> { fn call( &mut self, llty: &'ll Type, - fn_attrs: Option<&CodegenFnAttrs>, + fn_call_attrs: Option<&CodegenFnAttrs>, fn_abi: Option<&FnAbi<'tcx, Ty<'tcx>>>, llfn: &'ll Value, args: &[&'ll Value], @@ -1409,10 +1409,10 @@ impl<'a, 'll, 'tcx> BuilderMethods<'a, 'tcx> for Builder<'a, 'll, 'tcx> { } // Emit CFI pointer type membership test - self.cfi_type_test(fn_attrs, fn_abi, instance, llfn); + self.cfi_type_test(fn_call_attrs, fn_abi, instance, llfn); // Emit KCFI operand bundle - let kcfi_bundle = self.kcfi_operand_bundle(fn_attrs, fn_abi, instance, llfn); + let kcfi_bundle = self.kcfi_operand_bundle(fn_call_attrs, fn_abi, instance, llfn); if let Some(kcfi_bundle) = kcfi_bundle.as_ref().map(|b| b.as_ref()) { bundles.push(kcfi_bundle); } @@ -1429,6 +1429,29 @@ impl<'a, 'll, 'tcx> BuilderMethods<'a, 'tcx> for Builder<'a, 'll, 'tcx> { c"".as_ptr(), ) }; + + if let Some(instance) = instance { + // Attributes on the function definition being called + let fn_defn_attrs = self.cx.tcx.codegen_fn_attrs(instance.def_id()); + if let Some(fn_call_attrs) = fn_call_attrs + && !fn_call_attrs.target_features.is_empty() + // If there is an inline attribute and a target feature that matches + // we will add the attribute to the callsite otherwise we'll omit + // this and not add the attribute to prevent soundness issues. + && let Some(inlining_rule) = attributes::inline_attr(&self.cx, instance) + && self.cx.tcx.is_target_feature_call_safe( + &fn_call_attrs.target_features, + &fn_defn_attrs.target_features, + ) + { + attributes::apply_to_callsite( + call, + llvm::AttributePlace::Function, + &[inlining_rule], + ); + } + } + if let Some(fn_abi) = fn_abi { fn_abi.apply_attrs_callsite(self, call); } diff --git a/compiler/rustc_codegen_llvm/src/builder/gpu_offload.rs b/compiler/rustc_codegen_llvm/src/builder/gpu_offload.rs index 1280ab1442a..0737a18384b 100644 --- a/compiler/rustc_codegen_llvm/src/builder/gpu_offload.rs +++ b/compiler/rustc_codegen_llvm/src/builder/gpu_offload.rs @@ -193,7 +193,7 @@ fn gen_define_handling<'ll>( // reference) types. let num_ptr_types = types .iter() - .map(|&x| matches!(cx.type_kind(x), rustc_codegen_ssa::common::TypeKind::Pointer)) + .filter(|&x| matches!(cx.type_kind(x), rustc_codegen_ssa::common::TypeKind::Pointer)) .count(); // We do not know their size anymore at this level, so hardcode a placeholder. diff --git a/compiler/rustc_codegen_llvm/src/context.rs b/compiler/rustc_codegen_llvm/src/context.rs index 4fd6110ac4a..a69fa54a54a 100644 --- a/compiler/rustc_codegen_llvm/src/context.rs +++ b/compiler/rustc_codegen_llvm/src/context.rs @@ -217,6 +217,10 @@ pub(crate) unsafe fn create_module<'ll>( // LLVM 22.0 updated the default layout on avr: https://github.com/llvm/llvm-project/pull/153010 target_data_layout = target_data_layout.replace("n8:16", "n8") } + if sess.target.arch == "nvptx64" { + // LLVM 22 updated the NVPTX layout to indicate 256-bit vector load/store: https://github.com/llvm/llvm-project/pull/155198 + target_data_layout = target_data_layout.replace("-i256:256", ""); + } } // Ensure the data-layout values hardcoded remain the defaults. @@ -849,7 +853,7 @@ impl<'ll, 'tcx> MiscCodegenMethods<'tcx> for CodegenCx<'ll, 'tcx> { fn declare_c_main(&self, fn_type: Self::Type) -> Option<Self::Function> { let entry_name = self.sess().target.entry_name.as_ref(); if self.get_declared_value(entry_name).is_none() { - Some(self.declare_entry_fn( + let llfn = self.declare_entry_fn( entry_name, llvm::CallConv::from_conv( self.sess().target.entry_abi, @@ -857,7 +861,13 @@ impl<'ll, 'tcx> MiscCodegenMethods<'tcx> for CodegenCx<'ll, 'tcx> { ), llvm::UnnamedAddr::Global, fn_type, - )) + ); + attributes::apply_to_llfn( + llfn, + llvm::AttributePlace::Function, + attributes::target_features_attr(self, vec![]).as_slice(), + ); + Some(llfn) } else { // If the symbol already exists, it is an error: for example, the user wrote // #[no_mangle] extern "C" fn main(..) {..} diff --git a/compiler/rustc_codegen_llvm/src/intrinsic.rs b/compiler/rustc_codegen_llvm/src/intrinsic.rs index 49d3dedbeab..85f71f331a4 100644 --- a/compiler/rustc_codegen_llvm/src/intrinsic.rs +++ b/compiler/rustc_codegen_llvm/src/intrinsic.rs @@ -383,7 +383,9 @@ impl<'ll, 'tcx> IntrinsicCallBuilderMethods<'tcx> for Builder<'_, 'll, 'tcx> { | sym::rotate_left | sym::rotate_right | sym::saturating_add - | sym::saturating_sub => { + | sym::saturating_sub + | sym::unchecked_funnel_shl + | sym::unchecked_funnel_shr => { let ty = args[0].layout.ty; if !ty.is_integral() { tcx.dcx().emit_err(InvalidMonomorphization::BasicIntegerType { @@ -424,18 +426,26 @@ impl<'ll, 'tcx> IntrinsicCallBuilderMethods<'tcx> for Builder<'_, 'll, 'tcx> { sym::bitreverse => { self.call_intrinsic("llvm.bitreverse", &[llty], &[args[0].immediate()]) } - sym::rotate_left | sym::rotate_right => { - let is_left = name == sym::rotate_left; - let val = args[0].immediate(); - let raw_shift = args[1].immediate(); - // rotate = funnel shift with first two args the same + sym::rotate_left + | sym::rotate_right + | sym::unchecked_funnel_shl + | sym::unchecked_funnel_shr => { + let is_left = name == sym::rotate_left || name == sym::unchecked_funnel_shl; + let lhs = args[0].immediate(); + let (rhs, raw_shift) = + if name == sym::rotate_left || name == sym::rotate_right { + // rotate = funnel shift with first two args the same + (lhs, args[1].immediate()) + } else { + (args[1].immediate(), args[2].immediate()) + }; let llvm_name = format!("llvm.fsh{}", if is_left { 'l' } else { 'r' }); // llvm expects shift to be the same type as the values, but rust // always uses `u32`. - let raw_shift = self.intcast(raw_shift, self.val_ty(val), false); + let raw_shift = self.intcast(raw_shift, self.val_ty(lhs), false); - self.call_intrinsic(llvm_name, &[llty], &[val, val, raw_shift]) + self.call_intrinsic(llvm_name, &[llty], &[lhs, rhs, raw_shift]) } sym::saturating_add | sym::saturating_sub => { let is_add = name == sym::saturating_add; diff --git a/compiler/rustc_codegen_llvm/src/lib.rs b/compiler/rustc_codegen_llvm/src/lib.rs index 628cb34fd9e..6fb23d09843 100644 --- a/compiler/rustc_codegen_llvm/src/lib.rs +++ b/compiler/rustc_codegen_llvm/src/lib.rs @@ -211,11 +211,8 @@ impl WriteBackendMethods for LlvmCodegenBackend { ) -> CompiledModule { back::write::codegen(cgcx, module, config) } - fn prepare_thin( - module: ModuleCodegen<Self::Module>, - emit_summary: bool, - ) -> (String, Self::ThinBuffer) { - back::lto::prepare_thin(module, emit_summary) + fn prepare_thin(module: ModuleCodegen<Self::Module>) -> (String, Self::ThinBuffer) { + back::lto::prepare_thin(module) } fn serialize_module(module: ModuleCodegen<Self::Module>) -> (String, Self::ModuleBuffer) { (module.name, back::lto::ModuleBuffer::new(module.module_llvm.llmod())) diff --git a/compiler/rustc_codegen_llvm/src/llvm/ffi.rs b/compiler/rustc_codegen_llvm/src/llvm/ffi.rs index b66fc157b3c..0679f55ab7f 100644 --- a/compiler/rustc_codegen_llvm/src/llvm/ffi.rs +++ b/compiler/rustc_codegen_llvm/src/llvm/ffi.rs @@ -2602,7 +2602,6 @@ unsafe extern "C" { pub(crate) fn LLVMRustThinLTOBufferCreate( M: &Module, is_thin: bool, - emit_summary: bool, ) -> &'static mut ThinLTOBuffer; pub(crate) fn LLVMRustThinLTOBufferFree(M: &'static mut ThinLTOBuffer); pub(crate) fn LLVMRustThinLTOBufferPtr(M: &ThinLTOBuffer) -> *const c_char; diff --git a/compiler/rustc_codegen_ssa/messages.ftl b/compiler/rustc_codegen_ssa/messages.ftl index 44b9941691a..dd688b8b345 100644 --- a/compiler/rustc_codegen_ssa/messages.ftl +++ b/compiler/rustc_codegen_ssa/messages.ftl @@ -40,15 +40,15 @@ codegen_ssa_dynamic_linking_with_lto = .note = only 'staticlib', 'bin', and 'cdylib' outputs are supported with LTO codegen_ssa_error_calling_dlltool = - Error calling dlltool '{$dlltool_path}': {$error} + error calling dlltool '{$dlltool_path}': {$error} codegen_ssa_error_creating_import_library = - Error creating import library for {$lib_name}: {$error} + error creating import library for {$lib_name}: {$error} codegen_ssa_error_creating_remark_dir = failed to create remark directory: {$error} codegen_ssa_error_writing_def_file = - Error writing .DEF file: {$error} + error writing .DEF file: {$error} codegen_ssa_expected_name_value_pair = expected name value pair @@ -264,9 +264,9 @@ codegen_ssa_shuffle_indices_evaluation = could not evaluate shuffle_indices at c codegen_ssa_specify_libraries_to_link = use the `-l` flag to specify native libraries to link -codegen_ssa_static_library_native_artifacts = Link against the following native artifacts when linking against this static library. The order and any duplication can be significant on some platforms. +codegen_ssa_static_library_native_artifacts = link against the following native artifacts when linking against this static library. The order and any duplication can be significant on some platforms. -codegen_ssa_static_library_native_artifacts_to_file = Native artifacts to link against have been written to {$path}. The order and any duplication can be significant on some platforms. +codegen_ssa_static_library_native_artifacts_to_file = native artifacts to link against have been written to {$path}. The order and any duplication can be significant on some platforms. codegen_ssa_stripping_debug_info_failed = stripping debug info with `{$util}` failed: {$status} .note = {$output} @@ -364,13 +364,13 @@ codegen_ssa_unable_to_run = unable to run `{$util}`: {$error} codegen_ssa_unable_to_run_dsymutil = unable to run `dsymutil`: {$error} -codegen_ssa_unable_to_write_debugger_visualizer = Unable to write debugger visualizer file `{$path}`: {$error} +codegen_ssa_unable_to_write_debugger_visualizer = unable to write debugger visualizer file `{$path}`: {$error} codegen_ssa_unexpected_parameter_name = unexpected parameter name .label = expected `{$prefix_nops}` or `{$entry_nops}` codegen_ssa_unknown_archive_kind = - Don't know how to build archive of type: {$kind} + don't know how to build archive of type: {$kind} codegen_ssa_unknown_ctarget_feature = unknown and unstable feature specified for `-Ctarget-feature`: `{$feature}` diff --git a/compiler/rustc_codegen_ssa/src/back/link.rs b/compiler/rustc_codegen_ssa/src/back/link.rs index 19c919c0e4e..48b01ea2df1 100644 --- a/compiler/rustc_codegen_ssa/src/back/link.rs +++ b/compiler/rustc_codegen_ssa/src/back/link.rs @@ -58,6 +58,7 @@ use super::linker::{self, Linker}; use super::metadata::{MetadataPosition, create_wrapper_file}; use super::rpath::{self, RPathConfig}; use super::{apple, versioned_llvm_target}; +use crate::base::needs_allocator_shim_for_linking; use crate::{ CodegenResults, CompiledModule, CrateInfo, NativeLib, errors, looks_like_rust_object_file, }; @@ -2080,9 +2081,17 @@ fn add_local_crate_regular_objects(cmd: &mut dyn Linker, codegen_results: &Codeg } /// Add object files for allocator code linked once for the whole crate tree. -fn add_local_crate_allocator_objects(cmd: &mut dyn Linker, codegen_results: &CodegenResults) { - if let Some(obj) = codegen_results.allocator_module.as_ref().and_then(|m| m.object.as_ref()) { - cmd.add_object(obj); +fn add_local_crate_allocator_objects( + cmd: &mut dyn Linker, + codegen_results: &CodegenResults, + crate_type: CrateType, +) { + if needs_allocator_shim_for_linking(&codegen_results.crate_info.dependency_formats, crate_type) + { + if let Some(obj) = codegen_results.allocator_module.as_ref().and_then(|m| m.object.as_ref()) + { + cmd.add_object(obj); + } } } @@ -2281,7 +2290,7 @@ fn linker_with_args( codegen_results, metadata, ); - add_local_crate_allocator_objects(cmd, codegen_results); + add_local_crate_allocator_objects(cmd, codegen_results, crate_type); // Avoid linking to dynamic libraries unless they satisfy some undefined symbols // at the point at which they are specified on the command line. diff --git a/compiler/rustc_codegen_ssa/src/back/linker.rs b/compiler/rustc_codegen_ssa/src/back/linker.rs index df1e91b12f9..a2efd420a32 100644 --- a/compiler/rustc_codegen_ssa/src/back/linker.rs +++ b/compiler/rustc_codegen_ssa/src/back/linker.rs @@ -11,8 +11,9 @@ use rustc_metadata::{ }; use rustc_middle::bug; use rustc_middle::middle::dependency_format::Linkage; -use rustc_middle::middle::exported_symbols; -use rustc_middle::middle::exported_symbols::{ExportedSymbol, SymbolExportInfo, SymbolExportKind}; +use rustc_middle::middle::exported_symbols::{ + self, ExportedSymbol, SymbolExportInfo, SymbolExportKind, SymbolExportLevel, +}; use rustc_middle::ty::TyCtxt; use rustc_session::Session; use rustc_session::config::{self, CrateType, DebugInfo, LinkerPluginLto, Lto, OptLevel, Strip}; @@ -22,6 +23,8 @@ use tracing::{debug, warn}; use super::command::Command; use super::symbol_export; +use crate::back::symbol_export::allocator_shim_symbols; +use crate::base::needs_allocator_shim_for_linking; use crate::errors; #[cfg(test)] @@ -1827,7 +1830,7 @@ fn exported_symbols_for_non_proc_macro( let export_threshold = symbol_export::crates_export_threshold(&[crate_type]); for_each_exported_symbols_include_dep(tcx, crate_type, |symbol, info, cnum| { // Do not export mangled symbols from cdylibs and don't attempt to export compiler-builtins - // from any cdylib. The latter doesn't work anyway as we use hidden visibility for + // from any dylib. The latter doesn't work anyway as we use hidden visibility for // compiler-builtins. Most linkers silently ignore it, but ld64 gives a warning. if info.level.is_below_threshold(export_threshold) && !tcx.is_compiler_builtins(cnum) { symbols.push(( @@ -1838,6 +1841,14 @@ fn exported_symbols_for_non_proc_macro( } }); + // Mark allocator shim symbols as exported only if they were generated. + if export_threshold == SymbolExportLevel::Rust + && needs_allocator_shim_for_linking(tcx.dependency_formats(()), crate_type) + && tcx.allocator_kind(()).is_some() + { + symbols.extend(allocator_shim_symbols(tcx)); + } + symbols } diff --git a/compiler/rustc_codegen_ssa/src/back/lto.rs b/compiler/rustc_codegen_ssa/src/back/lto.rs index c95038375a1..e6df6a2469f 100644 --- a/compiler/rustc_codegen_ssa/src/back/lto.rs +++ b/compiler/rustc_codegen_ssa/src/back/lto.rs @@ -8,8 +8,9 @@ use rustc_middle::ty::TyCtxt; use rustc_session::config::{CrateType, Lto}; use tracing::info; -use crate::back::symbol_export::{self, symbol_name_for_instance_in_crate}; +use crate::back::symbol_export::{self, allocator_shim_symbols, symbol_name_for_instance_in_crate}; use crate::back::write::CodegenContext; +use crate::base::allocator_kind_for_codegen; use crate::errors::{DynamicLinkingWithLTO, LtoDisallowed, LtoDylib, LtoProcMacro}; use crate::traits::*; @@ -115,6 +116,11 @@ pub(super) fn exported_symbols_for_lto( } } + // Mark allocator shim symbols as exported only if they were generated. + if export_threshold == SymbolExportLevel::Rust && allocator_kind_for_codegen(tcx).is_some() { + symbols_below_threshold.extend(allocator_shim_symbols(tcx).map(|(name, _kind)| name)); + } + symbols_below_threshold } diff --git a/compiler/rustc_codegen_ssa/src/back/symbol_export.rs b/compiler/rustc_codegen_ssa/src/back/symbol_export.rs index d8a1480e911..b49e67217fb 100644 --- a/compiler/rustc_codegen_ssa/src/back/symbol_export.rs +++ b/compiler/rustc_codegen_ssa/src/back/symbol_export.rs @@ -18,7 +18,7 @@ use rustc_symbol_mangling::mangle_internal_symbol; use rustc_target::spec::TlsModel; use tracing::debug; -use crate::base::allocator_kind_for_codegen; +use crate::back::symbol_export; fn threshold(tcx: TyCtxt<'_>) -> SymbolExportLevel { crates_export_threshold(tcx.crate_types()) @@ -217,31 +217,6 @@ fn exported_non_generic_symbols_provider_local<'tcx>( )); } - // Mark allocator shim symbols as exported only if they were generated. - if allocator_kind_for_codegen(tcx).is_some() { - for symbol_name in ALLOCATOR_METHODS - .iter() - .map(|method| mangle_internal_symbol(tcx, global_fn_name(method.name).as_str())) - .chain([ - mangle_internal_symbol(tcx, "__rust_alloc_error_handler"), - mangle_internal_symbol(tcx, OomStrategy::SYMBOL), - mangle_internal_symbol(tcx, NO_ALLOC_SHIM_IS_UNSTABLE), - ]) - { - let exported_symbol = ExportedSymbol::NoDefId(SymbolName::new(tcx, &symbol_name)); - - symbols.push(( - exported_symbol, - SymbolExportInfo { - level: SymbolExportLevel::Rust, - kind: SymbolExportKind::Text, - used: false, - rustc_std_internal_symbol: true, - }, - )); - } - } - // Sort so we get a stable incr. comp. hash. symbols.sort_by_cached_key(|s| s.0.symbol_name_for_local_instance(tcx)); @@ -516,6 +491,31 @@ pub(crate) fn provide(providers: &mut Providers) { upstream_monomorphizations_for_provider; } +pub(crate) fn allocator_shim_symbols( + tcx: TyCtxt<'_>, +) -> impl Iterator<Item = (String, SymbolExportKind)> { + ALLOCATOR_METHODS + .iter() + .map(move |method| mangle_internal_symbol(tcx, global_fn_name(method.name).as_str())) + .chain([ + mangle_internal_symbol(tcx, "__rust_alloc_error_handler"), + mangle_internal_symbol(tcx, OomStrategy::SYMBOL), + mangle_internal_symbol(tcx, NO_ALLOC_SHIM_IS_UNSTABLE), + ]) + .map(move |symbol_name| { + let exported_symbol = ExportedSymbol::NoDefId(SymbolName::new(tcx, &symbol_name)); + + ( + symbol_export::exporting_symbol_name_for_instance_in_crate( + tcx, + exported_symbol, + LOCAL_CRATE, + ), + SymbolExportKind::Text, + ) + }) +} + fn symbol_export_level(tcx: TyCtxt<'_>, sym_def_id: DefId) -> SymbolExportLevel { // We export anything that's not mangled at the "C" layer as it probably has // to do with ABI concerns. We do not, however, apply such treatment to diff --git a/compiler/rustc_codegen_ssa/src/back/write.rs b/compiler/rustc_codegen_ssa/src/back/write.rs index 92582dcc399..cbaf67d7345 100644 --- a/compiler/rustc_codegen_ssa/src/back/write.rs +++ b/compiler/rustc_codegen_ssa/src/back/write.rs @@ -138,12 +138,23 @@ impl ModuleConfig { let emit_obj = if !should_emit_obj { EmitObj::None - } else if sess.target.obj_is_bitcode || sess.opts.cg.linker_plugin_lto.enabled() { + } else if sess.target.obj_is_bitcode + || (sess.opts.cg.linker_plugin_lto.enabled() && !no_builtins) + { // This case is selected if the target uses objects as bitcode, or // if linker plugin LTO is enabled. In the linker plugin LTO case // the assumption is that the final link-step will read the bitcode // and convert it to object code. This may be done by either the // native linker or rustc itself. + // + // Note, however, that the linker-plugin-lto requested here is + // explicitly ignored for `#![no_builtins]` crates. These crates are + // specifically ignored by rustc's LTO passes and wouldn't work if + // loaded into the linker. These crates define symbols that LLVM + // lowers intrinsics to, and these symbol dependencies aren't known + // until after codegen. As a result any crate marked + // `#![no_builtins]` is assumed to not participate in LTO and + // instead goes on to generate object code. EmitObj::Bitcode } else if need_bitcode_in_object(tcx) { EmitObj::ObjectCode(BitcodeSection::Full) @@ -322,8 +333,8 @@ pub struct CodegenContext<B: WriteBackendMethods> { pub crate_types: Vec<CrateType>, pub output_filenames: Arc<OutputFilenames>, pub invocation_temp: Option<String>, - pub regular_module_config: Arc<ModuleConfig>, - pub allocator_module_config: Arc<ModuleConfig>, + pub module_config: Arc<ModuleConfig>, + pub allocator_config: Arc<ModuleConfig>, pub tm_factory: TargetMachineFactoryFn<B>, pub msvc_imps_needed: bool, pub is_pe_coff: bool, @@ -361,13 +372,6 @@ impl<B: WriteBackendMethods> CodegenContext<B> { pub fn create_dcx(&self) -> DiagCtxt { DiagCtxt::new(Box::new(self.diag_emitter.clone())) } - - pub fn config(&self, kind: ModuleKind) -> &ModuleConfig { - match kind { - ModuleKind::Regular => &self.regular_module_config, - ModuleKind::Allocator => &self.allocator_module_config, - } - } } fn generate_thin_lto_work<B: ExtraBackendMethods>( @@ -431,6 +435,7 @@ pub(crate) fn start_async_codegen<B: ExtraBackendMethods>( backend: B, tcx: TyCtxt<'_>, target_cpu: String, + allocator_module: Option<ModuleCodegen<B::Module>>, ) -> OngoingCodegen<B> { let (coordinator_send, coordinator_receive) = channel(); @@ -454,6 +459,7 @@ pub(crate) fn start_async_codegen<B: ExtraBackendMethods>( coordinator_receive, Arc::new(regular_config), Arc::new(allocator_config), + allocator_module, coordinator_send.clone(), ); @@ -709,15 +715,6 @@ pub(crate) enum WorkItem<B: WriteBackendMethods> { } impl<B: WriteBackendMethods> WorkItem<B> { - fn module_kind(&self) -> ModuleKind { - match *self { - WorkItem::Optimize(ref m) => m.kind, - WorkItem::CopyPostLtoArtifacts(_) | WorkItem::FatLto { .. } | WorkItem::ThinLto(_) => { - ModuleKind::Regular - } - } - } - /// Generate a short description of this work item suitable for use as a thread name. fn short_description(&self) -> String { // `pthread_setname()` on *nix ignores anything beyond the first 15 @@ -806,7 +803,7 @@ pub(crate) fn compute_per_cgu_lto_type( let linker_does_lto = opts.cg.linker_plugin_lto.enabled(); // When we're automatically doing ThinLTO for multi-codegen-unit - // builds we don't actually want to LTO the allocator modules if + // builds we don't actually want to LTO the allocator module if // it shows up. This is due to various linker shenanigans that // we'll encounter later. let is_allocator = module_kind == ModuleKind::Allocator; @@ -832,11 +829,17 @@ pub(crate) fn compute_per_cgu_lto_type( fn execute_optimize_work_item<B: ExtraBackendMethods>( cgcx: &CodegenContext<B>, mut module: ModuleCodegen<B::Module>, - module_config: &ModuleConfig, ) -> WorkItemResult<B> { + let _timer = cgcx.prof.generic_activity_with_arg("codegen_module_optimize", &*module.name); + let dcx = cgcx.create_dcx(); let dcx = dcx.handle(); + let module_config = match module.kind { + ModuleKind::Regular => &cgcx.module_config, + ModuleKind::Allocator => &cgcx.allocator_config, + }; + B::optimize(cgcx, dcx, &mut module, module_config); // After we've done the initial round of optimizations we need to @@ -848,7 +851,7 @@ fn execute_optimize_work_item<B: ExtraBackendMethods>( // If we're doing some form of incremental LTO then we need to be sure to // save our module to disk first. - let bitcode = if cgcx.config(module.kind).emit_pre_lto_bc { + let bitcode = if module_config.emit_pre_lto_bc { let filename = pre_lto_bitcode_filename(&module.name); cgcx.incr_comp_session_dir.as_ref().map(|path| path.join(&filename)) } else { @@ -861,7 +864,7 @@ fn execute_optimize_work_item<B: ExtraBackendMethods>( WorkItemResult::Finished(module) } ComputedLtoType::Thin => { - let (name, thin_buffer) = B::prepare_thin(module, false); + let (name, thin_buffer) = B::prepare_thin(module); if let Some(path) = bitcode { fs::write(&path, thin_buffer.data()).unwrap_or_else(|e| { panic!("Error writing pre-lto-bitcode file `{}`: {}", path.display(), e); @@ -888,8 +891,11 @@ fn execute_optimize_work_item<B: ExtraBackendMethods>( fn execute_copy_from_cache_work_item<B: ExtraBackendMethods>( cgcx: &CodegenContext<B>, module: CachedModuleCodegen, - module_config: &ModuleConfig, ) -> WorkItemResult<B> { + let _timer = cgcx + .prof + .generic_activity_with_arg("codegen_copy_artifacts_from_incr_cache", &*module.name); + let incr_comp_session_dir = cgcx.incr_comp_session_dir.as_ref().unwrap(); let mut links_from_incr_cache = Vec::new(); @@ -948,6 +954,7 @@ fn execute_copy_from_cache_work_item<B: ExtraBackendMethods>( } }; + let module_config = &cgcx.module_config; let should_emit_obj = module_config.emit_obj != EmitObj::None; let assembly = load_from_incr_cache(module_config.emit_asm, OutputType::Assembly); let llvm_ir = load_from_incr_cache(module_config.emit_ir, OutputType::LlvmAssembly); @@ -959,8 +966,8 @@ fn execute_copy_from_cache_work_item<B: ExtraBackendMethods>( WorkItemResult::Finished(CompiledModule { links_from_incr_cache, - name: module.name, kind: ModuleKind::Regular, + name: module.name, object, dwarf_object, bytecode, @@ -975,8 +982,9 @@ fn execute_fat_lto_work_item<B: ExtraBackendMethods>( each_linked_rlib_for_lto: &[PathBuf], mut needs_fat_lto: Vec<FatLtoInput<B>>, import_only_modules: Vec<(SerializedModule<B::ModuleBuffer>, WorkProduct)>, - module_config: &ModuleConfig, ) -> WorkItemResult<B> { + let _timer = cgcx.prof.generic_activity_with_arg("codegen_module_perform_lto", "everything"); + for (module, wp) in import_only_modules { needs_fat_lto.push(FatLtoInput::Serialized { name: wp.cgu_name, buffer: module }) } @@ -987,17 +995,18 @@ fn execute_fat_lto_work_item<B: ExtraBackendMethods>( each_linked_rlib_for_lto, needs_fat_lto, ); - let module = B::codegen(cgcx, module, module_config); + let module = B::codegen(cgcx, module, &cgcx.module_config); WorkItemResult::Finished(module) } fn execute_thin_lto_work_item<B: ExtraBackendMethods>( cgcx: &CodegenContext<B>, module: lto::ThinModule<B>, - module_config: &ModuleConfig, ) -> WorkItemResult<B> { + let _timer = cgcx.prof.generic_activity_with_arg("codegen_module_perform_lto", module.name()); + let module = B::optimize_thin(cgcx, module); - let module = B::codegen(cgcx, module, module_config); + let module = B::codegen(cgcx, module, &cgcx.module_config); WorkItemResult::Finished(module) } @@ -1082,6 +1091,7 @@ fn start_executing_work<B: ExtraBackendMethods>( coordinator_receive: Receiver<Message<B>>, regular_config: Arc<ModuleConfig>, allocator_config: Arc<ModuleConfig>, + allocator_module: Option<ModuleCodegen<B::Module>>, tx_to_llvm_workers: Sender<Message<B>>, ) -> thread::JoinHandle<Result<CompiledModules, ()>> { let coordinator_send = tx_to_llvm_workers; @@ -1146,8 +1156,8 @@ fn start_executing_work<B: ExtraBackendMethods>( expanded_args: tcx.sess.expanded_args.clone(), diag_emitter: shared_emitter.clone(), output_filenames: Arc::clone(tcx.output_filenames(())), - regular_module_config: regular_config, - allocator_module_config: allocator_config, + module_config: regular_config, + allocator_config, tm_factory: backend.target_machine_factory(tcx.sess, ol, backend_features), msvc_imps_needed: msvc_imps_needed(tcx), is_pe_coff: tcx.sess.target.is_like_windows, @@ -1301,7 +1311,6 @@ fn start_executing_work<B: ExtraBackendMethods>( // This is where we collect codegen units that have gone all the way // through codegen and LLVM. let mut compiled_modules = vec![]; - let mut compiled_allocator_module = None; let mut needs_fat_lto = Vec::new(); let mut needs_thin_lto = Vec::new(); let mut lto_import_only_modules = Vec::new(); @@ -1342,6 +1351,17 @@ fn start_executing_work<B: ExtraBackendMethods>( let mut llvm_start_time: Option<VerboseTimingGuard<'_>> = None; + let compiled_allocator_module = allocator_module.and_then(|allocator_module| { + match execute_optimize_work_item(&cgcx, allocator_module) { + WorkItemResult::Finished(compiled_module) => return Some(compiled_module), + WorkItemResult::NeedsFatLto(fat_lto_input) => needs_fat_lto.push(fat_lto_input), + WorkItemResult::NeedsThinLto(name, thin_buffer) => { + needs_thin_lto.push((name, thin_buffer)) + } + } + None + }); + // Run the message loop while there's still anything that needs message // processing. Note that as soon as codegen is aborted we simply want to // wait for all existing work to finish, so many of the conditions here @@ -1575,15 +1595,7 @@ fn start_executing_work<B: ExtraBackendMethods>( match result { Ok(WorkItemResult::Finished(compiled_module)) => { - match compiled_module.kind { - ModuleKind::Regular => { - compiled_modules.push(compiled_module); - } - ModuleKind::Allocator => { - assert!(compiled_allocator_module.is_none()); - compiled_allocator_module = Some(compiled_module); - } - } + compiled_modules.push(compiled_module); } Ok(WorkItemResult::NeedsFatLto(fat_lto_input)) => { assert!(!started_lto); @@ -1711,46 +1723,22 @@ fn spawn_work<'a, B: ExtraBackendMethods>( let cgcx = cgcx.clone(); B::spawn_named_thread(cgcx.time_trace, work.short_description(), move || { - let result = std::panic::catch_unwind(AssertUnwindSafe(|| { - let module_config = cgcx.config(work.module_kind()); - - match work { - WorkItem::Optimize(m) => { - let _timer = - cgcx.prof.generic_activity_with_arg("codegen_module_optimize", &*m.name); - execute_optimize_work_item(&cgcx, m, module_config) - } - WorkItem::CopyPostLtoArtifacts(m) => { - let _timer = cgcx.prof.generic_activity_with_arg( - "codegen_copy_artifacts_from_incr_cache", - &*m.name, - ); - execute_copy_from_cache_work_item(&cgcx, m, module_config) - } - WorkItem::FatLto { - exported_symbols_for_lto, - each_linked_rlib_for_lto, - needs_fat_lto, - import_only_modules, - } => { - let _timer = cgcx - .prof - .generic_activity_with_arg("codegen_module_perform_lto", "everything"); - execute_fat_lto_work_item( - &cgcx, - &exported_symbols_for_lto, - &each_linked_rlib_for_lto, - needs_fat_lto, - import_only_modules, - module_config, - ) - } - WorkItem::ThinLto(m) => { - let _timer = - cgcx.prof.generic_activity_with_arg("codegen_module_perform_lto", m.name()); - execute_thin_lto_work_item(&cgcx, m, module_config) - } - } + let result = std::panic::catch_unwind(AssertUnwindSafe(|| match work { + WorkItem::Optimize(m) => execute_optimize_work_item(&cgcx, m), + WorkItem::CopyPostLtoArtifacts(m) => execute_copy_from_cache_work_item(&cgcx, m), + WorkItem::FatLto { + exported_symbols_for_lto, + each_linked_rlib_for_lto, + needs_fat_lto, + import_only_modules, + } => execute_fat_lto_work_item( + &cgcx, + &exported_symbols_for_lto, + &each_linked_rlib_for_lto, + needs_fat_lto, + import_only_modules, + ), + WorkItem::ThinLto(m) => execute_thin_lto_work_item(&cgcx, m), })); let msg = match result { diff --git a/compiler/rustc_codegen_ssa/src/base.rs b/compiler/rustc_codegen_ssa/src/base.rs index 8abaf201aba..45b028aa8ef 100644 --- a/compiler/rustc_codegen_ssa/src/base.rs +++ b/compiler/rustc_codegen_ssa/src/base.rs @@ -17,6 +17,7 @@ use rustc_hir::lang_items::LangItem; use rustc_hir::{ItemId, Target}; use rustc_middle::middle::codegen_fn_attrs::CodegenFnAttrs; use rustc_middle::middle::debugger_visualizer::{DebuggerVisualizerFile, DebuggerVisualizerType}; +use rustc_middle::middle::dependency_format::Dependencies; use rustc_middle::middle::exported_symbols::{self, SymbolExportKind}; use rustc_middle::middle::lang_items; use rustc_middle::mir::BinOp; @@ -630,14 +631,30 @@ pub fn allocator_kind_for_codegen(tcx: TyCtxt<'_>) -> Option<AllocatorKind> { // If the crate doesn't have an `allocator_kind` set then there's definitely // no shim to generate. Otherwise we also check our dependency graph for all // our output crate types. If anything there looks like its a `Dynamic` - // linkage, then it's already got an allocator shim and we'll be using that - // one instead. If nothing exists then it's our job to generate the - // allocator! - let any_dynamic_crate = tcx.dependency_formats(()).iter().any(|(_, list)| { + // linkage for all crate types we may link as, then it's already got an + // allocator shim and we'll be using that one instead. If nothing exists + // then it's our job to generate the allocator! If crate types disagree + // about whether an allocator shim is necessary or not, we generate one + // and let needs_allocator_shim_for_linking decide at link time whether or + // not to use it for any particular linker invocation. + let all_crate_types_any_dynamic_crate = tcx.dependency_formats(()).iter().all(|(_, list)| { use rustc_middle::middle::dependency_format::Linkage; list.iter().any(|&linkage| linkage == Linkage::Dynamic) }); - if any_dynamic_crate { None } else { tcx.allocator_kind(()) } + if all_crate_types_any_dynamic_crate { None } else { tcx.allocator_kind(()) } +} + +/// Decide if this particular crate type needs an allocator shim linked in. +/// This may return true even when allocator_kind_for_codegen returns false. In +/// this case no allocator shim shall be linked. +pub(crate) fn needs_allocator_shim_for_linking( + dependency_formats: &Dependencies, + crate_type: CrateType, +) -> bool { + use rustc_middle::middle::dependency_format::Linkage; + let any_dynamic_crate = + dependency_formats[&crate_type].iter().any(|&linkage| linkage == Linkage::Dynamic); + !any_dynamic_crate } pub fn codegen_crate<B: ExtraBackendMethods>( @@ -647,7 +664,7 @@ pub fn codegen_crate<B: ExtraBackendMethods>( ) -> OngoingCodegen<B> { // Skip crate items and just output metadata in -Z no-codegen mode. if tcx.sess.opts.unstable_opts.no_codegen || !tcx.sess.opts.output_types.should_codegen() { - let ongoing_codegen = start_async_codegen(backend, tcx, target_cpu); + let ongoing_codegen = start_async_codegen(backend, tcx, target_cpu, None); ongoing_codegen.codegen_finished(tcx); @@ -678,7 +695,27 @@ pub fn codegen_crate<B: ExtraBackendMethods>( } } - let ongoing_codegen = start_async_codegen(backend.clone(), tcx, target_cpu); + // Codegen an allocator shim, if necessary. + let allocator_module = if let Some(kind) = allocator_kind_for_codegen(tcx) { + let llmod_id = + cgu_name_builder.build_cgu_name(LOCAL_CRATE, &["crate"], Some("allocator")).to_string(); + + tcx.sess.time("write_allocator_module", || { + let module = backend.codegen_allocator( + tcx, + &llmod_id, + kind, + // If allocator_kind is Some then alloc_error_handler_kind must + // also be Some. + tcx.alloc_error_handler_kind(()).unwrap(), + ); + Some(ModuleCodegen::new_allocator(llmod_id, module)) + }) + } else { + None + }; + + let ongoing_codegen = start_async_codegen(backend.clone(), tcx, target_cpu, allocator_module); // For better throughput during parallel processing by LLVM, we used to sort // CGUs largest to smallest. This would lead to better thread utilization @@ -795,35 +832,6 @@ pub fn codegen_crate<B: ExtraBackendMethods>( } } - // Codegen an allocator shim, if necessary. - // Do this last to ensure the LLVM_passes timer doesn't start while no CGUs have been codegened - // yet for the backend to optimize. - if let Some(kind) = allocator_kind_for_codegen(tcx) { - let llmod_id = - cgu_name_builder.build_cgu_name(LOCAL_CRATE, &["crate"], Some("allocator")).to_string(); - let module_llvm = tcx.sess.time("write_allocator_module", || { - backend.codegen_allocator( - tcx, - &llmod_id, - kind, - // If allocator_kind is Some then alloc_error_handler_kind must - // also be Some. - tcx.alloc_error_handler_kind(()).unwrap(), - ) - }); - - ongoing_codegen.wait_for_signal_to_codegen_item(); - ongoing_codegen.check_for_errors(tcx.sess); - - // These modules are generally cheap and won't throw off scheduling. - let cost = 0; - submit_codegened_module_to_llvm( - &ongoing_codegen.coordinator, - ModuleCodegen::new_allocator(llmod_id, module_llvm), - cost, - ); - } - ongoing_codegen.codegen_finished(tcx); // Since the main thread is sometimes blocked during codegen, we keep track diff --git a/compiler/rustc_codegen_ssa/src/codegen_attrs.rs b/compiler/rustc_codegen_ssa/src/codegen_attrs.rs index 961bb788149..008340e614d 100644 --- a/compiler/rustc_codegen_ssa/src/codegen_attrs.rs +++ b/compiler/rustc_codegen_ssa/src/codegen_attrs.rs @@ -428,9 +428,16 @@ fn check_result( // llvm/llvm-project#70563). if !codegen_fn_attrs.target_features.is_empty() && matches!(codegen_fn_attrs.inline, InlineAttr::Always) + && !tcx.features().target_feature_inline_always() && let Some(span) = interesting_spans.inline { - tcx.dcx().span_err(span, "cannot use `#[inline(always)]` with `#[target_feature]`"); + feature_err( + tcx.sess, + sym::target_feature_inline_always, + span, + "cannot use `#[inline(always)]` with `#[target_feature]`", + ) + .emit(); } // warn that inline has no effect when no_sanitize is present diff --git a/compiler/rustc_codegen_ssa/src/lib.rs b/compiler/rustc_codegen_ssa/src/lib.rs index fe0500a5d4c..baba8f9ca3e 100644 --- a/compiler/rustc_codegen_ssa/src/lib.rs +++ b/compiler/rustc_codegen_ssa/src/lib.rs @@ -119,7 +119,7 @@ impl<M> ModuleCodegen<M> { }); CompiledModule { - name: self.name.clone(), + name: self.name, kind: self.kind, object, dwarf_object, diff --git a/compiler/rustc_codegen_ssa/src/mir/rvalue.rs b/compiler/rustc_codegen_ssa/src/mir/rvalue.rs index 8a67b8d6e5f..f6f2e3f2a3a 100644 --- a/compiler/rustc_codegen_ssa/src/mir/rvalue.rs +++ b/compiler/rustc_codegen_ssa/src/mir/rvalue.rs @@ -1,3 +1,4 @@ +use itertools::Itertools as _; use rustc_abi::{self as abi, FIRST_VARIANT}; use rustc_middle::ty::adjustment::PointerCoercion; use rustc_middle::ty::layout::{HasTyCtxt, HasTypingEnv, LayoutOf, TyAndLayout}; @@ -111,14 +112,13 @@ impl<'a, 'tcx, Bx: BuilderMethods<'a, 'tcx>> FunctionCx<'a, 'tcx, Bx> { let size = bx.const_usize(dest.layout.size.bytes()); // Use llvm.memset.p0i8.* to initialize all same byte arrays - if let Some(int) = bx.cx().const_to_opt_u128(v, false) { - let bytes = &int.to_le_bytes()[..cg_elem.layout.size.bytes_usize()]; - let first = bytes[0]; - if bytes[1..].iter().all(|&b| b == first) { - let fill = bx.cx().const_u8(first); - bx.memset(start, fill, size, dest.val.align, MemFlags::empty()); - return true; - } + if let Some(int) = bx.cx().const_to_opt_u128(v, false) + && let bytes = &int.to_le_bytes()[..cg_elem.layout.size.bytes_usize()] + && let Ok(&byte) = bytes.iter().all_equal_value() + { + let fill = bx.cx().const_u8(byte); + bx.memset(start, fill, size, dest.val.align, MemFlags::empty()); + return true; } // Use llvm.memset.p0i8.* to initialize byte arrays @@ -130,13 +130,10 @@ impl<'a, 'tcx, Bx: BuilderMethods<'a, 'tcx>> FunctionCx<'a, 'tcx, Bx> { false }; - match cg_elem.val { - OperandValue::Immediate(v) => { - if try_init_all_same(bx, v) { - return; - } - } - _ => (), + if let OperandValue::Immediate(v) = cg_elem.val + && try_init_all_same(bx, v) + { + return; } let count = self diff --git a/compiler/rustc_codegen_ssa/src/traits/write.rs b/compiler/rustc_codegen_ssa/src/traits/write.rs index cc7c4e46d7b..1ac1d7ef2e2 100644 --- a/compiler/rustc_codegen_ssa/src/traits/write.rs +++ b/compiler/rustc_codegen_ssa/src/traits/write.rs @@ -50,16 +50,12 @@ pub trait WriteBackendMethods: Clone + 'static { module: ModuleCodegen<Self::Module>, config: &ModuleConfig, ) -> CompiledModule; - fn prepare_thin( - module: ModuleCodegen<Self::Module>, - want_summary: bool, - ) -> (String, Self::ThinBuffer); + fn prepare_thin(module: ModuleCodegen<Self::Module>) -> (String, Self::ThinBuffer); fn serialize_module(module: ModuleCodegen<Self::Module>) -> (String, Self::ModuleBuffer); } pub trait ThinBufferMethods: Send + Sync { fn data(&self) -> &[u8]; - fn thin_link_data(&self) -> &[u8]; } pub trait ModuleBufferMethods: Send + Sync { diff --git a/compiler/rustc_const_eval/messages.ftl b/compiler/rustc_const_eval/messages.ftl index 60518dafbf2..700d7c26752 100644 --- a/compiler/rustc_const_eval/messages.ftl +++ b/compiler/rustc_const_eval/messages.ftl @@ -457,7 +457,7 @@ const_eval_validation_failure = it is undefined behavior to use this value const_eval_validation_failure_note = - The rules on what exactly is undefined behavior aren't clear, so this check might be overzealous. Please open an issue on the rustc repository if you believe it should not be considered undefined behavior. + the rules on what exactly is undefined behavior aren't clear, so this check might be overzealous. Please open an issue on the rustc repository if you believe it should not be considered undefined behavior. const_eval_validation_front_matter_invalid_value = constructing invalid value const_eval_validation_front_matter_invalid_value_with_path = constructing invalid value at {$path} diff --git a/compiler/rustc_driver_impl/messages.ftl b/compiler/rustc_driver_impl/messages.ftl index 2c6a0291ac2..b62cdc35f51 100644 --- a/compiler/rustc_driver_impl/messages.ftl +++ b/compiler/rustc_driver_impl/messages.ftl @@ -14,7 +14,7 @@ driver_impl_ice_version = rustc {$version} running on {$triple} driver_impl_rlink_corrupt_file = corrupt metadata encountered in `{$file}` -driver_impl_rlink_empty_version_number = The input does not contain version number +driver_impl_rlink_empty_version_number = the input does not contain version number driver_impl_rlink_encoding_version_mismatch = .rlink file was produced with encoding version `{$version_array}`, but the current version is `{$rlink_version}` @@ -24,6 +24,6 @@ driver_impl_rlink_rustc_version_mismatch = .rlink file was produced by rustc ver driver_impl_rlink_unable_to_read = failed to read rlink file: `{$err}` -driver_impl_rlink_wrong_file_type = The input does not look like a .rlink file +driver_impl_rlink_wrong_file_type = the input does not look like a .rlink file driver_impl_unstable_feature_usage = cannot dump feature usage metrics: {$error} diff --git a/compiler/rustc_errors/src/diagnostic.rs b/compiler/rustc_errors/src/diagnostic.rs index 96a4ed3218f..ae23ef1e255 100644 --- a/compiler/rustc_errors/src/diagnostic.rs +++ b/compiler/rustc_errors/src/diagnostic.rs @@ -945,11 +945,6 @@ impl<'a, G: EmissionGuarantee> Diag<'a, G> { None, "Span must not be empty and have no suggestion", ); - debug_assert_eq!( - parts.array_windows().find(|[a, b]| a.span.overlaps(b.span)), - None, - "suggestion must not have overlapping parts", - ); self.push_suggestion(CodeSuggestion { substitutions: vec![Substitution { parts }], diff --git a/compiler/rustc_errors/src/emitter.rs b/compiler/rustc_errors/src/emitter.rs index b94370e8e9b..93b1e6b7615 100644 --- a/compiler/rustc_errors/src/emitter.rs +++ b/compiler/rustc_errors/src/emitter.rs @@ -2354,7 +2354,6 @@ impl HumanEmitter { .sum(); let underline_start = (span_start_pos + start) as isize + offset; let underline_end = (span_start_pos + start + sub_len) as isize + offset; - assert!(underline_start >= 0 && underline_end >= 0); let padding: usize = max_line_num_len + 3; for p in underline_start..underline_end { if let DisplaySuggestion::Underline = show_code_change diff --git a/compiler/rustc_errors/src/lib.rs b/compiler/rustc_errors/src/lib.rs index 71fc54f0d33..8869799ce90 100644 --- a/compiler/rustc_errors/src/lib.rs +++ b/compiler/rustc_errors/src/lib.rs @@ -381,6 +381,17 @@ impl CodeSuggestion { // Assumption: all spans are in the same file, and all spans // are disjoint. Sort in ascending order. substitution.parts.sort_by_key(|part| part.span.lo()); + // Verify the assumption that all spans are disjoint + assert_eq!( + substitution.parts.array_windows().find(|[a, b]| a.span.overlaps(b.span)), + None, + "all spans must be disjoint", + ); + + // Account for cases where we are suggesting the same code that's already + // there. This shouldn't happen often, but in some cases for multipart + // suggestions it's much easier to handle it here than in the origin. + substitution.parts.retain(|p| is_different(sm, &p.snippet, p.span)); // Find the bounding span. let lo = substitution.parts.iter().map(|part| part.span.lo()).min()?; @@ -470,16 +481,12 @@ impl CodeSuggestion { _ => 1, }) .sum(); - if !is_different(sm, &part.snippet, part.span) { - // Account for cases where we are suggesting the same code that's already - // there. This shouldn't happen often, but in some cases for multipart - // suggestions it's much easier to handle it here than in the origin. - } else { - line_highlight.push(SubstitutionHighlight { - start: (cur_lo.col.0 as isize + acc) as usize, - end: (cur_lo.col.0 as isize + acc + len) as usize, - }); - } + + line_highlight.push(SubstitutionHighlight { + start: (cur_lo.col.0 as isize + acc) as usize, + end: (cur_lo.col.0 as isize + acc + len) as usize, + }); + buf.push_str(&part.snippet); let cur_hi = sm.lookup_char_pos(part.span.hi()); // Account for the difference between the width of the current code and the @@ -1160,7 +1167,7 @@ impl<'a> DiagCtxtHandle<'a> { // - It's only produce with JSON output. // - It's not emitted the usual way, via `emit_diagnostic`. // - The `$message_type` field is "unused_externs" rather than the usual - // "diagnosic". + // "diagnostic". // // We count it as a lint error because it has a lint level. The value // of `loud` (which comes from "unused-externs" or diff --git a/compiler/rustc_expand/src/mbe/macro_parser.rs b/compiler/rustc_expand/src/mbe/macro_parser.rs index 0324057e331..ab8e059b7b7 100644 --- a/compiler/rustc_expand/src/mbe/macro_parser.rs +++ b/compiler/rustc_expand/src/mbe/macro_parser.rs @@ -77,7 +77,7 @@ use std::rc::Rc; pub(crate) use NamedMatch::*; pub(crate) use ParseResult::*; -use rustc_ast::token::{self, DocComment, NonterminalKind, Token}; +use rustc_ast::token::{self, DocComment, NonterminalKind, Token, TokenKind}; use rustc_data_structures::fx::FxHashMap; use rustc_errors::ErrorGuaranteed; use rustc_lint_defs::pluralize; @@ -397,7 +397,23 @@ fn token_name_eq(t1: &Token, t2: &Token) -> bool { { ident1.name == ident2.name && is_raw1 == is_raw2 } else { - t1.kind == t2.kind + // Note: we SHOULD NOT use `t1.kind == t2.kind` here, and we should instead compare the + // tokens using the special comparison logic below. + // It makes sure that variants containing `InvisibleOrigin` will + // never compare equal to one another. + // + // When we had AST-based nonterminals we couldn't compare them, and the + // old `Nonterminal` type had an `eq` that always returned false, + // resulting in this restriction: + // <https://doc.rust-lang.org/nightly/reference/macros-by-example.html#forwarding-a-matched-fragment> + // This comparison logic emulates that behaviour. We could consider lifting this + // restriction now but there are still cases involving invisible + // delimiters that make it harder than it first appears. + match (t1.kind, t2.kind) { + (TokenKind::OpenInvisible(_) | TokenKind::CloseInvisible(_), _) + | (_, TokenKind::OpenInvisible(_) | TokenKind::CloseInvisible(_)) => false, + (a, b) => a == b, + } } } diff --git a/compiler/rustc_feature/src/unstable.rs b/compiler/rustc_feature/src/unstable.rs index 573e9aab055..4f35bf63a1a 100644 --- a/compiler/rustc_feature/src/unstable.rs +++ b/compiler/rustc_feature/src/unstable.rs @@ -642,6 +642,8 @@ declare_features! ( (unstable, super_let, "1.88.0", Some(139076)), /// Allows subtrait items to shadow supertrait items. (unstable, supertrait_item_shadowing, "1.86.0", Some(89151)), + /// Allows the use of target_feature when a function is marked inline(always). + (unstable, target_feature_inline_always, "CURRENT_RUSTC_VERSION", Some(145574)), /// Allows using `#[thread_local]` on `static` items. (unstable, thread_local, "1.0.0", Some(29594)), /// Allows defining `trait X = A + B;` alias items. diff --git a/compiler/rustc_hir_analysis/src/check/intrinsic.rs b/compiler/rustc_hir_analysis/src/check/intrinsic.rs index cfc6bc2f3a0..aa2d27ab809 100644 --- a/compiler/rustc_hir_analysis/src/check/intrinsic.rs +++ b/compiler/rustc_hir_analysis/src/check/intrinsic.rs @@ -449,6 +449,9 @@ pub(crate) fn check_intrinsic_type( } sym::unchecked_shl | sym::unchecked_shr => (2, 0, vec![param(0), param(1)], param(0)), sym::rotate_left | sym::rotate_right => (1, 0, vec![param(0), tcx.types.u32], param(0)), + sym::unchecked_funnel_shl | sym::unchecked_funnel_shr => { + (1, 0, vec![param(0), param(0), tcx.types.u32], param(0)) + } sym::unchecked_add | sym::unchecked_sub | sym::unchecked_mul => { (1, 0, vec![param(0), param(0)], param(0)) } diff --git a/compiler/rustc_hir_analysis/src/check/region.rs b/compiler/rustc_hir_analysis/src/check/region.rs index f5770b7312d..2ba7ed46f92 100644 --- a/compiler/rustc_hir_analysis/src/check/region.rs +++ b/compiler/rustc_hir_analysis/src/check/region.rs @@ -490,12 +490,8 @@ fn resolve_local<'tcx>( // // Iterate up to the enclosing destruction scope to find the same scope that will also // be used for the result of the block itself. - while let Some(s) = visitor.cx.var_parent { - let parent = visitor.scope_tree.parent_map.get(&s).cloned(); - if let Some(Scope { data: ScopeData::Destruction, .. }) = parent { - break; - } - visitor.cx.var_parent = parent; + if let Some(inner_scope) = visitor.cx.var_parent { + (visitor.cx.var_parent, _) = visitor.scope_tree.default_temporary_scope(inner_scope) } } } diff --git a/compiler/rustc_hir_typeck/src/coercion.rs b/compiler/rustc_hir_typeck/src/coercion.rs index e66601631fc..b99f811db1a 100644 --- a/compiler/rustc_hir_typeck/src/coercion.rs +++ b/compiler/rustc_hir_typeck/src/coercion.rs @@ -1897,7 +1897,7 @@ impl<'tcx, 'exprs, E: AsCoercionSite> CoerceMany<'tcx, 'exprs, E> { fcx.suggest_semicolon_at_end(cond_expr.span, &mut err); } } - }; + } // If this is due to an explicit `return`, suggest adding a return type. if let Some((fn_id, fn_decl)) = fcx.get_fn_decl(block_or_return_id) diff --git a/compiler/rustc_hir_typeck/src/expr.rs b/compiler/rustc_hir_typeck/src/expr.rs index a652e08905a..7adbee7ff28 100644 --- a/compiler/rustc_hir_typeck/src/expr.rs +++ b/compiler/rustc_hir_typeck/src/expr.rs @@ -2585,12 +2585,11 @@ impl<'a, 'tcx> FnCtxt<'a, 'tcx> { .filter(|item| item.is_fn() && !item.is_method()) .filter_map(|item| { // Only assoc fns that return `Self` - let fn_sig = self.tcx.fn_sig(item.def_id).skip_binder(); - let ret_ty = fn_sig.output(); - let ret_ty = self.tcx.normalize_erasing_late_bound_regions( - self.typing_env(self.param_env), - ret_ty, - ); + let fn_sig = self + .tcx + .fn_sig(item.def_id) + .instantiate(self.tcx, self.fresh_args_for_item(span, item.def_id)); + let ret_ty = self.tcx.instantiate_bound_regions_with_erased(fn_sig.output()); if !self.can_eq(self.param_env, ret_ty, adt_ty) { return None; } diff --git a/compiler/rustc_hir_typeck/src/fn_ctxt/checks.rs b/compiler/rustc_hir_typeck/src/fn_ctxt/checks.rs index 5aec50c8b53..94b635c41b4 100644 --- a/compiler/rustc_hir_typeck/src/fn_ctxt/checks.rs +++ b/compiler/rustc_hir_typeck/src/fn_ctxt/checks.rs @@ -1912,7 +1912,22 @@ impl<'a, 'tcx> FnCtxt<'a, 'tcx> { hir::StmtKind::Expr(ref expr) => { // Check with expected type of `()`. self.check_expr_has_type_or_error(expr, self.tcx.types.unit, |err| { - if expr.can_have_side_effects() { + if self.is_next_stmt_expr_continuation(stmt.hir_id) + && let hir::ExprKind::Match(..) | hir::ExprKind::If(..) = expr.kind + { + // We have something like `match () { _ => true } && true`. Suggest + // wrapping in parentheses. We find the statement or expression + // following the `match` (`&& true`) and see if it is something that + // can reasonably be interpreted as a binop following an expression. + err.multipart_suggestion( + "parentheses are required to parse this as an expression", + vec![ + (expr.span.shrink_to_lo(), "(".to_string()), + (expr.span.shrink_to_hi(), ")".to_string()), + ], + Applicability::MachineApplicable, + ); + } else if expr.can_have_side_effects() { self.suggest_semicolon_at_end(expr.span, err); } }); diff --git a/compiler/rustc_hir_typeck/src/fn_ctxt/suggestions.rs b/compiler/rustc_hir_typeck/src/fn_ctxt/suggestions.rs index aca3840712e..84ea2ec0f8a 100644 --- a/compiler/rustc_hir_typeck/src/fn_ctxt/suggestions.rs +++ b/compiler/rustc_hir_typeck/src/fn_ctxt/suggestions.rs @@ -1,3 +1,4 @@ +// ignore-tidy-filelength use core::cmp::min; use core::iter; @@ -766,56 +767,121 @@ impl<'a, 'tcx> FnCtxt<'a, 'tcx> { needs_block: bool, parent_is_closure: bool, ) { - if expected.is_unit() { - // `BlockTailExpression` only relevant if the tail expr would be - // useful on its own. - match expression.kind { - ExprKind::Call(..) - | ExprKind::MethodCall(..) - | ExprKind::Loop(..) - | ExprKind::If(..) - | ExprKind::Match(..) - | ExprKind::Block(..) - if expression.can_have_side_effects() - // If the expression is from an external macro, then do not suggest - // adding a semicolon, because there's nowhere to put it. - // See issue #81943. - && !expression.span.in_external_macro(self.tcx.sess.source_map()) => + if !expected.is_unit() { + return; + } + // `BlockTailExpression` only relevant if the tail expr would be + // useful on its own. + match expression.kind { + ExprKind::Call(..) + | ExprKind::MethodCall(..) + | ExprKind::Loop(..) + | ExprKind::If(..) + | ExprKind::Match(..) + | ExprKind::Block(..) + if expression.can_have_side_effects() + // If the expression is from an external macro, then do not suggest + // adding a semicolon, because there's nowhere to put it. + // See issue #81943. + && !expression.span.in_external_macro(self.tcx.sess.source_map()) => + { + if needs_block { + err.multipart_suggestion( + "consider using a semicolon here", + vec![ + (expression.span.shrink_to_lo(), "{ ".to_owned()), + (expression.span.shrink_to_hi(), "; }".to_owned()), + ], + Applicability::MachineApplicable, + ); + } else if let hir::Node::Block(block) = self.tcx.parent_hir_node(expression.hir_id) + && let hir::Node::Expr(expr) = self.tcx.parent_hir_node(block.hir_id) + && let hir::Node::Expr(if_expr) = self.tcx.parent_hir_node(expr.hir_id) + && let hir::ExprKind::If(_cond, _then, Some(_else)) = if_expr.kind + && let hir::Node::Stmt(stmt) = self.tcx.parent_hir_node(if_expr.hir_id) + && let hir::StmtKind::Expr(_) = stmt.kind + && self.is_next_stmt_expr_continuation(stmt.hir_id) { - if needs_block { - err.multipart_suggestion( - "consider using a semicolon here", - vec![ - (expression.span.shrink_to_lo(), "{ ".to_owned()), - (expression.span.shrink_to_hi(), "; }".to_owned()), - ], - Applicability::MachineApplicable, - ); - } else { - err.span_suggestion( - expression.span.shrink_to_hi(), - "consider using a semicolon here", - ";", - Applicability::MachineApplicable, - ); - } + err.multipart_suggestion( + "parentheses are required to parse this as an expression", + vec![ + (stmt.span.shrink_to_lo(), "(".to_string()), + (stmt.span.shrink_to_hi(), ")".to_string()), + ], + Applicability::MachineApplicable, + ); + } else { + err.span_suggestion( + expression.span.shrink_to_hi(), + "consider using a semicolon here", + ";", + Applicability::MachineApplicable, + ); } - ExprKind::Path(..) | ExprKind::Lit(_) - if parent_is_closure - && !expression.span.in_external_macro(self.tcx.sess.source_map()) => + } + ExprKind::Path(..) | ExprKind::Lit(_) + if parent_is_closure + && !expression.span.in_external_macro(self.tcx.sess.source_map()) => + { + err.span_suggestion_verbose( + expression.span.shrink_to_lo(), + "consider ignoring the value", + "_ = ", + Applicability::MachineApplicable, + ); + } + _ => { + if let hir::Node::Block(block) = self.tcx.parent_hir_node(expression.hir_id) + && let hir::Node::Expr(expr) = self.tcx.parent_hir_node(block.hir_id) + && let hir::Node::Expr(if_expr) = self.tcx.parent_hir_node(expr.hir_id) + && let hir::ExprKind::If(_cond, _then, Some(_else)) = if_expr.kind + && let hir::Node::Stmt(stmt) = self.tcx.parent_hir_node(if_expr.hir_id) + && let hir::StmtKind::Expr(_) = stmt.kind + && self.is_next_stmt_expr_continuation(stmt.hir_id) { - err.span_suggestion_verbose( - expression.span.shrink_to_lo(), - "consider ignoring the value", - "_ = ", + // The error is pointing at an arm of an if-expression, and we want to get the + // `Span` of the whole if-expression for the suggestion. This only works for a + // single level of nesting, which is fine. + // We have something like `if true { false } else { true } && true`. Suggest + // wrapping in parentheses. We find the statement or expression following the + // `if` (`&& true`) and see if it is something that can reasonably be + // interpreted as a binop following an expression. + err.multipart_suggestion( + "parentheses are required to parse this as an expression", + vec![ + (stmt.span.shrink_to_lo(), "(".to_string()), + (stmt.span.shrink_to_hi(), ")".to_string()), + ], Applicability::MachineApplicable, ); } - _ => (), } } } + pub(crate) fn is_next_stmt_expr_continuation(&self, hir_id: HirId) -> bool { + if let hir::Node::Block(b) = self.tcx.parent_hir_node(hir_id) + && let mut stmts = b.stmts.iter().skip_while(|s| s.hir_id != hir_id) + && let Some(_) = stmts.next() // The statement the statement that was passed in + && let Some(next) = match (stmts.next(), b.expr) { // The following statement + (Some(next), _) => match next.kind { + hir::StmtKind::Expr(next) | hir::StmtKind::Semi(next) => Some(next), + _ => None, + }, + (None, Some(next)) => Some(next), + _ => None, + } + && let hir::ExprKind::AddrOf(..) // prev_stmt && next + | hir::ExprKind::Unary(..) // prev_stmt * next + | hir::ExprKind::Err(_) = next.kind + // prev_stmt + next + { + true + } else { + false + } + } + /// A possible error is to forget to add a return type that is needed: /// /// ```compile_fail,E0308 diff --git a/compiler/rustc_index/src/interval.rs b/compiler/rustc_index/src/interval.rs index 0225c5c4f32..dda5253e7c5 100644 --- a/compiler/rustc_index/src/interval.rs +++ b/compiler/rustc_index/src/interval.rs @@ -140,6 +140,30 @@ impl<I: Idx> IntervalSet<I> { result } + /// Specialized version of `insert` when we know that the inserted point is *after* any + /// contained. + pub fn append(&mut self, point: I) { + let point = point.index() as u32; + + if let Some((_, last_end)) = self.map.last_mut() { + assert!(*last_end <= point); + if point == *last_end { + // The point is already in the set. + } else if point == *last_end + 1 { + *last_end = point; + } else { + self.map.push((point, point)); + } + } else { + self.map.push((point, point)); + } + + debug_assert!( + self.check_invariants(), + "wrong intervals after append {point:?} to {self:?}" + ); + } + pub fn contains(&self, needle: I) -> bool { let needle = needle.index() as u32; let Some(last) = self.map.partition_point(|r| r.0 <= needle).checked_sub(1) else { @@ -176,6 +200,32 @@ impl<I: Idx> IntervalSet<I> { }) } + pub fn disjoint(&self, other: &IntervalSet<I>) -> bool + where + I: Step, + { + let helper = move || { + let mut self_iter = self.iter_intervals(); + let mut other_iter = other.iter_intervals(); + + let mut self_current = self_iter.next()?; + let mut other_current = other_iter.next()?; + + loop { + if self_current.end <= other_current.start { + self_current = self_iter.next()?; + continue; + } + if other_current.end <= self_current.start { + other_current = other_iter.next()?; + continue; + } + return Some(false); + } + }; + helper().unwrap_or(true) + } + pub fn is_empty(&self) -> bool { self.map.is_empty() } @@ -325,6 +375,10 @@ impl<R: Idx, C: Step + Idx> SparseIntervalMatrix<R, C> { self.ensure_row(row).insert(point) } + pub fn append(&mut self, row: R, point: C) { + self.ensure_row(row).append(point) + } + pub fn contains(&self, row: R, point: C) -> bool { self.row(row).is_some_and(|r| r.contains(point)) } diff --git a/compiler/rustc_infer/src/infer/context.rs b/compiler/rustc_infer/src/infer/context.rs index bb9c8850093..8265fccabc9 100644 --- a/compiler/rustc_infer/src/infer/context.rs +++ b/compiler/rustc_infer/src/infer/context.rs @@ -1,4 +1,4 @@ -///! Definition of `InferCtxtLike` from the librarified type layer. +//! Definition of `InferCtxtLike` from the librarified type layer. use rustc_hir::def_id::DefId; use rustc_middle::traits::ObligationCause; use rustc_middle::ty::relate::RelateResult; diff --git a/compiler/rustc_interface/src/interface.rs b/compiler/rustc_interface/src/interface.rs index 8f131f45bbd..4c820b8877b 100644 --- a/compiler/rustc_interface/src/interface.rs +++ b/compiler/rustc_interface/src/interface.rs @@ -13,7 +13,7 @@ use rustc_lint::LintStore; use rustc_middle::ty; use rustc_middle::ty::CurrentGcx; use rustc_middle::util::Providers; -use rustc_parse::new_parser_from_source_str; +use rustc_parse::new_parser_from_simple_source_str; use rustc_parse::parser::attr::AllowLeadingUnsafe; use rustc_query_impl::QueryCtxt; use rustc_query_system::query::print_query_stack; @@ -68,7 +68,7 @@ pub(crate) fn parse_cfg(dcx: DiagCtxtHandle<'_>, cfgs: Vec<String>) -> Cfg { }; } - match new_parser_from_source_str(&psess, filename, s.to_string()) { + match new_parser_from_simple_source_str(&psess, filename, s.to_string()) { Ok(mut parser) => match parser.parse_meta_item(AllowLeadingUnsafe::No) { Ok(meta_item) if parser.token == token::Eof => { if meta_item.path.segments.len() != 1 { @@ -166,7 +166,7 @@ pub(crate) fn parse_check_cfg(dcx: DiagCtxtHandle<'_>, specs: Vec<String>) -> Ch error!("expected `cfg(name, values(\"value1\", \"value2\", ... \"valueN\"))`") }; - let mut parser = match new_parser_from_source_str(&psess, filename, s.to_string()) { + let mut parser = match new_parser_from_simple_source_str(&psess, filename, s.to_string()) { Ok(parser) => parser, Err(errs) => { errs.into_iter().for_each(|err| err.cancel()); diff --git a/compiler/rustc_lexer/src/lib.rs b/compiler/rustc_lexer/src/lib.rs index d10b1920343..c29ab569f47 100644 --- a/compiler/rustc_lexer/src/lib.rs +++ b/compiler/rustc_lexer/src/lib.rs @@ -331,24 +331,37 @@ pub fn is_whitespace(c: char) -> bool { matches!( c, - // Usual ASCII suspects - '\u{0009}' // \t - | '\u{000A}' // \n + // End-of-line characters + | '\u{000A}' // line feed (\n) | '\u{000B}' // vertical tab | '\u{000C}' // form feed - | '\u{000D}' // \r - | '\u{0020}' // space - - // NEXT LINE from latin1 - | '\u{0085}' + | '\u{000D}' // carriage return (\r) + | '\u{0085}' // next line (from latin1) + | '\u{2028}' // LINE SEPARATOR + | '\u{2029}' // PARAGRAPH SEPARATOR - // Bidi markers + // `Default_Ignorable_Code_Point` characters | '\u{200E}' // LEFT-TO-RIGHT MARK | '\u{200F}' // RIGHT-TO-LEFT MARK - // Dedicated whitespace characters from Unicode - | '\u{2028}' // LINE SEPARATOR - | '\u{2029}' // PARAGRAPH SEPARATOR + // Horizontal space characters + | '\u{0009}' // tab (\t) + | '\u{0020}' // space + ) +} + +/// True if `c` is considered horizontal whitespace according to Rust language definition. +pub fn is_horizontal_whitespace(c: char) -> bool { + // This is Pattern_White_Space. + // + // Note that this set is stable (ie, it doesn't change with different + // Unicode versions), so it's ok to just hard-code the values. + + matches!( + c, + // Horizontal space characters + '\u{0009}' // tab (\t) + | '\u{0020}' // space ) } @@ -538,7 +551,7 @@ impl Cursor<'_> { debug_assert!(length_opening >= 3); // whitespace between the opening and the infostring. - self.eat_while(|ch| ch != '\n' && is_whitespace(ch)); + self.eat_while(|ch| ch != '\n' && is_horizontal_whitespace(ch)); // copied from `eat_identifier`, but allows `-` and `.` in infostring to allow something like // `---Cargo.toml` as a valid opener @@ -547,7 +560,7 @@ impl Cursor<'_> { self.eat_while(|c| is_id_continue(c) || c == '-' || c == '.'); } - self.eat_while(|ch| ch != '\n' && is_whitespace(ch)); + self.eat_while(|ch| ch != '\n' && is_horizontal_whitespace(ch)); let invalid_infostring = self.first() != '\n'; let mut found = false; @@ -588,7 +601,7 @@ impl Cursor<'_> { // on a standalone line. Might be wrong. while let Some(closing) = rest.find("---") { let preceding_chars_start = rest[..closing].rfind("\n").map_or(0, |i| i + 1); - if rest[preceding_chars_start..closing].chars().all(is_whitespace) { + if rest[preceding_chars_start..closing].chars().all(is_horizontal_whitespace) { // candidate found potential_closing = Some(closing); break; diff --git a/compiler/rustc_lint/Cargo.toml b/compiler/rustc_lint/Cargo.toml index 7718f16984d..3a50aac50cb 100644 --- a/compiler/rustc_lint/Cargo.toml +++ b/compiler/rustc_lint/Cargo.toml @@ -5,6 +5,7 @@ edition = "2024" [dependencies] # tidy-alphabetical-start +bitflags = "2.4.1" rustc_abi = { path = "../rustc_abi" } rustc_ast = { path = "../rustc_ast" } rustc_ast_pretty = { path = "../rustc_ast_pretty" } diff --git a/compiler/rustc_lint/src/foreign_modules.rs b/compiler/rustc_lint/src/foreign_modules.rs index 3267e70f1de..ad73e15e31f 100644 --- a/compiler/rustc_lint/src/foreign_modules.rs +++ b/compiler/rustc_lint/src/foreign_modules.rs @@ -136,7 +136,6 @@ impl ClashingExternDeclarations { ty::TypingEnv::non_body_analysis(tcx, this_fi.owner_id), existing_decl_ty, this_decl_ty, - types::CItemKind::Declaration, ) { let orig = name_of_extern_decl(tcx, existing_did); @@ -214,10 +213,9 @@ fn structurally_same_type<'tcx>( typing_env: ty::TypingEnv<'tcx>, a: Ty<'tcx>, b: Ty<'tcx>, - ckind: types::CItemKind, ) -> bool { let mut seen_types = UnordSet::default(); - let result = structurally_same_type_impl(&mut seen_types, tcx, typing_env, a, b, ckind); + let result = structurally_same_type_impl(&mut seen_types, tcx, typing_env, a, b); if cfg!(debug_assertions) && result { // Sanity-check: must have same ABI, size and alignment. // `extern` blocks cannot be generic, so we'll always get a layout here. @@ -236,7 +234,6 @@ fn structurally_same_type_impl<'tcx>( typing_env: ty::TypingEnv<'tcx>, a: Ty<'tcx>, b: Ty<'tcx>, - ckind: types::CItemKind, ) -> bool { debug!("structurally_same_type_impl(tcx, a = {:?}, b = {:?})", a, b); @@ -307,7 +304,6 @@ fn structurally_same_type_impl<'tcx>( typing_env, tcx.type_of(a_did).instantiate(tcx, a_gen_args), tcx.type_of(b_did).instantiate(tcx, b_gen_args), - ckind, ) }, ) @@ -315,25 +311,19 @@ fn structurally_same_type_impl<'tcx>( (ty::Array(a_ty, a_len), ty::Array(b_ty, b_len)) => { // For arrays, we also check the length. a_len == b_len - && structurally_same_type_impl( - seen_types, tcx, typing_env, *a_ty, *b_ty, ckind, - ) + && structurally_same_type_impl(seen_types, tcx, typing_env, *a_ty, *b_ty) } (ty::Slice(a_ty), ty::Slice(b_ty)) => { - structurally_same_type_impl(seen_types, tcx, typing_env, *a_ty, *b_ty, ckind) + structurally_same_type_impl(seen_types, tcx, typing_env, *a_ty, *b_ty) } (ty::RawPtr(a_ty, a_mutbl), ty::RawPtr(b_ty, b_mutbl)) => { a_mutbl == b_mutbl - && structurally_same_type_impl( - seen_types, tcx, typing_env, *a_ty, *b_ty, ckind, - ) + && structurally_same_type_impl(seen_types, tcx, typing_env, *a_ty, *b_ty) } (ty::Ref(_a_region, a_ty, a_mut), ty::Ref(_b_region, b_ty, b_mut)) => { // For structural sameness, we don't need the region to be same. a_mut == b_mut - && structurally_same_type_impl( - seen_types, tcx, typing_env, *a_ty, *b_ty, ckind, - ) + && structurally_same_type_impl(seen_types, tcx, typing_env, *a_ty, *b_ty) } (ty::FnDef(..), ty::FnDef(..)) => { let a_poly_sig = a.fn_sig(tcx); @@ -347,7 +337,7 @@ fn structurally_same_type_impl<'tcx>( (a_sig.abi, a_sig.safety, a_sig.c_variadic) == (b_sig.abi, b_sig.safety, b_sig.c_variadic) && a_sig.inputs().iter().eq_by(b_sig.inputs().iter(), |a, b| { - structurally_same_type_impl(seen_types, tcx, typing_env, *a, *b, ckind) + structurally_same_type_impl(seen_types, tcx, typing_env, *a, *b) }) && structurally_same_type_impl( seen_types, @@ -355,7 +345,6 @@ fn structurally_same_type_impl<'tcx>( typing_env, a_sig.output(), b_sig.output(), - ckind, ) } (ty::Tuple(..), ty::Tuple(..)) => { @@ -383,14 +372,14 @@ fn structurally_same_type_impl<'tcx>( // An Adt and a primitive or pointer type. This can be FFI-safe if non-null // enum layout optimisation is being applied. (ty::Adt(..) | ty::Pat(..), _) if is_primitive_or_pointer(b) => { - if let Some(a_inner) = types::repr_nullable_ptr(tcx, typing_env, a, ckind) { + if let Some(a_inner) = types::repr_nullable_ptr(tcx, typing_env, a) { a_inner == b } else { false } } (_, ty::Adt(..) | ty::Pat(..)) if is_primitive_or_pointer(a) => { - if let Some(b_inner) = types::repr_nullable_ptr(tcx, typing_env, b, ckind) { + if let Some(b_inner) = types::repr_nullable_ptr(tcx, typing_env, b) { b_inner == a } else { false diff --git a/compiler/rustc_lint/src/lib.rs b/compiler/rustc_lint/src/lib.rs index bdbac7fc4d1..9bb53fea54a 100644 --- a/compiler/rustc_lint/src/lib.rs +++ b/compiler/rustc_lint/src/lib.rs @@ -194,8 +194,7 @@ late_lint_methods!( DefaultCouldBeDerived: DefaultCouldBeDerived::default(), DerefIntoDynSupertrait: DerefIntoDynSupertrait, DropForgetUseless: DropForgetUseless, - ImproperCTypesDeclarations: ImproperCTypesDeclarations, - ImproperCTypesDefinitions: ImproperCTypesDefinitions, + ImproperCTypesLint: ImproperCTypesLint, InvalidFromUtf8: InvalidFromUtf8, VariantSizeDifferences: VariantSizeDifferences, PathStatements: PathStatements, diff --git a/compiler/rustc_lint/src/types.rs b/compiler/rustc_lint/src/types.rs index f8a692313f0..a72b802eb5d 100644 --- a/compiler/rustc_lint/src/types.rs +++ b/compiler/rustc_lint/src/types.rs @@ -1,35 +1,28 @@ use std::iter; -use std::ops::ControlFlow; -use rustc_abi::{BackendRepr, TagEncoding, VariantIdx, Variants, WrappingRange}; -use rustc_data_structures::fx::FxHashSet; -use rustc_errors::DiagMessage; -use rustc_hir::intravisit::VisitorExt; -use rustc_hir::{AmbigArg, Expr, ExprKind, HirId, LangItem}; +use rustc_abi::{BackendRepr, TagEncoding, Variants, WrappingRange}; +use rustc_hir::{Expr, ExprKind, HirId, LangItem}; use rustc_middle::bug; use rustc_middle::ty::layout::{LayoutOf, SizeSkeleton}; -use rustc_middle::ty::{ - self, Adt, AdtKind, GenericArgsRef, Ty, TyCtxt, TypeSuperVisitable, TypeVisitable, - TypeVisitableExt, -}; +use rustc_middle::ty::{self, Ty, TyCtxt, TypeVisitableExt}; use rustc_session::{declare_lint, declare_lint_pass, impl_lint_pass}; -use rustc_span::def_id::LocalDefId; use rustc_span::{Span, Symbol, sym}; use tracing::debug; use {rustc_ast as ast, rustc_hir as hir}; -mod improper_ctypes; +mod improper_ctypes; // these filed do the implementation for ImproperCTypesDefinitions,ImproperCTypesDeclarations +pub(crate) use improper_ctypes::ImproperCTypesLint; use crate::lints::{ AmbiguousWidePointerComparisons, AmbiguousWidePointerComparisonsAddrMetadataSuggestion, AmbiguousWidePointerComparisonsAddrSuggestion, AmbiguousWidePointerComparisonsCastSuggestion, AmbiguousWidePointerComparisonsExpectSuggestion, AtomicOrderingFence, AtomicOrderingLoad, - AtomicOrderingStore, ImproperCTypes, InvalidAtomicOrderingDiag, InvalidNanComparisons, + AtomicOrderingStore, InvalidAtomicOrderingDiag, InvalidNanComparisons, InvalidNanComparisonsSuggestion, UnpredictableFunctionPointerComparisons, - UnpredictableFunctionPointerComparisonsSuggestion, UnusedComparisons, UsesPowerAlignment, + UnpredictableFunctionPointerComparisonsSuggestion, UnusedComparisons, VariantSizeDifferencesDiag, }; -use crate::{LateContext, LateLintPass, LintContext, fluent_generated as fluent}; +use crate::{LateContext, LateLintPass, LintContext}; mod literal; @@ -690,144 +683,6 @@ impl<'tcx> LateLintPass<'tcx> for TypeLimits { } } -declare_lint! { - /// The `improper_ctypes` lint detects incorrect use of types in foreign - /// modules. - /// - /// ### Example - /// - /// ```rust - /// unsafe extern "C" { - /// static STATIC: String; - /// } - /// ``` - /// - /// {{produces}} - /// - /// ### Explanation - /// - /// The compiler has several checks to verify that types used in `extern` - /// blocks are safe and follow certain rules to ensure proper - /// compatibility with the foreign interfaces. This lint is issued when it - /// detects a probable mistake in a definition. The lint usually should - /// provide a description of the issue, along with possibly a hint on how - /// to resolve it. - IMPROPER_CTYPES, - Warn, - "proper use of libc types in foreign modules" -} - -declare_lint_pass!(ImproperCTypesDeclarations => [IMPROPER_CTYPES]); - -declare_lint! { - /// The `improper_ctypes_definitions` lint detects incorrect use of - /// [`extern` function] definitions. - /// - /// [`extern` function]: https://doc.rust-lang.org/reference/items/functions.html#extern-function-qualifier - /// - /// ### Example - /// - /// ```rust - /// # #![allow(unused)] - /// pub extern "C" fn str_type(p: &str) { } - /// ``` - /// - /// {{produces}} - /// - /// ### Explanation - /// - /// There are many parameter and return types that may be specified in an - /// `extern` function that are not compatible with the given ABI. This - /// lint is an alert that these types should not be used. The lint usually - /// should provide a description of the issue, along with possibly a hint - /// on how to resolve it. - IMPROPER_CTYPES_DEFINITIONS, - Warn, - "proper use of libc types in foreign item definitions" -} - -declare_lint! { - /// The `uses_power_alignment` lint detects specific `repr(C)` - /// aggregates on AIX. - /// In its platform C ABI, AIX uses the "power" (as in PowerPC) alignment - /// rule (detailed in https://www.ibm.com/docs/en/xl-c-and-cpp-aix/16.1?topic=data-using-alignment-modes#alignment), - /// which can also be set for XLC by `#pragma align(power)` or - /// `-qalign=power`. Aggregates with a floating-point type as the - /// recursively first field (as in "at offset 0") modify the layout of - /// *subsequent* fields of the associated structs to use an alignment value - /// where the floating-point type is aligned on a 4-byte boundary. - /// - /// Effectively, subsequent floating-point fields act as-if they are `repr(packed(4))`. This - /// would be unsound to do in a `repr(C)` type without all the restrictions that come with - /// `repr(packed)`. Rust instead chooses a layout that maintains soundness of Rust code, at the - /// expense of incompatibility with C code. - /// - /// ### Example - /// - /// ```rust,ignore (fails on non-powerpc64-ibm-aix) - /// #[repr(C)] - /// pub struct Floats { - /// a: f64, - /// b: u8, - /// c: f64, - /// } - /// ``` - /// - /// This will produce: - /// - /// ```text - /// warning: repr(C) does not follow the power alignment rule. This may affect platform C ABI compatibility for this type - /// --> <source>:5:3 - /// | - /// 5 | c: f64, - /// | ^^^^^^ - /// | - /// = note: `#[warn(uses_power_alignment)]` on by default - /// ``` - /// - /// ### Explanation - /// - /// The power alignment rule specifies that the above struct has the - /// following alignment: - /// - offset_of!(Floats, a) == 0 - /// - offset_of!(Floats, b) == 8 - /// - offset_of!(Floats, c) == 12 - /// - /// However, Rust currently aligns `c` at `offset_of!(Floats, c) == 16`. - /// Using offset 12 would be unsound since `f64` generally must be 8-aligned on this target. - /// Thus, a warning is produced for the above struct. - USES_POWER_ALIGNMENT, - Warn, - "Structs do not follow the power alignment rule under repr(C)" -} - -declare_lint_pass!(ImproperCTypesDefinitions => [IMPROPER_CTYPES_DEFINITIONS, USES_POWER_ALIGNMENT]); - -#[derive(Clone, Copy)] -pub(crate) enum CItemKind { - Declaration, - Definition, -} - -struct ImproperCTypesVisitor<'a, 'tcx> { - cx: &'a LateContext<'tcx>, - mode: CItemKind, -} - -/// Accumulator for recursive ffi type checking -struct CTypesVisitorState<'tcx> { - cache: FxHashSet<Ty<'tcx>>, - /// The original type being checked, before we recursed - /// to any other types it contains. - base_ty: Ty<'tcx>, -} - -enum FfiResult<'tcx> { - FfiSafe, - FfiPhantom(Ty<'tcx>), - FfiUnsafe { ty: Ty<'tcx>, reason: DiagMessage, help: Option<DiagMessage> }, -} - pub(crate) fn nonnull_optimization_guaranteed<'tcx>( tcx: TyCtxt<'tcx>, def: ty::AdtDef<'tcx>, @@ -855,14 +710,13 @@ fn ty_is_known_nonnull<'tcx>( tcx: TyCtxt<'tcx>, typing_env: ty::TypingEnv<'tcx>, ty: Ty<'tcx>, - mode: CItemKind, ) -> bool { let ty = tcx.try_normalize_erasing_regions(typing_env, ty).unwrap_or(ty); match ty.kind() { ty::FnPtr(..) => true, ty::Ref(..) => true, - ty::Adt(def, _) if def.is_box() && matches!(mode, CItemKind::Definition) => true, + ty::Adt(def, _) if def.is_box() => true, ty::Adt(def, args) if def.repr().transparent() && !def.is_union() => { let marked_non_null = nonnull_optimization_guaranteed(tcx, *def); @@ -878,10 +732,10 @@ fn ty_is_known_nonnull<'tcx>( def.variants() .iter() .filter_map(|variant| transparent_newtype_field(tcx, variant)) - .any(|field| ty_is_known_nonnull(tcx, typing_env, field.ty(tcx, args), mode)) + .any(|field| ty_is_known_nonnull(tcx, typing_env, field.ty(tcx, args))) } ty::Pat(base, pat) => { - ty_is_known_nonnull(tcx, typing_env, *base, mode) + ty_is_known_nonnull(tcx, typing_env, *base) || pat_ty_is_known_nonnull(tcx, typing_env, *pat) } _ => false, @@ -992,7 +846,6 @@ pub(crate) fn repr_nullable_ptr<'tcx>( tcx: TyCtxt<'tcx>, typing_env: ty::TypingEnv<'tcx>, ty: Ty<'tcx>, - ckind: CItemKind, ) -> Option<Ty<'tcx>> { debug!("is_repr_nullable_ptr(tcx, ty = {:?})", ty); match ty.kind() { @@ -1017,7 +870,7 @@ pub(crate) fn repr_nullable_ptr<'tcx>( _ => return None, }; - if !ty_is_known_nonnull(tcx, typing_env, field_ty, ckind) { + if !ty_is_known_nonnull(tcx, typing_env, field_ty) { return None; } @@ -1076,710 +929,6 @@ fn get_nullable_type_from_pat<'tcx>( } } -impl<'a, 'tcx> ImproperCTypesVisitor<'a, 'tcx> { - /// Check if the type is array and emit an unsafe type lint. - fn check_for_array_ty(&mut self, sp: Span, ty: Ty<'tcx>) -> bool { - if let ty::Array(..) = ty.kind() { - self.emit_ffi_unsafe_type_lint( - ty, - sp, - fluent::lint_improper_ctypes_array_reason, - Some(fluent::lint_improper_ctypes_array_help), - ); - true - } else { - false - } - } - - /// Checks if the given field's type is "ffi-safe". - fn check_field_type_for_ffi( - &self, - acc: &mut CTypesVisitorState<'tcx>, - field: &ty::FieldDef, - args: GenericArgsRef<'tcx>, - ) -> FfiResult<'tcx> { - let field_ty = field.ty(self.cx.tcx, args); - let field_ty = self - .cx - .tcx - .try_normalize_erasing_regions(self.cx.typing_env(), field_ty) - .unwrap_or(field_ty); - self.check_type_for_ffi(acc, field_ty) - } - - /// Checks if the given `VariantDef`'s field types are "ffi-safe". - fn check_variant_for_ffi( - &self, - acc: &mut CTypesVisitorState<'tcx>, - ty: Ty<'tcx>, - def: ty::AdtDef<'tcx>, - variant: &ty::VariantDef, - args: GenericArgsRef<'tcx>, - ) -> FfiResult<'tcx> { - use FfiResult::*; - let transparent_with_all_zst_fields = if def.repr().transparent() { - if let Some(field) = transparent_newtype_field(self.cx.tcx, variant) { - // Transparent newtypes have at most one non-ZST field which needs to be checked.. - match self.check_field_type_for_ffi(acc, field, args) { - FfiUnsafe { ty, .. } if ty.is_unit() => (), - r => return r, - } - - false - } else { - // ..or have only ZST fields, which is FFI-unsafe (unless those fields are all - // `PhantomData`). - true - } - } else { - false - }; - - // We can't completely trust `repr(C)` markings, so make sure the fields are actually safe. - let mut all_phantom = !variant.fields.is_empty(); - for field in &variant.fields { - all_phantom &= match self.check_field_type_for_ffi(acc, field, args) { - FfiSafe => false, - // `()` fields are FFI-safe! - FfiUnsafe { ty, .. } if ty.is_unit() => false, - FfiPhantom(..) => true, - r @ FfiUnsafe { .. } => return r, - } - } - - if all_phantom { - FfiPhantom(ty) - } else if transparent_with_all_zst_fields { - FfiUnsafe { ty, reason: fluent::lint_improper_ctypes_struct_zst, help: None } - } else { - FfiSafe - } - } - - /// Checks if the given type is "ffi-safe" (has a stable, well-defined - /// representation which can be exported to C code). - fn check_type_for_ffi( - &self, - acc: &mut CTypesVisitorState<'tcx>, - ty: Ty<'tcx>, - ) -> FfiResult<'tcx> { - use FfiResult::*; - - let tcx = self.cx.tcx; - - // Protect against infinite recursion, for example - // `struct S(*mut S);`. - // FIXME: A recursion limit is necessary as well, for irregular - // recursive types. - if !acc.cache.insert(ty) { - return FfiSafe; - } - - match *ty.kind() { - ty::Adt(def, args) => { - if let Some(boxed) = ty.boxed_ty() - && matches!(self.mode, CItemKind::Definition) - { - if boxed.is_sized(tcx, self.cx.typing_env()) { - return FfiSafe; - } else { - return FfiUnsafe { - ty, - reason: fluent::lint_improper_ctypes_box, - help: None, - }; - } - } - if def.is_phantom_data() { - return FfiPhantom(ty); - } - match def.adt_kind() { - AdtKind::Struct | AdtKind::Union => { - if let Some(sym::cstring_type | sym::cstr_type) = - tcx.get_diagnostic_name(def.did()) - && !acc.base_ty.is_mutable_ptr() - { - return FfiUnsafe { - ty, - reason: fluent::lint_improper_ctypes_cstr_reason, - help: Some(fluent::lint_improper_ctypes_cstr_help), - }; - } - - if !def.repr().c() && !def.repr().transparent() { - return FfiUnsafe { - ty, - reason: if def.is_struct() { - fluent::lint_improper_ctypes_struct_layout_reason - } else { - fluent::lint_improper_ctypes_union_layout_reason - }, - help: if def.is_struct() { - Some(fluent::lint_improper_ctypes_struct_layout_help) - } else { - Some(fluent::lint_improper_ctypes_union_layout_help) - }, - }; - } - - if def.non_enum_variant().field_list_has_applicable_non_exhaustive() { - return FfiUnsafe { - ty, - reason: if def.is_struct() { - fluent::lint_improper_ctypes_struct_non_exhaustive - } else { - fluent::lint_improper_ctypes_union_non_exhaustive - }, - help: None, - }; - } - - if def.non_enum_variant().fields.is_empty() { - return FfiUnsafe { - ty, - reason: if def.is_struct() { - fluent::lint_improper_ctypes_struct_fieldless_reason - } else { - fluent::lint_improper_ctypes_union_fieldless_reason - }, - help: if def.is_struct() { - Some(fluent::lint_improper_ctypes_struct_fieldless_help) - } else { - Some(fluent::lint_improper_ctypes_union_fieldless_help) - }, - }; - } - - self.check_variant_for_ffi(acc, ty, def, def.non_enum_variant(), args) - } - AdtKind::Enum => { - if def.variants().is_empty() { - // Empty enums are okay... although sort of useless. - return FfiSafe; - } - // Check for a repr() attribute to specify the size of the - // discriminant. - if !def.repr().c() && !def.repr().transparent() && def.repr().int.is_none() - { - // Special-case types like `Option<extern fn()>` and `Result<extern fn(), ()>` - if let Some(ty) = - repr_nullable_ptr(self.cx.tcx, self.cx.typing_env(), ty, self.mode) - { - return self.check_type_for_ffi(acc, ty); - } - - return FfiUnsafe { - ty, - reason: fluent::lint_improper_ctypes_enum_repr_reason, - help: Some(fluent::lint_improper_ctypes_enum_repr_help), - }; - } - - use improper_ctypes::check_non_exhaustive_variant; - - let non_exhaustive = def.variant_list_has_applicable_non_exhaustive(); - // Check the contained variants. - let ret = def.variants().iter().try_for_each(|variant| { - check_non_exhaustive_variant(non_exhaustive, variant) - .map_break(|reason| FfiUnsafe { ty, reason, help: None })?; - - match self.check_variant_for_ffi(acc, ty, def, variant, args) { - FfiSafe => ControlFlow::Continue(()), - r => ControlFlow::Break(r), - } - }); - if let ControlFlow::Break(result) = ret { - return result; - } - - FfiSafe - } - } - } - - ty::Char => FfiUnsafe { - ty, - reason: fluent::lint_improper_ctypes_char_reason, - help: Some(fluent::lint_improper_ctypes_char_help), - }, - - // It's just extra invariants on the type that you need to uphold, - // but only the base type is relevant for being representable in FFI. - ty::Pat(base, ..) => self.check_type_for_ffi(acc, base), - - // Primitive types with a stable representation. - ty::Bool | ty::Int(..) | ty::Uint(..) | ty::Float(..) | ty::Never => FfiSafe, - - ty::Slice(_) => FfiUnsafe { - ty, - reason: fluent::lint_improper_ctypes_slice_reason, - help: Some(fluent::lint_improper_ctypes_slice_help), - }, - - ty::Dynamic(..) => { - FfiUnsafe { ty, reason: fluent::lint_improper_ctypes_dyn, help: None } - } - - ty::Str => FfiUnsafe { - ty, - reason: fluent::lint_improper_ctypes_str_reason, - help: Some(fluent::lint_improper_ctypes_str_help), - }, - - ty::Tuple(..) => FfiUnsafe { - ty, - reason: fluent::lint_improper_ctypes_tuple_reason, - help: Some(fluent::lint_improper_ctypes_tuple_help), - }, - - ty::RawPtr(ty, _) | ty::Ref(_, ty, _) - if { - matches!(self.mode, CItemKind::Definition) - && ty.is_sized(self.cx.tcx, self.cx.typing_env()) - } => - { - FfiSafe - } - - ty::RawPtr(ty, _) - if match ty.kind() { - ty::Tuple(tuple) => tuple.is_empty(), - _ => false, - } => - { - FfiSafe - } - - ty::RawPtr(ty, _) | ty::Ref(_, ty, _) => self.check_type_for_ffi(acc, ty), - - ty::Array(inner_ty, _) => self.check_type_for_ffi(acc, inner_ty), - - ty::FnPtr(sig_tys, hdr) => { - let sig = sig_tys.with(hdr); - if sig.abi().is_rustic_abi() { - return FfiUnsafe { - ty, - reason: fluent::lint_improper_ctypes_fnptr_reason, - help: Some(fluent::lint_improper_ctypes_fnptr_help), - }; - } - - let sig = tcx.instantiate_bound_regions_with_erased(sig); - for arg in sig.inputs() { - match self.check_type_for_ffi(acc, *arg) { - FfiSafe => {} - r => return r, - } - } - - let ret_ty = sig.output(); - if ret_ty.is_unit() { - return FfiSafe; - } - - self.check_type_for_ffi(acc, ret_ty) - } - - ty::Foreign(..) => FfiSafe, - - // While opaque types are checked for earlier, if a projection in a struct field - // normalizes to an opaque type, then it will reach this branch. - ty::Alias(ty::Opaque, ..) => { - FfiUnsafe { ty, reason: fluent::lint_improper_ctypes_opaque, help: None } - } - - // `extern "C" fn` functions can have type parameters, which may or may not be FFI-safe, - // so they are currently ignored for the purposes of this lint. - ty::Param(..) | ty::Alias(ty::Projection | ty::Inherent, ..) - if matches!(self.mode, CItemKind::Definition) => - { - FfiSafe - } - - ty::UnsafeBinder(_) => todo!("FIXME(unsafe_binder)"), - - ty::Param(..) - | ty::Alias(ty::Projection | ty::Inherent | ty::Free, ..) - | ty::Infer(..) - | ty::Bound(..) - | ty::Error(_) - | ty::Closure(..) - | ty::CoroutineClosure(..) - | ty::Coroutine(..) - | ty::CoroutineWitness(..) - | ty::Placeholder(..) - | ty::FnDef(..) => bug!("unexpected type in foreign function: {:?}", ty), - } - } - - fn emit_ffi_unsafe_type_lint( - &mut self, - ty: Ty<'tcx>, - sp: Span, - note: DiagMessage, - help: Option<DiagMessage>, - ) { - let lint = match self.mode { - CItemKind::Declaration => IMPROPER_CTYPES, - CItemKind::Definition => IMPROPER_CTYPES_DEFINITIONS, - }; - let desc = match self.mode { - CItemKind::Declaration => "block", - CItemKind::Definition => "fn", - }; - let span_note = if let ty::Adt(def, _) = ty.kind() - && let Some(sp) = self.cx.tcx.hir_span_if_local(def.did()) - { - Some(sp) - } else { - None - }; - self.cx.emit_span_lint( - lint, - sp, - ImproperCTypes { ty, desc, label: sp, help, note, span_note }, - ); - } - - fn check_for_opaque_ty(&mut self, sp: Span, ty: Ty<'tcx>) -> bool { - struct ProhibitOpaqueTypes; - impl<'tcx> ty::TypeVisitor<TyCtxt<'tcx>> for ProhibitOpaqueTypes { - type Result = ControlFlow<Ty<'tcx>>; - - fn visit_ty(&mut self, ty: Ty<'tcx>) -> Self::Result { - if !ty.has_opaque_types() { - return ControlFlow::Continue(()); - } - - if let ty::Alias(ty::Opaque, ..) = ty.kind() { - ControlFlow::Break(ty) - } else { - ty.super_visit_with(self) - } - } - } - - if let Some(ty) = self - .cx - .tcx - .try_normalize_erasing_regions(self.cx.typing_env(), ty) - .unwrap_or(ty) - .visit_with(&mut ProhibitOpaqueTypes) - .break_value() - { - self.emit_ffi_unsafe_type_lint(ty, sp, fluent::lint_improper_ctypes_opaque, None); - true - } else { - false - } - } - - fn check_type_for_ffi_and_report_errors( - &mut self, - sp: Span, - ty: Ty<'tcx>, - is_static: bool, - is_return_type: bool, - ) { - if self.check_for_opaque_ty(sp, ty) { - // We've already emitted an error due to an opaque type. - return; - } - - let ty = self.cx.tcx.try_normalize_erasing_regions(self.cx.typing_env(), ty).unwrap_or(ty); - - // C doesn't really support passing arrays by value - the only way to pass an array by value - // is through a struct. So, first test that the top level isn't an array, and then - // recursively check the types inside. - if !is_static && self.check_for_array_ty(sp, ty) { - return; - } - - // Don't report FFI errors for unit return types. This check exists here, and not in - // the caller (where it would make more sense) so that normalization has definitely - // happened. - if is_return_type && ty.is_unit() { - return; - } - - let mut acc = CTypesVisitorState { cache: FxHashSet::default(), base_ty: ty }; - match self.check_type_for_ffi(&mut acc, ty) { - FfiResult::FfiSafe => {} - FfiResult::FfiPhantom(ty) => { - self.emit_ffi_unsafe_type_lint( - ty, - sp, - fluent::lint_improper_ctypes_only_phantomdata, - None, - ); - } - FfiResult::FfiUnsafe { ty, reason, help } => { - self.emit_ffi_unsafe_type_lint(ty, sp, reason, help); - } - } - } - - /// Check if a function's argument types and result type are "ffi-safe". - /// - /// For a external ABI function, argument types and the result type are walked to find fn-ptr - /// types that have external ABIs, as these still need checked. - fn check_fn(&mut self, def_id: LocalDefId, decl: &'tcx hir::FnDecl<'_>) { - let sig = self.cx.tcx.fn_sig(def_id).instantiate_identity(); - let sig = self.cx.tcx.instantiate_bound_regions_with_erased(sig); - - for (input_ty, input_hir) in iter::zip(sig.inputs(), decl.inputs) { - for (fn_ptr_ty, span) in self.find_fn_ptr_ty_with_external_abi(input_hir, *input_ty) { - self.check_type_for_ffi_and_report_errors(span, fn_ptr_ty, false, false); - } - } - - if let hir::FnRetTy::Return(ret_hir) = decl.output { - for (fn_ptr_ty, span) in self.find_fn_ptr_ty_with_external_abi(ret_hir, sig.output()) { - self.check_type_for_ffi_and_report_errors(span, fn_ptr_ty, false, true); - } - } - } - - /// Check if a function's argument types and result type are "ffi-safe". - fn check_foreign_fn(&mut self, def_id: LocalDefId, decl: &'tcx hir::FnDecl<'_>) { - let sig = self.cx.tcx.fn_sig(def_id).instantiate_identity(); - let sig = self.cx.tcx.instantiate_bound_regions_with_erased(sig); - - for (input_ty, input_hir) in iter::zip(sig.inputs(), decl.inputs) { - self.check_type_for_ffi_and_report_errors(input_hir.span, *input_ty, false, false); - } - - if let hir::FnRetTy::Return(ret_hir) = decl.output { - self.check_type_for_ffi_and_report_errors(ret_hir.span, sig.output(), false, true); - } - } - - fn check_foreign_static(&mut self, id: hir::OwnerId, span: Span) { - let ty = self.cx.tcx.type_of(id).instantiate_identity(); - self.check_type_for_ffi_and_report_errors(span, ty, true, false); - } - - /// Find any fn-ptr types with external ABIs in `ty`. - /// - /// For example, `Option<extern "C" fn()>` returns `extern "C" fn()` - fn find_fn_ptr_ty_with_external_abi( - &self, - hir_ty: &hir::Ty<'tcx>, - ty: Ty<'tcx>, - ) -> Vec<(Ty<'tcx>, Span)> { - struct FnPtrFinder<'tcx> { - spans: Vec<Span>, - tys: Vec<Ty<'tcx>>, - } - - impl<'tcx> hir::intravisit::Visitor<'_> for FnPtrFinder<'tcx> { - fn visit_ty(&mut self, ty: &'_ hir::Ty<'_, AmbigArg>) { - debug!(?ty); - if let hir::TyKind::FnPtr(hir::FnPtrTy { abi, .. }) = ty.kind - && !abi.is_rustic_abi() - { - self.spans.push(ty.span); - } - - hir::intravisit::walk_ty(self, ty) - } - } - - impl<'tcx> ty::TypeVisitor<TyCtxt<'tcx>> for FnPtrFinder<'tcx> { - type Result = (); - - fn visit_ty(&mut self, ty: Ty<'tcx>) -> Self::Result { - if let ty::FnPtr(_, hdr) = ty.kind() - && !hdr.abi.is_rustic_abi() - { - self.tys.push(ty); - } - - ty.super_visit_with(self) - } - } - - let mut visitor = FnPtrFinder { spans: Vec::new(), tys: Vec::new() }; - ty.visit_with(&mut visitor); - visitor.visit_ty_unambig(hir_ty); - - iter::zip(visitor.tys.drain(..), visitor.spans.drain(..)).collect() - } -} - -impl<'tcx> LateLintPass<'tcx> for ImproperCTypesDeclarations { - fn check_foreign_item(&mut self, cx: &LateContext<'tcx>, it: &hir::ForeignItem<'tcx>) { - let mut vis = ImproperCTypesVisitor { cx, mode: CItemKind::Declaration }; - let abi = cx.tcx.hir_get_foreign_abi(it.hir_id()); - - match it.kind { - hir::ForeignItemKind::Fn(sig, _, _) => { - if abi.is_rustic_abi() { - vis.check_fn(it.owner_id.def_id, sig.decl) - } else { - vis.check_foreign_fn(it.owner_id.def_id, sig.decl); - } - } - hir::ForeignItemKind::Static(ty, _, _) if !abi.is_rustic_abi() => { - vis.check_foreign_static(it.owner_id, ty.span); - } - hir::ForeignItemKind::Static(..) | hir::ForeignItemKind::Type => (), - } - } -} - -impl ImproperCTypesDefinitions { - fn check_ty_maybe_containing_foreign_fnptr<'tcx>( - &mut self, - cx: &LateContext<'tcx>, - hir_ty: &'tcx hir::Ty<'_>, - ty: Ty<'tcx>, - ) { - let mut vis = ImproperCTypesVisitor { cx, mode: CItemKind::Definition }; - for (fn_ptr_ty, span) in vis.find_fn_ptr_ty_with_external_abi(hir_ty, ty) { - vis.check_type_for_ffi_and_report_errors(span, fn_ptr_ty, true, false); - } - } - - fn check_arg_for_power_alignment<'tcx>( - &mut self, - cx: &LateContext<'tcx>, - ty: Ty<'tcx>, - ) -> bool { - assert!(cx.tcx.sess.target.os == "aix"); - // Structs (under repr(C)) follow the power alignment rule if: - // - the first field of the struct is a floating-point type that - // is greater than 4-bytes, or - // - the first field of the struct is an aggregate whose - // recursively first field is a floating-point type greater than - // 4 bytes. - if ty.is_floating_point() && ty.primitive_size(cx.tcx).bytes() > 4 { - return true; - } else if let Adt(adt_def, _) = ty.kind() - && adt_def.is_struct() - && adt_def.repr().c() - && !adt_def.repr().packed() - && adt_def.repr().align.is_none() - { - let struct_variant = adt_def.variant(VariantIdx::ZERO); - // Within a nested struct, all fields are examined to correctly - // report if any fields after the nested struct within the - // original struct are misaligned. - for struct_field in &struct_variant.fields { - let field_ty = cx.tcx.type_of(struct_field.did).instantiate_identity(); - if self.check_arg_for_power_alignment(cx, field_ty) { - return true; - } - } - } - return false; - } - - fn check_struct_for_power_alignment<'tcx>( - &mut self, - cx: &LateContext<'tcx>, - item: &'tcx hir::Item<'tcx>, - ) { - let adt_def = cx.tcx.adt_def(item.owner_id.to_def_id()); - // repr(C) structs also with packed or aligned representation - // should be ignored. - if adt_def.repr().c() - && !adt_def.repr().packed() - && adt_def.repr().align.is_none() - && cx.tcx.sess.target.os == "aix" - && !adt_def.all_fields().next().is_none() - { - let struct_variant_data = item.expect_struct().2; - for field_def in struct_variant_data.fields().iter().skip(1) { - // Struct fields (after the first field) are checked for the - // power alignment rule, as fields after the first are likely - // to be the fields that are misaligned. - let def_id = field_def.def_id; - let ty = cx.tcx.type_of(def_id).instantiate_identity(); - if self.check_arg_for_power_alignment(cx, ty) { - cx.emit_span_lint(USES_POWER_ALIGNMENT, field_def.span, UsesPowerAlignment); - } - } - } - } -} - -/// `ImproperCTypesDefinitions` checks items outside of foreign items (e.g. stuff that isn't in -/// `extern "C" { }` blocks): -/// -/// - `extern "<abi>" fn` definitions are checked in the same way as the -/// `ImproperCtypesDeclarations` visitor checks functions if `<abi>` is external (e.g. "C"). -/// - All other items which contain types (e.g. other functions, struct definitions, etc) are -/// checked for extern fn-ptrs with external ABIs. -impl<'tcx> LateLintPass<'tcx> for ImproperCTypesDefinitions { - fn check_item(&mut self, cx: &LateContext<'tcx>, item: &'tcx hir::Item<'tcx>) { - match item.kind { - hir::ItemKind::Static(_, _, ty, _) - | hir::ItemKind::Const(_, _, ty, _) - | hir::ItemKind::TyAlias(_, _, ty) => { - self.check_ty_maybe_containing_foreign_fnptr( - cx, - ty, - cx.tcx.type_of(item.owner_id).instantiate_identity(), - ); - } - // See `check_fn`.. - hir::ItemKind::Fn { .. } => {} - // Structs are checked based on if they follow the power alignment - // rule (under repr(C)). - hir::ItemKind::Struct(..) => { - self.check_struct_for_power_alignment(cx, item); - } - // See `check_field_def`.. - hir::ItemKind::Union(..) | hir::ItemKind::Enum(..) => {} - // Doesn't define something that can contain a external type to be checked. - hir::ItemKind::Impl(..) - | hir::ItemKind::TraitAlias(..) - | hir::ItemKind::Trait(..) - | hir::ItemKind::GlobalAsm { .. } - | hir::ItemKind::ForeignMod { .. } - | hir::ItemKind::Mod(..) - | hir::ItemKind::Macro(..) - | hir::ItemKind::Use(..) - | hir::ItemKind::ExternCrate(..) => {} - } - } - - fn check_field_def(&mut self, cx: &LateContext<'tcx>, field: &'tcx hir::FieldDef<'tcx>) { - self.check_ty_maybe_containing_foreign_fnptr( - cx, - field.ty, - cx.tcx.type_of(field.def_id).instantiate_identity(), - ); - } - - fn check_fn( - &mut self, - cx: &LateContext<'tcx>, - kind: hir::intravisit::FnKind<'tcx>, - decl: &'tcx hir::FnDecl<'_>, - _: &'tcx hir::Body<'_>, - _: Span, - id: LocalDefId, - ) { - use hir::intravisit::FnKind; - - let abi = match kind { - FnKind::ItemFn(_, _, header, ..) => header.abi, - FnKind::Method(_, sig, ..) => sig.header.abi, - _ => return, - }; - - let mut vis = ImproperCTypesVisitor { cx, mode: CItemKind::Definition }; - if abi.is_rustic_abi() { - vis.check_fn(id, decl); - } else { - vis.check_foreign_fn(id, decl); - } - } -} - declare_lint_pass!(VariantSizeDifferences => [VARIANT_SIZE_DIFFERENCES]); impl<'tcx> LateLintPass<'tcx> for VariantSizeDifferences { diff --git a/compiler/rustc_lint/src/types/improper_ctypes.rs b/compiler/rustc_lint/src/types/improper_ctypes.rs index 13afa540afc..7ca57b0094e 100644 --- a/compiler/rustc_lint/src/types/improper_ctypes.rs +++ b/compiler/rustc_lint/src/types/improper_ctypes.rs @@ -1,10 +1,141 @@ +use std::iter; use std::ops::ControlFlow; +use bitflags::bitflags; +use rustc_abi::VariantIdx; +use rustc_data_structures::fx::FxHashSet; use rustc_errors::DiagMessage; use rustc_hir::def::CtorKind; -use rustc_middle::ty; +use rustc_hir::intravisit::VisitorExt; +use rustc_hir::{self as hir, AmbigArg}; +use rustc_middle::bug; +use rustc_middle::ty::{ + self, Adt, AdtDef, AdtKind, GenericArgsRef, Ty, TyCtxt, TypeSuperVisitable, TypeVisitable, + TypeVisitableExt, +}; +use rustc_session::{declare_lint, declare_lint_pass}; +use rustc_span::def_id::LocalDefId; +use rustc_span::{Span, sym}; +use tracing::debug; -use crate::fluent_generated as fluent; +use super::repr_nullable_ptr; +use crate::lints::{ImproperCTypes, UsesPowerAlignment}; +use crate::{LateContext, LateLintPass, LintContext, fluent_generated as fluent}; + +declare_lint! { + /// The `improper_ctypes` lint detects incorrect use of types in foreign + /// modules. + /// + /// ### Example + /// + /// ```rust + /// unsafe extern "C" { + /// static STATIC: String; + /// } + /// ``` + /// + /// {{produces}} + /// + /// ### Explanation + /// + /// The compiler has several checks to verify that types used in `extern` + /// blocks are safe and follow certain rules to ensure proper + /// compatibility with the foreign interfaces. This lint is issued when it + /// detects a probable mistake in a definition. The lint usually should + /// provide a description of the issue, along with possibly a hint on how + /// to resolve it. + IMPROPER_CTYPES, + Warn, + "proper use of libc types in foreign modules" +} + +declare_lint! { + /// The `improper_ctypes_definitions` lint detects incorrect use of + /// [`extern` function] definitions. + /// + /// [`extern` function]: https://doc.rust-lang.org/reference/items/functions.html#extern-function-qualifier + /// + /// ### Example + /// + /// ```rust + /// # #![allow(unused)] + /// pub extern "C" fn str_type(p: &str) { } + /// ``` + /// + /// {{produces}} + /// + /// ### Explanation + /// + /// There are many parameter and return types that may be specified in an + /// `extern` function that are not compatible with the given ABI. This + /// lint is an alert that these types should not be used. The lint usually + /// should provide a description of the issue, along with possibly a hint + /// on how to resolve it. + IMPROPER_CTYPES_DEFINITIONS, + Warn, + "proper use of libc types in foreign item definitions" +} + +declare_lint! { + /// The `uses_power_alignment` lint detects specific `repr(C)` + /// aggregates on AIX. + /// In its platform C ABI, AIX uses the "power" (as in PowerPC) alignment + /// rule (detailed in https://www.ibm.com/docs/en/xl-c-and-cpp-aix/16.1?topic=data-using-alignment-modes#alignment), + /// which can also be set for XLC by `#pragma align(power)` or + /// `-qalign=power`. Aggregates with a floating-point type as the + /// recursively first field (as in "at offset 0") modify the layout of + /// *subsequent* fields of the associated structs to use an alignment value + /// where the floating-point type is aligned on a 4-byte boundary. + /// + /// Effectively, subsequent floating-point fields act as-if they are `repr(packed(4))`. This + /// would be unsound to do in a `repr(C)` type without all the restrictions that come with + /// `repr(packed)`. Rust instead chooses a layout that maintains soundness of Rust code, at the + /// expense of incompatibility with C code. + /// + /// ### Example + /// + /// ```rust,ignore (fails on non-powerpc64-ibm-aix) + /// #[repr(C)] + /// pub struct Floats { + /// a: f64, + /// b: u8, + /// c: f64, + /// } + /// ``` + /// + /// This will produce: + /// + /// ```text + /// warning: repr(C) does not follow the power alignment rule. This may affect platform C ABI compatibility for this type + /// --> <source>:5:3 + /// | + /// 5 | c: f64, + /// | ^^^^^^ + /// | + /// = note: `#[warn(uses_power_alignment)]` on by default + /// ``` + /// + /// ### Explanation + /// + /// The power alignment rule specifies that the above struct has the + /// following alignment: + /// - offset_of!(Floats, a) == 0 + /// - offset_of!(Floats, b) == 8 + /// - offset_of!(Floats, c) == 12 + /// + /// However, Rust currently aligns `c` at `offset_of!(Floats, c) == 16`. + /// Using offset 12 would be unsound since `f64` generally must be 8-aligned on this target. + /// Thus, a warning is produced for the above struct. + USES_POWER_ALIGNMENT, + Warn, + "Structs do not follow the power alignment rule under repr(C)" +} + +declare_lint_pass!(ImproperCTypesLint => [ + IMPROPER_CTYPES, + IMPROPER_CTYPES_DEFINITIONS, + USES_POWER_ALIGNMENT +]); /// Check a variant of a non-exhaustive enum for improper ctypes /// @@ -41,3 +172,884 @@ fn variant_has_complex_ctor(variant: &ty::VariantDef) -> bool { // CtorKind::Const means a "unit" ctor !matches!(variant.ctor_kind(), Some(CtorKind::Const)) } + +/// Per-struct-field function that checks if a struct definition follows +/// the Power alignment Rule (see the `check_struct_for_power_alignment` function). +fn check_arg_for_power_alignment<'tcx>(cx: &LateContext<'tcx>, ty: Ty<'tcx>) -> bool { + let tcx = cx.tcx; + assert!(tcx.sess.target.os == "aix"); + // Structs (under repr(C)) follow the power alignment rule if: + // - the first field of the struct is a floating-point type that + // is greater than 4-bytes, or + // - the first field of the struct is an aggregate whose + // recursively first field is a floating-point type greater than + // 4 bytes. + if ty.is_floating_point() && ty.primitive_size(tcx).bytes() > 4 { + return true; + } else if let Adt(adt_def, _) = ty.kind() + && adt_def.is_struct() + && adt_def.repr().c() + && !adt_def.repr().packed() + && adt_def.repr().align.is_none() + { + let struct_variant = adt_def.variant(VariantIdx::ZERO); + // Within a nested struct, all fields are examined to correctly + // report if any fields after the nested struct within the + // original struct are misaligned. + for struct_field in &struct_variant.fields { + let field_ty = tcx.type_of(struct_field.did).instantiate_identity(); + if check_arg_for_power_alignment(cx, field_ty) { + return true; + } + } + } + return false; +} + +/// Check a struct definition for respect of the Power alignment Rule (as in PowerPC), +/// which should be respected in the "aix" target OS. +/// To do so, we must follow one of the two following conditions: +/// - The first field of the struct must be floating-point type that +/// is greater than 4-bytes. +/// - The first field of the struct must be an aggregate whose +/// recursively first field is a floating-point type greater than +/// 4 bytes. +fn check_struct_for_power_alignment<'tcx>( + cx: &LateContext<'tcx>, + item: &'tcx hir::Item<'tcx>, + adt_def: AdtDef<'tcx>, +) { + let tcx = cx.tcx; + // repr(C) structs also with packed or aligned representation + // should be ignored. + if adt_def.repr().c() + && !adt_def.repr().packed() + && adt_def.repr().align.is_none() + && tcx.sess.target.os == "aix" + && !adt_def.all_fields().next().is_none() + { + let struct_variant_data = item.expect_struct().2; + for field_def in struct_variant_data.fields().iter().skip(1) { + // Struct fields (after the first field) are checked for the + // power alignment rule, as fields after the first are likely + // to be the fields that are misaligned. + let def_id = field_def.def_id; + let ty = tcx.type_of(def_id).instantiate_identity(); + if check_arg_for_power_alignment(cx, ty) { + cx.emit_span_lint(USES_POWER_ALIGNMENT, field_def.span, UsesPowerAlignment); + } + } + } +} + +#[derive(Clone, Copy)] +enum CItemKind { + Declaration, + Definition, +} + +enum FfiResult<'tcx> { + FfiSafe, + FfiPhantom(Ty<'tcx>), + FfiUnsafe { ty: Ty<'tcx>, reason: DiagMessage, help: Option<DiagMessage> }, +} + +/// The result when a type has been checked but perhaps not completely. `None` indicates that +/// FFI safety/unsafety has not yet been determined, `Some(res)` indicates that the safety/unsafety +/// in the `FfiResult` is final. +type PartialFfiResult<'tcx> = Option<FfiResult<'tcx>>; + +bitflags! { + #[derive(Clone, Copy, Debug, PartialEq, Eq)] + struct VisitorState: u8 { + /// For use in (externally-linked) static variables. + const STATIC = 0b000001; + /// For use in functions in general. + const FUNC = 0b000010; + /// For variables in function returns (implicitly: not for static variables). + const FN_RETURN = 0b000100; + /// For variables in functions/variables which are defined in rust. + const DEFINED = 0b001000; + /// For times where we are only defining the type of something + /// (struct/enum/union definitions, FnPtrs). + const THEORETICAL = 0b010000; + } +} + +impl VisitorState { + // The values that can be set. + const STATIC_TY: Self = Self::STATIC; + const ARGUMENT_TY_IN_DEFINITION: Self = + Self::from_bits(Self::FUNC.bits() | Self::DEFINED.bits()).unwrap(); + const RETURN_TY_IN_DEFINITION: Self = + Self::from_bits(Self::FUNC.bits() | Self::FN_RETURN.bits() | Self::DEFINED.bits()).unwrap(); + const ARGUMENT_TY_IN_DECLARATION: Self = Self::FUNC; + const RETURN_TY_IN_DECLARATION: Self = + Self::from_bits(Self::FUNC.bits() | Self::FN_RETURN.bits()).unwrap(); + const ARGUMENT_TY_IN_FNPTR: Self = + Self::from_bits(Self::FUNC.bits() | Self::THEORETICAL.bits()).unwrap(); + const RETURN_TY_IN_FNPTR: Self = + Self::from_bits(Self::FUNC.bits() | Self::THEORETICAL.bits() | Self::FN_RETURN.bits()) + .unwrap(); + + /// Get the proper visitor state for a given function's arguments. + fn argument_from_fnmode(fn_mode: CItemKind) -> Self { + match fn_mode { + CItemKind::Definition => VisitorState::ARGUMENT_TY_IN_DEFINITION, + CItemKind::Declaration => VisitorState::ARGUMENT_TY_IN_DECLARATION, + } + } + + /// Get the proper visitor state for a given function's return type. + fn return_from_fnmode(fn_mode: CItemKind) -> Self { + match fn_mode { + CItemKind::Definition => VisitorState::RETURN_TY_IN_DEFINITION, + CItemKind::Declaration => VisitorState::RETURN_TY_IN_DECLARATION, + } + } + + /// Whether the type is used in a function. + fn is_in_function(self) -> bool { + let ret = self.contains(Self::FUNC); + if ret { + debug_assert!(!self.contains(Self::STATIC)); + } + ret + } + /// Whether the type is used (directly or not) in a function, in return position. + fn is_in_function_return(self) -> bool { + let ret = self.contains(Self::FN_RETURN); + if ret { + debug_assert!(self.is_in_function()); + } + ret + } + /// Whether the type is used (directly or not) in a defined function. + /// In other words, whether or not we allow non-FFI-safe types behind a C pointer, + /// to be treated as an opaque type on the other side of the FFI boundary. + fn is_in_defined_function(self) -> bool { + self.contains(Self::DEFINED) && self.is_in_function() + } + + /// Whether the type is used (directly or not) in a function pointer type. + /// Here, we also allow non-FFI-safe types behind a C pointer, + /// to be treated as an opaque type on the other side of the FFI boundary. + fn is_in_fnptr(self) -> bool { + self.contains(Self::THEORETICAL) && self.is_in_function() + } + + /// Whether we can expect type parameters and co in a given type. + fn can_expect_ty_params(self) -> bool { + // rust-defined functions, as well as FnPtrs + self.contains(Self::THEORETICAL) || self.is_in_defined_function() + } +} + +/// Visitor used to recursively traverse MIR types and evaluate FFI-safety. +/// It uses ``check_*`` methods as entrypoints to be called elsewhere, +/// and ``visit_*`` methods to recurse. +struct ImproperCTypesVisitor<'a, 'tcx> { + cx: &'a LateContext<'tcx>, + /// To prevent problems with recursive types, + /// add a types-in-check cache. + cache: FxHashSet<Ty<'tcx>>, + /// The original type being checked, before we recursed + /// to any other types it contains. + base_ty: Ty<'tcx>, + base_fn_mode: CItemKind, +} + +impl<'a, 'tcx> ImproperCTypesVisitor<'a, 'tcx> { + fn new(cx: &'a LateContext<'tcx>, base_ty: Ty<'tcx>, base_fn_mode: CItemKind) -> Self { + Self { cx, base_ty, base_fn_mode, cache: FxHashSet::default() } + } + + /// Checks if the given field's type is "ffi-safe". + fn check_field_type_for_ffi( + &mut self, + state: VisitorState, + field: &ty::FieldDef, + args: GenericArgsRef<'tcx>, + ) -> FfiResult<'tcx> { + let field_ty = field.ty(self.cx.tcx, args); + let field_ty = self + .cx + .tcx + .try_normalize_erasing_regions(self.cx.typing_env(), field_ty) + .unwrap_or(field_ty); + self.visit_type(state, field_ty) + } + + /// Checks if the given `VariantDef`'s field types are "ffi-safe". + fn check_variant_for_ffi( + &mut self, + state: VisitorState, + ty: Ty<'tcx>, + def: ty::AdtDef<'tcx>, + variant: &ty::VariantDef, + args: GenericArgsRef<'tcx>, + ) -> FfiResult<'tcx> { + use FfiResult::*; + let transparent_with_all_zst_fields = if def.repr().transparent() { + if let Some(field) = super::transparent_newtype_field(self.cx.tcx, variant) { + // Transparent newtypes have at most one non-ZST field which needs to be checked.. + match self.check_field_type_for_ffi(state, field, args) { + FfiUnsafe { ty, .. } if ty.is_unit() => (), + r => return r, + } + + false + } else { + // ..or have only ZST fields, which is FFI-unsafe (unless those fields are all + // `PhantomData`). + true + } + } else { + false + }; + + // We can't completely trust `repr(C)` markings, so make sure the fields are actually safe. + let mut all_phantom = !variant.fields.is_empty(); + for field in &variant.fields { + all_phantom &= match self.check_field_type_for_ffi(state, field, args) { + FfiSafe => false, + // `()` fields are FFI-safe! + FfiUnsafe { ty, .. } if ty.is_unit() => false, + FfiPhantom(..) => true, + r @ FfiUnsafe { .. } => return r, + } + } + + if all_phantom { + FfiPhantom(ty) + } else if transparent_with_all_zst_fields { + FfiUnsafe { ty, reason: fluent::lint_improper_ctypes_struct_zst, help: None } + } else { + FfiSafe + } + } + + /// Checks if the given type is "ffi-safe" (has a stable, well-defined + /// representation which can be exported to C code). + fn visit_type(&mut self, state: VisitorState, ty: Ty<'tcx>) -> FfiResult<'tcx> { + use FfiResult::*; + + let tcx = self.cx.tcx; + + // Protect against infinite recursion, for example + // `struct S(*mut S);`. + // FIXME: A recursion limit is necessary as well, for irregular + // recursive types. + if !self.cache.insert(ty) { + return FfiSafe; + } + + match *ty.kind() { + ty::Adt(def, args) => { + if let Some(boxed) = ty.boxed_ty() + && ( + // FIXME(ctypes): this logic is broken, but it still fits the current tests + state.is_in_defined_function() + || (state.is_in_fnptr() + && matches!(self.base_fn_mode, CItemKind::Definition)) + ) + { + if boxed.is_sized(tcx, self.cx.typing_env()) { + return FfiSafe; + } else { + return FfiUnsafe { + ty, + reason: fluent::lint_improper_ctypes_box, + help: None, + }; + } + } + if def.is_phantom_data() { + return FfiPhantom(ty); + } + match def.adt_kind() { + AdtKind::Struct | AdtKind::Union => { + if let Some(sym::cstring_type | sym::cstr_type) = + tcx.get_diagnostic_name(def.did()) + && !self.base_ty.is_mutable_ptr() + { + return FfiUnsafe { + ty, + reason: fluent::lint_improper_ctypes_cstr_reason, + help: Some(fluent::lint_improper_ctypes_cstr_help), + }; + } + + if !def.repr().c() && !def.repr().transparent() { + return FfiUnsafe { + ty, + reason: if def.is_struct() { + fluent::lint_improper_ctypes_struct_layout_reason + } else { + fluent::lint_improper_ctypes_union_layout_reason + }, + help: if def.is_struct() { + Some(fluent::lint_improper_ctypes_struct_layout_help) + } else { + Some(fluent::lint_improper_ctypes_union_layout_help) + }, + }; + } + + if def.non_enum_variant().field_list_has_applicable_non_exhaustive() { + return FfiUnsafe { + ty, + reason: if def.is_struct() { + fluent::lint_improper_ctypes_struct_non_exhaustive + } else { + fluent::lint_improper_ctypes_union_non_exhaustive + }, + help: None, + }; + } + + if def.non_enum_variant().fields.is_empty() { + return FfiUnsafe { + ty, + reason: if def.is_struct() { + fluent::lint_improper_ctypes_struct_fieldless_reason + } else { + fluent::lint_improper_ctypes_union_fieldless_reason + }, + help: if def.is_struct() { + Some(fluent::lint_improper_ctypes_struct_fieldless_help) + } else { + Some(fluent::lint_improper_ctypes_union_fieldless_help) + }, + }; + } + + self.check_variant_for_ffi(state, ty, def, def.non_enum_variant(), args) + } + AdtKind::Enum => { + if def.variants().is_empty() { + // Empty enums are okay... although sort of useless. + return FfiSafe; + } + // Check for a repr() attribute to specify the size of the + // discriminant. + if !def.repr().c() && !def.repr().transparent() && def.repr().int.is_none() + { + // Special-case types like `Option<extern fn()>` and `Result<extern fn(), ()>` + if let Some(ty) = + repr_nullable_ptr(self.cx.tcx, self.cx.typing_env(), ty) + { + return self.visit_type(state, ty); + } + + return FfiUnsafe { + ty, + reason: fluent::lint_improper_ctypes_enum_repr_reason, + help: Some(fluent::lint_improper_ctypes_enum_repr_help), + }; + } + + let non_exhaustive = def.variant_list_has_applicable_non_exhaustive(); + // Check the contained variants. + let ret = def.variants().iter().try_for_each(|variant| { + check_non_exhaustive_variant(non_exhaustive, variant) + .map_break(|reason| FfiUnsafe { ty, reason, help: None })?; + + match self.check_variant_for_ffi(state, ty, def, variant, args) { + FfiSafe => ControlFlow::Continue(()), + r => ControlFlow::Break(r), + } + }); + if let ControlFlow::Break(result) = ret { + return result; + } + + FfiSafe + } + } + } + + ty::Char => FfiUnsafe { + ty, + reason: fluent::lint_improper_ctypes_char_reason, + help: Some(fluent::lint_improper_ctypes_char_help), + }, + + // It's just extra invariants on the type that you need to uphold, + // but only the base type is relevant for being representable in FFI. + ty::Pat(base, ..) => self.visit_type(state, base), + + // Primitive types with a stable representation. + ty::Bool | ty::Int(..) | ty::Uint(..) | ty::Float(..) | ty::Never => FfiSafe, + + ty::Slice(_) => FfiUnsafe { + ty, + reason: fluent::lint_improper_ctypes_slice_reason, + help: Some(fluent::lint_improper_ctypes_slice_help), + }, + + ty::Dynamic(..) => { + FfiUnsafe { ty, reason: fluent::lint_improper_ctypes_dyn, help: None } + } + + ty::Str => FfiUnsafe { + ty, + reason: fluent::lint_improper_ctypes_str_reason, + help: Some(fluent::lint_improper_ctypes_str_help), + }, + + ty::Tuple(..) => FfiUnsafe { + ty, + reason: fluent::lint_improper_ctypes_tuple_reason, + help: Some(fluent::lint_improper_ctypes_tuple_help), + }, + + ty::RawPtr(ty, _) | ty::Ref(_, ty, _) + if { + (state.is_in_defined_function() || state.is_in_fnptr()) + && ty.is_sized(self.cx.tcx, self.cx.typing_env()) + } => + { + FfiSafe + } + + ty::RawPtr(ty, _) + if match ty.kind() { + ty::Tuple(tuple) => tuple.is_empty(), + _ => false, + } => + { + FfiSafe + } + + ty::RawPtr(ty, _) | ty::Ref(_, ty, _) => self.visit_type(state, ty), + + ty::Array(inner_ty, _) => self.visit_type(state, inner_ty), + + ty::FnPtr(sig_tys, hdr) => { + let sig = sig_tys.with(hdr); + if sig.abi().is_rustic_abi() { + return FfiUnsafe { + ty, + reason: fluent::lint_improper_ctypes_fnptr_reason, + help: Some(fluent::lint_improper_ctypes_fnptr_help), + }; + } + + let sig = tcx.instantiate_bound_regions_with_erased(sig); + for arg in sig.inputs() { + match self.visit_type(VisitorState::ARGUMENT_TY_IN_FNPTR, *arg) { + FfiSafe => {} + r => return r, + } + } + + let ret_ty = sig.output(); + if ret_ty.is_unit() { + return FfiSafe; + } + + self.visit_type(VisitorState::RETURN_TY_IN_FNPTR, ret_ty) + } + + ty::Foreign(..) => FfiSafe, + + // While opaque types are checked for earlier, if a projection in a struct field + // normalizes to an opaque type, then it will reach this branch. + ty::Alias(ty::Opaque, ..) => { + FfiUnsafe { ty, reason: fluent::lint_improper_ctypes_opaque, help: None } + } + + // `extern "C" fn` functions can have type parameters, which may or may not be FFI-safe, + // so they are currently ignored for the purposes of this lint. + ty::Param(..) | ty::Alias(ty::Projection | ty::Inherent, ..) + if state.can_expect_ty_params() => + { + FfiSafe + } + + ty::UnsafeBinder(_) => todo!("FIXME(unsafe_binder)"), + + ty::Param(..) + | ty::Alias(ty::Projection | ty::Inherent | ty::Free, ..) + | ty::Infer(..) + | ty::Bound(..) + | ty::Error(_) + | ty::Closure(..) + | ty::CoroutineClosure(..) + | ty::Coroutine(..) + | ty::CoroutineWitness(..) + | ty::Placeholder(..) + | ty::FnDef(..) => bug!("unexpected type in foreign function: {:?}", ty), + } + } + + fn visit_for_opaque_ty(&mut self, ty: Ty<'tcx>) -> PartialFfiResult<'tcx> { + struct ProhibitOpaqueTypes; + impl<'tcx> ty::TypeVisitor<TyCtxt<'tcx>> for ProhibitOpaqueTypes { + type Result = ControlFlow<Ty<'tcx>>; + + fn visit_ty(&mut self, ty: Ty<'tcx>) -> Self::Result { + if !ty.has_opaque_types() { + return ControlFlow::Continue(()); + } + + if let ty::Alias(ty::Opaque, ..) = ty.kind() { + ControlFlow::Break(ty) + } else { + ty.super_visit_with(self) + } + } + } + + if let Some(ty) = self + .cx + .tcx + .try_normalize_erasing_regions(self.cx.typing_env(), ty) + .unwrap_or(ty) + .visit_with(&mut ProhibitOpaqueTypes) + .break_value() + { + Some(FfiResult::FfiUnsafe { + ty, + reason: fluent::lint_improper_ctypes_opaque, + help: None, + }) + } else { + None + } + } + + /// Check if the type is array and emit an unsafe type lint. + fn check_for_array_ty(&mut self, ty: Ty<'tcx>) -> PartialFfiResult<'tcx> { + if let ty::Array(..) = ty.kind() { + Some(FfiResult::FfiUnsafe { + ty, + reason: fluent::lint_improper_ctypes_array_reason, + help: Some(fluent::lint_improper_ctypes_array_help), + }) + } else { + None + } + } + + /// Determine the FFI-safety of a single (MIR) type, given the context of how it is used. + fn check_type(&mut self, state: VisitorState, ty: Ty<'tcx>) -> FfiResult<'tcx> { + if let Some(res) = self.visit_for_opaque_ty(ty) { + return res; + } + + let ty = self.cx.tcx.try_normalize_erasing_regions(self.cx.typing_env(), ty).unwrap_or(ty); + + // C doesn't really support passing arrays by value - the only way to pass an array by value + // is through a struct. So, first test that the top level isn't an array, and then + // recursively check the types inside. + if state.is_in_function() { + if let Some(res) = self.check_for_array_ty(ty) { + return res; + } + } + + // Don't report FFI errors for unit return types. This check exists here, and not in + // the caller (where it would make more sense) so that normalization has definitely + // happened. + if state.is_in_function_return() && ty.is_unit() { + return FfiResult::FfiSafe; + } + + self.visit_type(state, ty) + } +} + +impl<'tcx> ImproperCTypesLint { + /// Find any fn-ptr types with external ABIs in `ty`, and FFI-checks them. + /// For example, `Option<extern "C" fn()>` FFI-checks `extern "C" fn()`. + fn check_type_for_external_abi_fnptr( + &mut self, + cx: &LateContext<'tcx>, + state: VisitorState, + hir_ty: &hir::Ty<'tcx>, + ty: Ty<'tcx>, + fn_mode: CItemKind, + ) { + struct FnPtrFinder<'tcx> { + spans: Vec<Span>, + tys: Vec<Ty<'tcx>>, + } + + impl<'tcx> hir::intravisit::Visitor<'_> for FnPtrFinder<'tcx> { + fn visit_ty(&mut self, ty: &'_ hir::Ty<'_, AmbigArg>) { + debug!(?ty); + if let hir::TyKind::FnPtr(hir::FnPtrTy { abi, .. }) = ty.kind + && !abi.is_rustic_abi() + { + self.spans.push(ty.span); + } + + hir::intravisit::walk_ty(self, ty) + } + } + + impl<'tcx> ty::TypeVisitor<TyCtxt<'tcx>> for FnPtrFinder<'tcx> { + type Result = (); + + fn visit_ty(&mut self, ty: Ty<'tcx>) -> Self::Result { + if let ty::FnPtr(_, hdr) = ty.kind() + && !hdr.abi.is_rustic_abi() + { + self.tys.push(ty); + } + + ty.super_visit_with(self) + } + } + + let mut visitor = FnPtrFinder { spans: Vec::new(), tys: Vec::new() }; + ty.visit_with(&mut visitor); + visitor.visit_ty_unambig(hir_ty); + + let all_types = iter::zip(visitor.tys.drain(..), visitor.spans.drain(..)); + for (fn_ptr_ty, span) in all_types { + let mut visitor = ImproperCTypesVisitor::new(cx, fn_ptr_ty, fn_mode); + // FIXME(ctypes): make a check_for_fnptr + let ffi_res = visitor.check_type(state, fn_ptr_ty); + + self.process_ffi_result(cx, span, ffi_res, fn_mode); + } + } + + /// Regardless of a function's need to be "ffi-safe", look for fn-ptr argument/return types + /// that need to be checked for ffi-safety. + fn check_fn_for_external_abi_fnptr( + &mut self, + cx: &LateContext<'tcx>, + fn_mode: CItemKind, + def_id: LocalDefId, + decl: &'tcx hir::FnDecl<'_>, + ) { + let sig = cx.tcx.fn_sig(def_id).instantiate_identity(); + let sig = cx.tcx.instantiate_bound_regions_with_erased(sig); + + for (input_ty, input_hir) in iter::zip(sig.inputs(), decl.inputs) { + let state = VisitorState::argument_from_fnmode(fn_mode); + self.check_type_for_external_abi_fnptr(cx, state, input_hir, *input_ty, fn_mode); + } + + if let hir::FnRetTy::Return(ret_hir) = decl.output { + let state = VisitorState::return_from_fnmode(fn_mode); + self.check_type_for_external_abi_fnptr(cx, state, ret_hir, sig.output(), fn_mode); + } + } + + /// For a local definition of a #[repr(C)] struct/enum/union, check that it is indeed FFI-safe. + fn check_reprc_adt( + &mut self, + cx: &LateContext<'tcx>, + item: &'tcx hir::Item<'tcx>, + adt_def: AdtDef<'tcx>, + ) { + debug_assert!( + adt_def.repr().c() && !adt_def.repr().packed() && adt_def.repr().align.is_none() + ); + + // FIXME(ctypes): this following call is awkward. + // is there a way to perform its logic in MIR space rather than HIR space? + // (so that its logic can be absorbed into visitor.visit_struct_or_union) + check_struct_for_power_alignment(cx, item, adt_def); + } + + fn check_foreign_static(&mut self, cx: &LateContext<'tcx>, id: hir::OwnerId, span: Span) { + let ty = cx.tcx.type_of(id).instantiate_identity(); + let mut visitor = ImproperCTypesVisitor::new(cx, ty, CItemKind::Declaration); + let ffi_res = visitor.check_type(VisitorState::STATIC_TY, ty); + self.process_ffi_result(cx, span, ffi_res, CItemKind::Declaration); + } + + /// Check if a function's argument types and result type are "ffi-safe". + fn check_foreign_fn( + &mut self, + cx: &LateContext<'tcx>, + fn_mode: CItemKind, + def_id: LocalDefId, + decl: &'tcx hir::FnDecl<'_>, + ) { + let sig = cx.tcx.fn_sig(def_id).instantiate_identity(); + let sig = cx.tcx.instantiate_bound_regions_with_erased(sig); + + for (input_ty, input_hir) in iter::zip(sig.inputs(), decl.inputs) { + let state = VisitorState::argument_from_fnmode(fn_mode); + let mut visitor = ImproperCTypesVisitor::new(cx, *input_ty, fn_mode); + let ffi_res = visitor.check_type(state, *input_ty); + self.process_ffi_result(cx, input_hir.span, ffi_res, fn_mode); + } + + if let hir::FnRetTy::Return(ret_hir) = decl.output { + let state = VisitorState::return_from_fnmode(fn_mode); + let mut visitor = ImproperCTypesVisitor::new(cx, sig.output(), fn_mode); + let ffi_res = visitor.check_type(state, sig.output()); + self.process_ffi_result(cx, ret_hir.span, ffi_res, fn_mode); + } + } + + fn process_ffi_result( + &self, + cx: &LateContext<'tcx>, + sp: Span, + res: FfiResult<'tcx>, + fn_mode: CItemKind, + ) { + match res { + FfiResult::FfiSafe => {} + FfiResult::FfiPhantom(ty) => { + self.emit_ffi_unsafe_type_lint( + cx, + ty, + sp, + fluent::lint_improper_ctypes_only_phantomdata, + None, + fn_mode, + ); + } + FfiResult::FfiUnsafe { ty, reason, help } => { + self.emit_ffi_unsafe_type_lint(cx, ty, sp, reason, help, fn_mode); + } + } + } + + fn emit_ffi_unsafe_type_lint( + &self, + cx: &LateContext<'tcx>, + ty: Ty<'tcx>, + sp: Span, + note: DiagMessage, + help: Option<DiagMessage>, + fn_mode: CItemKind, + ) { + let lint = match fn_mode { + CItemKind::Declaration => IMPROPER_CTYPES, + CItemKind::Definition => IMPROPER_CTYPES_DEFINITIONS, + }; + let desc = match fn_mode { + CItemKind::Declaration => "block", + CItemKind::Definition => "fn", + }; + let span_note = if let ty::Adt(def, _) = ty.kind() + && let Some(sp) = cx.tcx.hir_span_if_local(def.did()) + { + Some(sp) + } else { + None + }; + cx.emit_span_lint(lint, sp, ImproperCTypes { ty, desc, label: sp, help, note, span_note }); + } +} + +/// `ImproperCTypesDefinitions` checks items outside of foreign items (e.g. stuff that isn't in +/// `extern "C" { }` blocks): +/// +/// - `extern "<abi>" fn` definitions are checked in the same way as the +/// `ImproperCtypesDeclarations` visitor checks functions if `<abi>` is external (e.g. "C"). +/// - All other items which contain types (e.g. other functions, struct definitions, etc) are +/// checked for extern fn-ptrs with external ABIs. +impl<'tcx> LateLintPass<'tcx> for ImproperCTypesLint { + fn check_foreign_item(&mut self, cx: &LateContext<'tcx>, it: &hir::ForeignItem<'tcx>) { + let abi = cx.tcx.hir_get_foreign_abi(it.hir_id()); + + match it.kind { + hir::ForeignItemKind::Fn(sig, _, _) => { + // fnptrs are a special case, they always need to be treated as + // "the element rendered unsafe" because their unsafety doesn't affect + // their surroundings, and their type is often declared inline + if !abi.is_rustic_abi() { + self.check_foreign_fn(cx, CItemKind::Declaration, it.owner_id.def_id, sig.decl); + } else { + self.check_fn_for_external_abi_fnptr( + cx, + CItemKind::Declaration, + it.owner_id.def_id, + sig.decl, + ); + } + } + hir::ForeignItemKind::Static(ty, _, _) if !abi.is_rustic_abi() => { + self.check_foreign_static(cx, it.owner_id, ty.span); + } + hir::ForeignItemKind::Static(..) | hir::ForeignItemKind::Type => (), + } + } + + fn check_item(&mut self, cx: &LateContext<'tcx>, item: &'tcx hir::Item<'tcx>) { + match item.kind { + hir::ItemKind::Static(_, _, ty, _) + | hir::ItemKind::Const(_, _, ty, _) + | hir::ItemKind::TyAlias(_, _, ty) => { + self.check_type_for_external_abi_fnptr( + cx, + VisitorState::STATIC_TY, + ty, + cx.tcx.type_of(item.owner_id).instantiate_identity(), + CItemKind::Definition, + ); + } + // See `check_fn` for declarations, `check_foreign_items` for definitions in extern blocks + hir::ItemKind::Fn { .. } => {} + hir::ItemKind::Struct(..) | hir::ItemKind::Union(..) | hir::ItemKind::Enum(..) => { + // looking for extern FnPtr:s is delegated to `check_field_def`. + let adt_def: AdtDef<'tcx> = cx.tcx.adt_def(item.owner_id.to_def_id()); + + if adt_def.repr().c() && !adt_def.repr().packed() && adt_def.repr().align.is_none() + { + self.check_reprc_adt(cx, item, adt_def); + } + } + + // Doesn't define something that can contain a external type to be checked. + hir::ItemKind::Impl(..) + | hir::ItemKind::TraitAlias(..) + | hir::ItemKind::Trait(..) + | hir::ItemKind::GlobalAsm { .. } + | hir::ItemKind::ForeignMod { .. } + | hir::ItemKind::Mod(..) + | hir::ItemKind::Macro(..) + | hir::ItemKind::Use(..) + | hir::ItemKind::ExternCrate(..) => {} + } + } + + fn check_field_def(&mut self, cx: &LateContext<'tcx>, field: &'tcx hir::FieldDef<'tcx>) { + self.check_type_for_external_abi_fnptr( + cx, + VisitorState::STATIC_TY, + field.ty, + cx.tcx.type_of(field.def_id).instantiate_identity(), + CItemKind::Definition, + ); + } + + fn check_fn( + &mut self, + cx: &LateContext<'tcx>, + kind: hir::intravisit::FnKind<'tcx>, + decl: &'tcx hir::FnDecl<'_>, + _: &'tcx hir::Body<'_>, + _: Span, + id: LocalDefId, + ) { + use hir::intravisit::FnKind; + + let abi = match kind { + FnKind::ItemFn(_, _, header, ..) => header.abi, + FnKind::Method(_, sig, ..) => sig.header.abi, + _ => return, + }; + + // fnptrs are a special case, they always need to be treated as + // "the element rendered unsafe" because their unsafety doesn't affect + // their surroundings, and their type is often declared inline + if !abi.is_rustic_abi() { + self.check_foreign_fn(cx, CItemKind::Definition, id, decl); + } else { + self.check_fn_for_external_abi_fnptr(cx, CItemKind::Definition, id, decl); + } + } +} diff --git a/compiler/rustc_lint_defs/src/builtin.rs b/compiler/rustc_lint_defs/src/builtin.rs index ce92263babd..b79075ac09b 100644 --- a/compiler/rustc_lint_defs/src/builtin.rs +++ b/compiler/rustc_lint_defs/src/builtin.rs @@ -5141,3 +5141,57 @@ declare_lint! { "detects tail calls of functions marked with `#[track_caller]`", @feature_gate = explicit_tail_calls; } +declare_lint! { + /// The `inline_always_mismatching_target_features` lint will trigger when a + /// function with the `#[inline(always)]` and `#[target_feature(enable = "...")]` + /// attributes is called and cannot be inlined due to missing target features in the caller. + /// + /// ### Example + /// + /// ```rust,ignore (fails on x86_64) + /// #[inline(always)] + /// #[target_feature(enable = "fp16")] + /// unsafe fn callee() { + /// // operations using fp16 types + /// } + /// + /// // Caller does not enable the required target feature + /// fn caller() { + /// unsafe { callee(); } + /// } + /// + /// fn main() { + /// caller(); + /// } + /// ``` + /// + /// This will produce: + /// + /// ```text + /// warning: call to `#[inline(always)]`-annotated `callee` requires the same target features. Function will not have `alwaysinline` attribute applied + /// --> $DIR/builtin.rs:5192:14 + /// | + /// 10 | unsafe { callee(); } + /// | ^^^^^^^^ + /// | + /// note: `fp16` target feature enabled in `callee` here but missing from `caller` + /// --> $DIR/builtin.rs:5185:1 + /// | + /// 3 | #[target_feature(enable = "fp16")] + /// | ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ + /// 4 | unsafe fn callee() { + /// | ------------------ + /// = note: `#[warn(inline_always_mismatching_target_features)]` on by default + /// warning: 1 warning emitted + /// ``` + /// + /// ### Explanation + /// + /// Inlining a function with a target feature attribute into a caller that + /// lacks the corresponding target feature can lead to unsound behavior. + /// LLVM may select the wrong instructions or registers, or reorder + /// operations, potentially resulting in runtime errors. + pub INLINE_ALWAYS_MISMATCHING_TARGET_FEATURES, + Warn, + r#"detects when a function annotated with `#[inline(always)]` and `#[target_feature(enable = "..")]` is inlined into a caller without the required target feature"#, +} diff --git a/compiler/rustc_llvm/build.rs b/compiler/rustc_llvm/build.rs index d01f79dcade..225ab8c846d 100644 --- a/compiler/rustc_llvm/build.rs +++ b/compiler/rustc_llvm/build.rs @@ -174,10 +174,10 @@ fn main() { // Prevent critical warnings when we're compiling from rust-lang/rust CI, // except on MSVC, as the compiler throws warnings that are only reported - // for this platform. See https://github.com/rust-lang/rust/pull/145031#issuecomment-3162677202 - // FIXME(llvm22): It looks like the specific problem code has been removed - // in https://github.com/llvm/llvm-project/commit/e8fc808bf8e78a3c80d1f8e293a92677b92366dd, - // retry msvc once we bump our LLVM version. + // for this platform. See https://github.com/rust-lang/rust/pull/145031#issuecomment-3162677202. + // Moreover, LLVM generally guarantees warning-freedom only when building with Clang, as other + // compilers have too many false positives. This is typically the case for MSVC, which throws + // many false-positive warnings. We keep it excluded, for these reasons. if std::env::var_os("CI").is_some() && !target.contains("msvc") { cfg.warnings_into_errors(true); } diff --git a/compiler/rustc_llvm/llvm-wrapper/PassWrapper.cpp b/compiler/rustc_llvm/llvm-wrapper/PassWrapper.cpp index dd492325814..3bb1533c2fe 100644 --- a/compiler/rustc_llvm/llvm-wrapper/PassWrapper.cpp +++ b/compiler/rustc_llvm/llvm-wrapper/PassWrapper.cpp @@ -1568,12 +1568,11 @@ extern "C" bool LLVMRustPrepareThinLTOImport(const LLVMRustThinLTOData *Data, return true; } -extern "C" LLVMRustThinLTOBuffer * -LLVMRustThinLTOBufferCreate(LLVMModuleRef M, bool is_thin, bool emit_summary) { +extern "C" LLVMRustThinLTOBuffer *LLVMRustThinLTOBufferCreate(LLVMModuleRef M, + bool is_thin) { auto Ret = std::make_unique<LLVMRustThinLTOBuffer>(); { auto OS = raw_string_ostream(Ret->data); - auto ThinLinkOS = raw_string_ostream(Ret->thin_link_data); { if (is_thin) { PassBuilder PB; @@ -1587,11 +1586,7 @@ LLVMRustThinLTOBufferCreate(LLVMModuleRef M, bool is_thin, bool emit_summary) { PB.registerLoopAnalyses(LAM); PB.crossRegisterProxies(LAM, FAM, CGAM, MAM); ModulePassManager MPM; - // We only pass ThinLinkOS to be filled in if we want the summary, - // because otherwise LLVM does extra work and may double-emit some - // errors or warnings. - MPM.addPass( - ThinLTOBitcodeWriterPass(OS, emit_summary ? &ThinLinkOS : nullptr)); + MPM.addPass(ThinLTOBitcodeWriterPass(OS, nullptr)); MPM.run(*unwrap(M), MAM); } else { WriteBitcodeToFile(*unwrap(M), OS); diff --git a/compiler/rustc_log/Cargo.toml b/compiler/rustc_log/Cargo.toml index c673d51a1d4..2332ff9b323 100644 --- a/compiler/rustc_log/Cargo.toml +++ b/compiler/rustc_log/Cargo.toml @@ -5,8 +5,8 @@ edition = "2024" [dependencies] # tidy-alphabetical-start -tracing = "0.1.28" -tracing-core = "=0.1.30" # FIXME(Nilstrieb) tracing has a deadlock: https://github.com/tokio-rs/tracing/issues/2635 +# tracing > 0.1.37 have huge binary size / instructions regression +tracing = "=0.1.37" tracing-subscriber = { version = "0.3.3", default-features = false, features = ["fmt", "env-filter", "smallvec", "parking_lot", "ansi"] } tracing-tree = "0.3.1" # tidy-alphabetical-end diff --git a/compiler/rustc_log/src/lib.rs b/compiler/rustc_log/src/lib.rs index df648bbd489..26475eec1c1 100644 --- a/compiler/rustc_log/src/lib.rs +++ b/compiler/rustc_log/src/lib.rs @@ -38,7 +38,7 @@ use std::fmt::{self, Display}; use std::io::{self, IsTerminal}; use tracing::dispatcher::SetGlobalDefaultError; -use tracing_core::{Event, Subscriber}; +use tracing::{Event, Subscriber}; use tracing_subscriber::filter::{Directive, EnvFilter, LevelFilter}; use tracing_subscriber::fmt::FmtContext; use tracing_subscriber::fmt::format::{self, FormatEvent, FormatFields}; diff --git a/compiler/rustc_metadata/src/creader.rs b/compiler/rustc_metadata/src/creader.rs index 5d776ea581d..9e23da88f5e 100644 --- a/compiler/rustc_metadata/src/creader.rs +++ b/compiler/rustc_metadata/src/creader.rs @@ -412,7 +412,7 @@ impl CStore { match (&left_name_val, &right_name_val) { (Some(l), Some(r)) => match l.1.opt.cmp(&r.1.opt) { cmp::Ordering::Equal => { - if l.0.tech_value != r.0.tech_value { + if !l.1.consistent(&tcx.sess.opts, Some(&r.1)) { report_diff( &l.0.prefix, &l.0.name, @@ -424,20 +424,28 @@ impl CStore { right_name_val = None; } cmp::Ordering::Greater => { - report_diff(&r.0.prefix, &r.0.name, None, Some(&r.1.value_name)); + if !r.1.consistent(&tcx.sess.opts, None) { + report_diff(&r.0.prefix, &r.0.name, None, Some(&r.1.value_name)); + } right_name_val = None; } cmp::Ordering::Less => { - report_diff(&l.0.prefix, &l.0.name, Some(&l.1.value_name), None); + if !l.1.consistent(&tcx.sess.opts, None) { + report_diff(&l.0.prefix, &l.0.name, Some(&l.1.value_name), None); + } left_name_val = None; } }, (Some(l), None) => { - report_diff(&l.0.prefix, &l.0.name, Some(&l.1.value_name), None); + if !l.1.consistent(&tcx.sess.opts, None) { + report_diff(&l.0.prefix, &l.0.name, Some(&l.1.value_name), None); + } left_name_val = None; } (None, Some(r)) => { - report_diff(&r.0.prefix, &r.0.name, None, Some(&r.1.value_name)); + if !r.1.consistent(&tcx.sess.opts, None) { + report_diff(&r.0.prefix, &r.0.name, None, Some(&r.1.value_name)); + } right_name_val = None; } (None, None) => break, diff --git a/compiler/rustc_middle/src/middle/codegen_fn_attrs.rs b/compiler/rustc_middle/src/middle/codegen_fn_attrs.rs index 866736f74a0..8b4503073b0 100644 --- a/compiler/rustc_middle/src/middle/codegen_fn_attrs.rs +++ b/compiler/rustc_middle/src/middle/codegen_fn_attrs.rs @@ -82,7 +82,7 @@ pub enum TargetFeatureKind { Forced, } -#[derive(Copy, Clone, Debug, TyEncodable, TyDecodable, HashStable)] +#[derive(Copy, Clone, Debug, Eq, PartialEq, TyEncodable, TyDecodable, HashStable)] pub struct TargetFeature { /// The name of the target feature (e.g. "avx") pub name: Symbol, diff --git a/compiler/rustc_middle/src/middle/region.rs b/compiler/rustc_middle/src/middle/region.rs index 857d041224f..5367e5edd49 100644 --- a/compiler/rustc_middle/src/middle/region.rs +++ b/compiler/rustc_middle/src/middle/region.rs @@ -299,4 +299,43 @@ impl ScopeTree { true } + + /// Returns the scope of non-lifetime-extended temporaries within a given scope, as well as + /// whether we've recorded a potential backwards-incompatible change to lint on. + /// Returns `None` when no enclosing temporary scope is found, such as for static items. + pub fn default_temporary_scope(&self, inner: Scope) -> (Option<Scope>, Option<Scope>) { + let mut id = inner; + let mut backwards_incompatible = None; + + while let Some(&p) = self.parent_map.get(&id) { + match p.data { + ScopeData::Destruction => { + debug!("temporary_scope({inner:?}) = {id:?} [enclosing]"); + return (Some(id), backwards_incompatible); + } + ScopeData::IfThenRescope | ScopeData::MatchGuard => { + debug!("temporary_scope({inner:?}) = {p:?} [enclosing]"); + return (Some(p), backwards_incompatible); + } + ScopeData::Node + | ScopeData::CallSite + | ScopeData::Arguments + | ScopeData::IfThen + | ScopeData::Remainder(_) => { + // If we haven't already passed through a backwards-incompatible node, + // then check if we are passing through one now and record it if so. + // This is for now only working for cases where a temporary lifetime is + // *shortened*. + if backwards_incompatible.is_none() { + backwards_incompatible = + self.backwards_incompatible_scope.get(&p.local_id).copied(); + } + id = p + } + } + } + + debug!("temporary_scope({inner:?}) = None"); + (None, backwards_incompatible) + } } diff --git a/compiler/rustc_middle/src/mir/consts.rs b/compiler/rustc_middle/src/mir/consts.rs index 164433aede2..5764a9c84ee 100644 --- a/compiler/rustc_middle/src/mir/consts.rs +++ b/compiler/rustc_middle/src/mir/consts.rs @@ -579,7 +579,7 @@ impl<'tcx> Display for Const<'tcx> { } /////////////////////////////////////////////////////////////////////////// -/// Const-related utilities +// Const-related utilities impl<'tcx> TyCtxt<'tcx> { pub fn span_as_caller_location(self, span: Span) -> ConstValue { diff --git a/compiler/rustc_middle/src/mir/statement.rs b/compiler/rustc_middle/src/mir/statement.rs index 6e52bc775ef..ec2a8e86077 100644 --- a/compiler/rustc_middle/src/mir/statement.rs +++ b/compiler/rustc_middle/src/mir/statement.rs @@ -408,14 +408,14 @@ impl<'tcx> Place<'tcx> { self.as_ref().project_deeper(more_projections, tcx) } - pub fn ty_from<D: ?Sized>( + pub fn ty_from<D>( local: Local, projection: &[PlaceElem<'tcx>], local_decls: &D, tcx: TyCtxt<'tcx>, ) -> PlaceTy<'tcx> where - D: HasLocalDecls<'tcx>, + D: ?Sized + HasLocalDecls<'tcx>, { PlaceTy::from_ty(local_decls.local_decls()[local].ty).multi_projection_ty(tcx, projection) } @@ -529,9 +529,9 @@ impl<'tcx> PlaceRef<'tcx> { Place { local: self.local, projection: tcx.mk_place_elems(new_projections) } } - pub fn ty<D: ?Sized>(&self, local_decls: &D, tcx: TyCtxt<'tcx>) -> PlaceTy<'tcx> + pub fn ty<D>(&self, local_decls: &D, tcx: TyCtxt<'tcx>) -> PlaceTy<'tcx> where - D: HasLocalDecls<'tcx>, + D: ?Sized + HasLocalDecls<'tcx>, { Place::ty_from(self.local, self.projection, local_decls, tcx) } @@ -630,9 +630,9 @@ impl<'tcx> Operand<'tcx> { if let ty::FnDef(def_id, args) = *const_ty.kind() { Some((def_id, args)) } else { None } } - pub fn ty<D: ?Sized>(&self, local_decls: &D, tcx: TyCtxt<'tcx>) -> Ty<'tcx> + pub fn ty<D>(&self, local_decls: &D, tcx: TyCtxt<'tcx>) -> Ty<'tcx> where - D: HasLocalDecls<'tcx>, + D: ?Sized + HasLocalDecls<'tcx>, { match self { &Operand::Copy(ref l) | &Operand::Move(ref l) => l.ty(local_decls, tcx).ty, @@ -640,9 +640,9 @@ impl<'tcx> Operand<'tcx> { } } - pub fn span<D: ?Sized>(&self, local_decls: &D) -> Span + pub fn span<D>(&self, local_decls: &D) -> Span where - D: HasLocalDecls<'tcx>, + D: ?Sized + HasLocalDecls<'tcx>, { match self { &Operand::Copy(ref l) | &Operand::Move(ref l) => { @@ -674,7 +674,7 @@ impl<'tcx> ConstOperand<'tcx> { } /////////////////////////////////////////////////////////////////////////// -/// Rvalues +// Rvalues pub enum RvalueInitializationState { Shallow, @@ -721,9 +721,9 @@ impl<'tcx> Rvalue<'tcx> { } } - pub fn ty<D: ?Sized>(&self, local_decls: &D, tcx: TyCtxt<'tcx>) -> Ty<'tcx> + pub fn ty<D>(&self, local_decls: &D, tcx: TyCtxt<'tcx>) -> Ty<'tcx> where - D: HasLocalDecls<'tcx>, + D: ?Sized + HasLocalDecls<'tcx>, { match *self { Rvalue::Use(ref operand) => operand.ty(local_decls, tcx), diff --git a/compiler/rustc_middle/src/mir/visit.rs b/compiler/rustc_middle/src/mir/visit.rs index 904d78d69b6..b498b7b8912 100644 --- a/compiler/rustc_middle/src/mir/visit.rs +++ b/compiler/rustc_middle/src/mir/visit.rs @@ -1415,6 +1415,24 @@ impl PlaceContext { ) } + /// Returns `true` if this place context may be used to know the address of the given place. + #[inline] + pub fn may_observe_address(self) -> bool { + matches!( + self, + PlaceContext::NonMutatingUse( + NonMutatingUseContext::SharedBorrow + | NonMutatingUseContext::RawBorrow + | NonMutatingUseContext::FakeBorrow + ) | PlaceContext::MutatingUse( + MutatingUseContext::Drop + | MutatingUseContext::Borrow + | MutatingUseContext::RawBorrow + | MutatingUseContext::AsmOutput + ) + ) + } + /// Returns `true` if this place context represents a storage live or storage dead marker. #[inline] pub fn is_storage_marker(self) -> bool { diff --git a/compiler/rustc_middle/src/ty/codec.rs b/compiler/rustc_middle/src/ty/codec.rs index 95759d1f31a..3f37595d0ee 100644 --- a/compiler/rustc_middle/src/ty/codec.rs +++ b/compiler/rustc_middle/src/ty/codec.rs @@ -216,10 +216,7 @@ impl<'tcx, E: TyEncoder<'tcx>> Encodable<E> for ty::ParamEnv<'tcx> { #[inline] fn decode_arena_allocable<'tcx, D: TyDecoder<'tcx>, T: ArenaAllocatable<'tcx> + Decodable<D>>( decoder: &mut D, -) -> &'tcx T -where - D: TyDecoder<'tcx>, -{ +) -> &'tcx T { decoder.interner().arena.alloc(Decodable::decode(decoder)) } @@ -230,10 +227,7 @@ fn decode_arena_allocable_slice< T: ArenaAllocatable<'tcx> + Decodable<D>, >( decoder: &mut D, -) -> &'tcx [T] -where - D: TyDecoder<'tcx>, -{ +) -> &'tcx [T] { decoder.interner().arena.alloc_from_iter(<Vec<T> as Decodable<D>>::decode(decoder)) } diff --git a/compiler/rustc_middle/src/ty/context.rs b/compiler/rustc_middle/src/ty/context.rs index 72ab6ac612c..97c8a0336c2 100644 --- a/compiler/rustc_middle/src/ty/context.rs +++ b/compiler/rustc_middle/src/ty/context.rs @@ -738,8 +738,8 @@ impl<'tcx> Interner for TyCtxt<'tcx> { ) } - type ProbeRef = &'tcx inspect::Probe<TyCtxt<'tcx>>; - fn mk_probe_ref(self, probe: inspect::Probe<Self>) -> &'tcx inspect::Probe<TyCtxt<'tcx>> { + type Probe = &'tcx inspect::Probe<TyCtxt<'tcx>>; + fn mk_probe(self, probe: inspect::Probe<Self>) -> &'tcx inspect::Probe<TyCtxt<'tcx>> { self.arena.alloc(probe) } fn evaluate_root_goal_for_proof_tree_raw( diff --git a/compiler/rustc_middle/src/ty/rvalue_scopes.rs b/compiler/rustc_middle/src/ty/rvalue_scopes.rs index 7dfe2d28051..8b92e48ed1a 100644 --- a/compiler/rustc_middle/src/ty/rvalue_scopes.rs +++ b/compiler/rustc_middle/src/ty/rvalue_scopes.rs @@ -35,41 +35,8 @@ impl RvalueScopes { // if there's one. Static items, for instance, won't // have an enclosing scope, hence no scope will be // returned. - let mut id = Scope { local_id: expr_id, data: ScopeData::Node }; - let mut backwards_incompatible = None; - - while let Some(&p) = region_scope_tree.parent_map.get(&id) { - match p.data { - ScopeData::Destruction => { - debug!("temporary_scope({expr_id:?}) = {id:?} [enclosing]"); - return (Some(id), backwards_incompatible); - } - ScopeData::IfThenRescope | ScopeData::MatchGuard => { - debug!("temporary_scope({expr_id:?}) = {p:?} [enclosing]"); - return (Some(p), backwards_incompatible); - } - ScopeData::Node - | ScopeData::CallSite - | ScopeData::Arguments - | ScopeData::IfThen - | ScopeData::Remainder(_) => { - // If we haven't already passed through a backwards-incompatible node, - // then check if we are passing through one now and record it if so. - // This is for now only working for cases where a temporary lifetime is - // *shortened*. - if backwards_incompatible.is_none() { - backwards_incompatible = region_scope_tree - .backwards_incompatible_scope - .get(&p.local_id) - .copied(); - } - id = p - } - } - } - - debug!("temporary_scope({expr_id:?}) = None"); - (None, backwards_incompatible) + region_scope_tree + .default_temporary_scope(Scope { local_id: expr_id, data: ScopeData::Node }) } /// Make an association between a sub-expression and an extended lifetime diff --git a/compiler/rustc_mir_dataflow/src/impls/liveness.rs b/compiler/rustc_mir_dataflow/src/impls/liveness.rs index 6ec1b03a34e..5eba474a60c 100644 --- a/compiler/rustc_mir_dataflow/src/impls/liveness.rs +++ b/compiler/rustc_mir_dataflow/src/impls/liveness.rs @@ -92,7 +92,7 @@ impl<'tcx> Visitor<'tcx> for TransferFunction<'_> { } match DefUse::for_place(*place, context) { - Some(DefUse::Def) => { + DefUse::Def => { if let PlaceContext::MutatingUse( MutatingUseContext::Call | MutatingUseContext::AsmOutput, ) = context @@ -105,8 +105,8 @@ impl<'tcx> Visitor<'tcx> for TransferFunction<'_> { self.0.kill(place.local); } } - Some(DefUse::Use) => self.0.gen_(place.local), - None => {} + DefUse::Use => self.0.gen_(place.local), + DefUse::PartialWrite | DefUse::NonUse => {} } self.visit_projection(place.as_ref(), context, location); @@ -131,23 +131,29 @@ impl<'tcx> Visitor<'tcx> for YieldResumeEffect<'_> { } #[derive(Eq, PartialEq, Clone)] -enum DefUse { +pub enum DefUse { + /// Full write to the local. Def, + /// Read of any part of the local. Use, + /// Partial write to the local. + PartialWrite, + /// Non-use, like debuginfo. + NonUse, } impl DefUse { fn apply(state: &mut DenseBitSet<Local>, place: Place<'_>, context: PlaceContext) { match DefUse::for_place(place, context) { - Some(DefUse::Def) => state.kill(place.local), - Some(DefUse::Use) => state.gen_(place.local), - None => {} + DefUse::Def => state.kill(place.local), + DefUse::Use => state.gen_(place.local), + DefUse::PartialWrite | DefUse::NonUse => {} } } - fn for_place(place: Place<'_>, context: PlaceContext) -> Option<DefUse> { + pub fn for_place(place: Place<'_>, context: PlaceContext) -> DefUse { match context { - PlaceContext::NonUse(_) => None, + PlaceContext::NonUse(_) => DefUse::NonUse, PlaceContext::MutatingUse( MutatingUseContext::Call @@ -156,21 +162,20 @@ impl DefUse { | MutatingUseContext::Store | MutatingUseContext::Deinit, ) => { + // Treat derefs as a use of the base local. `*p = 4` is not a def of `p` but a use. if place.is_indirect() { - // Treat derefs as a use of the base local. `*p = 4` is not a def of `p` but a - // use. - Some(DefUse::Use) + DefUse::Use } else if place.projection.is_empty() { - Some(DefUse::Def) + DefUse::Def } else { - None + DefUse::PartialWrite } } // Setting the discriminant is not a use because it does no reading, but it is also not // a def because it does not overwrite the whole place PlaceContext::MutatingUse(MutatingUseContext::SetDiscriminant) => { - place.is_indirect().then_some(DefUse::Use) + if place.is_indirect() { DefUse::Use } else { DefUse::PartialWrite } } // All other contexts are uses... @@ -188,7 +193,7 @@ impl DefUse { | NonMutatingUseContext::PlaceMention | NonMutatingUseContext::FakeBorrow | NonMutatingUseContext::SharedBorrow, - ) => Some(DefUse::Use), + ) => DefUse::Use, PlaceContext::MutatingUse(MutatingUseContext::Projection) | PlaceContext::NonMutatingUse(NonMutatingUseContext::Projection) => { diff --git a/compiler/rustc_mir_dataflow/src/impls/mod.rs b/compiler/rustc_mir_dataflow/src/impls/mod.rs index 3f29b819a6d..6d573e1c00e 100644 --- a/compiler/rustc_mir_dataflow/src/impls/mod.rs +++ b/compiler/rustc_mir_dataflow/src/impls/mod.rs @@ -9,7 +9,8 @@ pub use self::initialized::{ MaybeUninitializedPlaces, MaybeUninitializedPlacesDomain, }; pub use self::liveness::{ - MaybeLiveLocals, MaybeTransitiveLiveLocals, TransferFunction as LivenessTransferFunction, + DefUse, MaybeLiveLocals, MaybeTransitiveLiveLocals, + TransferFunction as LivenessTransferFunction, }; pub use self::storage_liveness::{ MaybeRequiresStorage, MaybeStorageDead, MaybeStorageLive, always_storage_live_locals, diff --git a/compiler/rustc_mir_dataflow/src/points.rs b/compiler/rustc_mir_dataflow/src/points.rs index 70d1a34b5fb..e3d1e04a319 100644 --- a/compiler/rustc_mir_dataflow/src/points.rs +++ b/compiler/rustc_mir_dataflow/src/points.rs @@ -1,9 +1,5 @@ -use rustc_index::bit_set::DenseBitSet; -use rustc_index::interval::SparseIntervalMatrix; use rustc_index::{Idx, IndexVec}; -use rustc_middle::mir::{self, BasicBlock, Body, Location}; - -use crate::framework::{Analysis, Results, ResultsVisitor, visit_results}; +use rustc_middle::mir::{BasicBlock, Body, Location}; /// Maps between a `Location` and a `PointIndex` (and vice versa). pub struct DenseLocationMap { @@ -93,65 +89,3 @@ rustc_index::newtype_index! { #[debug_format = "PointIndex({})"] pub struct PointIndex {} } - -/// Add points depending on the result of the given dataflow analysis. -pub fn save_as_intervals<'tcx, N, A>( - elements: &DenseLocationMap, - body: &mir::Body<'tcx>, - mut analysis: A, - results: Results<A::Domain>, -) -> SparseIntervalMatrix<N, PointIndex> -where - N: Idx, - A: Analysis<'tcx, Domain = DenseBitSet<N>>, -{ - let values = SparseIntervalMatrix::new(elements.num_points()); - let mut visitor = Visitor { elements, values }; - visit_results( - body, - body.basic_blocks.reverse_postorder().iter().copied(), - &mut analysis, - &results, - &mut visitor, - ); - visitor.values -} - -struct Visitor<'a, N: Idx> { - elements: &'a DenseLocationMap, - values: SparseIntervalMatrix<N, PointIndex>, -} - -impl<'tcx, A, N> ResultsVisitor<'tcx, A> for Visitor<'_, N> -where - A: Analysis<'tcx, Domain = DenseBitSet<N>>, - N: Idx, -{ - fn visit_after_primary_statement_effect<'mir>( - &mut self, - _analysis: &mut A, - state: &A::Domain, - _statement: &'mir mir::Statement<'tcx>, - location: Location, - ) { - let point = self.elements.point_from_location(location); - // Use internal iterator manually as it is much more efficient. - state.iter().for_each(|node| { - self.values.insert(node, point); - }); - } - - fn visit_after_primary_terminator_effect<'mir>( - &mut self, - _analysis: &mut A, - state: &A::Domain, - _terminator: &'mir mir::Terminator<'tcx>, - location: Location, - ) { - let point = self.elements.point_from_location(location); - // Use internal iterator manually as it is much more efficient. - state.iter().for_each(|node| { - self.values.insert(node, point); - }); - } -} diff --git a/compiler/rustc_mir_dataflow/src/value_analysis.rs b/compiler/rustc_mir_dataflow/src/value_analysis.rs index 005e7973130..9a00831dc01 100644 --- a/compiler/rustc_mir_dataflow/src/value_analysis.rs +++ b/compiler/rustc_mir_dataflow/src/value_analysis.rs @@ -6,7 +6,7 @@ use rustc_data_structures::fx::{FxHashMap, FxIndexSet, StdEntry}; use rustc_data_structures::stack::ensure_sufficient_stack; use rustc_index::IndexVec; use rustc_index::bit_set::DenseBitSet; -use rustc_middle::mir::visit::{MutatingUseContext, PlaceContext, Visitor}; +use rustc_middle::mir::visit::{PlaceContext, Visitor}; use rustc_middle::mir::*; use rustc_middle::ty::{self, Ty, TyCtxt}; use tracing::debug; @@ -917,12 +917,7 @@ pub fn excluded_locals(body: &Body<'_>) -> DenseBitSet<Local> { impl<'tcx> Visitor<'tcx> for Collector { fn visit_place(&mut self, place: &Place<'tcx>, context: PlaceContext, _location: Location) { - if (context.is_borrow() - || context.is_address_of() - || context.is_drop() - || context == PlaceContext::MutatingUse(MutatingUseContext::AsmOutput)) - && !place.is_indirect() - { + if context.may_observe_address() && !place.is_indirect() { // A pointer to a place could be used to access other places with the same local, // hence we have to exclude the local completely. self.result.insert(place.local); diff --git a/compiler/rustc_mir_transform/src/check_inline_always_target_features.rs b/compiler/rustc_mir_transform/src/check_inline_always_target_features.rs new file mode 100644 index 00000000000..abad28f0a8f --- /dev/null +++ b/compiler/rustc_mir_transform/src/check_inline_always_target_features.rs @@ -0,0 +1,88 @@ +use rustc_hir::attrs::InlineAttr; +use rustc_middle::middle::codegen_fn_attrs::{TargetFeature, TargetFeatureKind}; +use rustc_middle::mir::{Body, TerminatorKind}; +use rustc_middle::ty::{self, TyCtxt}; + +use crate::pass_manager::MirLint; + +pub(super) struct CheckInlineAlwaysTargetFeature; + +impl<'tcx> MirLint<'tcx> for CheckInlineAlwaysTargetFeature { + fn run_lint(&self, tcx: TyCtxt<'tcx>, body: &Body<'tcx>) { + check_inline_always_target_features(tcx, body) + } +} + +/// `#[target_feature]`-annotated functions can be marked `#[inline]` and will only be inlined if +/// the target features match (as well as all of the other inlining heuristics). `#[inline(always)]` +/// will always inline regardless of matching target features, which can result in errors from LLVM. +/// However, it is desirable to be able to always annotate certain functions (e.g. SIMD intrinsics) +/// as `#[inline(always)]` but check the target features match in Rust to avoid the LLVM errors. +/// +/// We check the caller and callee target features to ensure that this can +/// be done or emit a lint. +#[inline] +fn check_inline_always_target_features<'tcx>(tcx: TyCtxt<'tcx>, body: &Body<'tcx>) { + let caller_def_id = body.source.def_id().expect_local(); + if !tcx.def_kind(caller_def_id).has_codegen_attrs() { + return; + } + + let caller_codegen_fn_attrs = tcx.codegen_fn_attrs(caller_def_id); + + for bb in body.basic_blocks.iter() { + let terminator = bb.terminator(); + match &terminator.kind { + TerminatorKind::Call { func, .. } | TerminatorKind::TailCall { func, .. } => { + let fn_ty = func.ty(body, tcx); + let ty::FnDef(callee_def_id, _) = *fn_ty.kind() else { + continue; + }; + + if !tcx.def_kind(callee_def_id).has_codegen_attrs() { + continue; + } + let callee_codegen_fn_attrs = tcx.codegen_fn_attrs(callee_def_id); + if callee_codegen_fn_attrs.inline != InlineAttr::Always + || callee_codegen_fn_attrs.target_features.is_empty() + { + continue; + } + + // Scan the users defined target features and ensure they + // match the caller. + if tcx.is_target_feature_call_safe( + &callee_codegen_fn_attrs.target_features, + &caller_codegen_fn_attrs + .target_features + .iter() + .cloned() + .chain(tcx.sess.target_features.iter().map(|feat| TargetFeature { + name: *feat, + kind: TargetFeatureKind::Implied, + })) + .collect::<Vec<_>>(), + ) { + continue; + } + + let callee_only: Vec<_> = callee_codegen_fn_attrs + .target_features + .iter() + .filter(|it| !caller_codegen_fn_attrs.target_features.contains(it)) + .filter(|it| !matches!(it.kind, TargetFeatureKind::Implied)) + .map(|it| it.name.as_str()) + .collect(); + + crate::errors::emit_inline_always_target_feature_diagnostic( + tcx, + terminator.source_info.span, + callee_def_id, + caller_def_id.into(), + &callee_only, + ); + } + _ => (), + } + } +} diff --git a/compiler/rustc_mir_transform/src/coverage/expansion.rs b/compiler/rustc_mir_transform/src/coverage/expansion.rs index 91e0528f52f..851bbaeed48 100644 --- a/compiler/rustc_mir_transform/src/coverage/expansion.rs +++ b/compiler/rustc_mir_transform/src/coverage/expansion.rs @@ -82,7 +82,7 @@ impl ExpnNode { Self { expn_id, - expn_kind: expn_data.kind.clone(), + expn_kind: expn_data.kind, call_site, call_site_expn_id, diff --git a/compiler/rustc_mir_transform/src/dest_prop.rs b/compiler/rustc_mir_transform/src/dest_prop.rs index cf7425251e8..9ba2d274691 100644 --- a/compiler/rustc_mir_transform/src/dest_prop.rs +++ b/compiler/rustc_mir_transform/src/dest_prop.rs @@ -59,6 +59,12 @@ //! The first two conditions are simple structural requirements on the `Assign` statements that can //! be trivially checked. The third requirement however is more difficult and costly to check. //! +//! ## Current implementation +//! +//! The current implementation relies on live range computation to check for conflicts. We only +//! allow to merge locals that have disjoint live ranges. The live range are defined with +//! half-statement granularity, so as to make all writes be live for at least a half statement. +//! //! ## Future Improvements //! //! There are a number of ways in which this pass could be improved in the future: @@ -117,9 +123,8 @@ //! - Layout optimizations for coroutines have been added to improve code generation for //! async/await, which are very similar in spirit to what this optimization does. //! -//! Also, rustc now has a simple NRVO pass (see `nrvo.rs`), which handles a subset of the cases that -//! this destination propagation pass handles, proving that similar optimizations can be performed -//! on MIR. +//! [The next approach][attempt 4] computes a conflict matrix between locals by forbidding merging +//! locals with competing writes or with one write while the other is live. //! //! ## Pre/Post Optimization //! @@ -130,20 +135,18 @@ //! [attempt 1]: https://github.com/rust-lang/rust/pull/47954 //! [attempt 2]: https://github.com/rust-lang/rust/pull/71003 //! [attempt 3]: https://github.com/rust-lang/rust/pull/72632 +//! [attempt 4]: https://github.com/rust-lang/rust/pull/96451 -use rustc_data_structures::fx::{FxIndexMap, IndexEntry, IndexOccupiedEntry}; +use rustc_data_structures::union_find::UnionFind; use rustc_index::bit_set::DenseBitSet; use rustc_index::interval::SparseIntervalMatrix; -use rustc_middle::bug; +use rustc_index::{IndexVec, newtype_index}; use rustc_middle::mir::visit::{MutVisitor, PlaceContext, Visitor}; -use rustc_middle::mir::{ - Body, HasLocalDecls, InlineAsmOperand, Local, LocalKind, Location, MirDumper, Operand, - PassWhere, Place, Rvalue, Statement, StatementKind, TerminatorKind, traversal, -}; +use rustc_middle::mir::*; use rustc_middle::ty::TyCtxt; -use rustc_mir_dataflow::Analysis; -use rustc_mir_dataflow::impls::MaybeLiveLocals; -use rustc_mir_dataflow::points::{DenseLocationMap, PointIndex, save_as_intervals}; +use rustc_mir_dataflow::impls::{DefUse, MaybeLiveLocals}; +use rustc_mir_dataflow::points::DenseLocationMap; +use rustc_mir_dataflow::{Analysis, Results}; use tracing::{debug, trace}; pub(super) struct DestinationPropagation; @@ -161,84 +164,81 @@ impl<'tcx> crate::MirPass<'tcx> for DestinationPropagation { sess.mir_opt_level() >= 3 } + #[tracing::instrument(level = "trace", skip(self, tcx, body))] fn run_pass(&self, tcx: TyCtxt<'tcx>, body: &mut Body<'tcx>) { let def_id = body.source.def_id(); - let mut candidates = Candidates::default(); - let mut write_info = WriteInfo::default(); - trace!(func = ?tcx.def_path_str(def_id)); + trace!(?def_id); let borrowed = rustc_mir_dataflow::impls::borrowed_locals(body); + let candidates = Candidates::find(body, &borrowed); + trace!(?candidates); + if candidates.c.is_empty() { + return; + } + let live = MaybeLiveLocals.iterate_to_fixpoint(tcx, body, Some("MaybeLiveLocals-DestProp")); + let points = DenseLocationMap::new(body); - let mut live = save_as_intervals(&points, body, live.analysis, live.results); - - // In order to avoid having to collect data for every single pair of locals in the body, we - // do not allow doing more than one merge for places that are derived from the same local at - // once. To avoid missed opportunities, we instead iterate to a fixed point - we'll refer to - // each of these iterations as a "round." - // - // Reaching a fixed point could in theory take up to `min(l, s)` rounds - however, we do not - // expect to see MIR like that. To verify this, a test was run against `[rust-lang/regex]` - - // the average MIR body saw 1.32 full iterations of this loop. The most that was hit were 30 - // for a single function. Only 80/2801 (2.9%) of functions saw at least 5. - // - // [rust-lang/regex]: - // https://github.com/rust-lang/regex/tree/b5372864e2df6a2f5e543a556a62197f50ca3650 - let mut round_count = 0; - loop { - // PERF: Can we do something smarter than recalculating the candidates and liveness - // results? - candidates.reset_and_find(body, &borrowed); - trace!(?candidates); - dest_prop_mir_dump(tcx, body, &points, &live, round_count); - - FilterInformation::filter_liveness( - &mut candidates, - &points, - &live, - &mut write_info, - body, - ); - - // Because we only filter once per round, it is unsound to use a local for more than - // one merge operation within a single round of optimizations. We store here which ones - // we have already used. - let mut merged_locals: DenseBitSet<Local> = - DenseBitSet::new_empty(body.local_decls.len()); - - // This is the set of merges we will apply this round. It is a subset of the candidates. - let mut merges = FxIndexMap::default(); - - for (src, candidates) in candidates.c.iter() { - if merged_locals.contains(*src) { - continue; - } - let Some(dest) = candidates.iter().find(|dest| !merged_locals.contains(**dest)) - else { - continue; - }; - - // Replace `src` by `dest` everywhere. - merges.insert(*src, *dest); - merged_locals.insert(*src); - merged_locals.insert(*dest); - - // Update liveness information based on the merge we just performed. - // Every location where `src` was live, `dest` will be live. - live.union_rows(*src, *dest); + let mut relevant = RelevantLocals::compute(&candidates, body.local_decls.len()); + let mut live = save_as_intervals(&points, body, &relevant, live.results); + + dest_prop_mir_dump(tcx, body, &points, &live, &relevant); + + let mut merged_locals = DenseBitSet::new_empty(body.local_decls.len()); + + for (src, dst) in candidates.c.into_iter() { + trace!(?src, ?dst); + + let Some(mut src) = relevant.find(src) else { continue }; + let Some(mut dst) = relevant.find(dst) else { continue }; + if src == dst { + continue; } - trace!(merging = ?merges); - if merges.is_empty() { - break; + let Some(src_live_ranges) = live.row(src) else { continue }; + let Some(dst_live_ranges) = live.row(dst) else { continue }; + trace!(?src, ?src_live_ranges); + trace!(?dst, ?dst_live_ranges); + + if src_live_ranges.disjoint(dst_live_ranges) { + // We want to replace `src` by `dst`. + let mut orig_src = relevant.original[src]; + let mut orig_dst = relevant.original[dst]; + + // The return place and function arguments are required and cannot be renamed. + // This check cannot be made during candidate collection, as we may want to + // unify the same non-required local with several required locals. + match (is_local_required(orig_src, body), is_local_required(orig_dst, body)) { + // Renaming `src` is ok. + (false, _) => {} + // Renaming `src` is wrong, but renaming `dst` is ok. + (true, false) => { + std::mem::swap(&mut src, &mut dst); + std::mem::swap(&mut orig_src, &mut orig_dst); + } + // Neither local can be renamed, so skip this case. + (true, true) => continue, + } + + trace!(?src, ?dst, "merge"); + merged_locals.insert(orig_src); + merged_locals.insert(orig_dst); + + // Replace `src` by `dst`. + let head = relevant.union(src, dst); + live.union_rows(/* read */ src, /* write */ head); + live.union_rows(/* read */ dst, /* write */ head); } - round_count += 1; + } + trace!(?merged_locals); + trace!(?relevant.renames); - apply_merges(body, tcx, merges, merged_locals); + if merged_locals.is_empty() { + return; } - trace!(round_count); + apply_merges(body, tcx, relevant, merged_locals); } fn is_required(&self) -> bool { @@ -246,30 +246,6 @@ impl<'tcx> crate::MirPass<'tcx> for DestinationPropagation { } } -#[derive(Debug, Default)] -struct Candidates { - /// The set of candidates we are considering in this optimization. - /// - /// We will always merge the key into at most one of its values. - /// - /// Whether a place ends up in the key or the value does not correspond to whether it appears as - /// the lhs or rhs of any assignment. As a matter of fact, the places in here might never appear - /// in an assignment at all. This happens because if we see an assignment like this: - /// - /// ```ignore (syntax-highlighting-only) - /// _1.0 = _2.0 - /// ``` - /// - /// We will still report that we would like to merge `_1` and `_2` in an attempt to allow us to - /// remove that assignment. - c: FxIndexMap<Local, Vec<Local>>, - - /// A reverse index of the `c` set; if the `c` set contains `a => Place { local: b, proj }`, - /// then this contains `b => a`. - // PERF: Possibly these should be `SmallVec`s? - reverse: FxIndexMap<Local, Vec<Local>>, -} - ////////////////////////////////////////////////////////// // Merging // @@ -278,16 +254,16 @@ struct Candidates { fn apply_merges<'tcx>( body: &mut Body<'tcx>, tcx: TyCtxt<'tcx>, - merges: FxIndexMap<Local, Local>, + relevant: RelevantLocals, merged_locals: DenseBitSet<Local>, ) { - let mut merger = Merger { tcx, merges, merged_locals }; + let mut merger = Merger { tcx, relevant, merged_locals }; merger.visit_body_preserves_cfg(body); } struct Merger<'tcx> { tcx: TyCtxt<'tcx>, - merges: FxIndexMap<Local, Local>, + relevant: RelevantLocals, merged_locals: DenseBitSet<Local>, } @@ -297,8 +273,8 @@ impl<'tcx> MutVisitor<'tcx> for Merger<'tcx> { } fn visit_local(&mut self, local: &mut Local, _: PlaceContext, _location: Location) { - if let Some(dest) = self.merges.get(local) { - *local = *dest; + if let Some(relevant) = self.relevant.find(*local) { + *local = self.relevant.original[relevant]; } } @@ -336,414 +312,95 @@ impl<'tcx> MutVisitor<'tcx> for Merger<'tcx> { } ////////////////////////////////////////////////////////// -// Liveness filtering +// Relevant locals // -// This section enforces bullet point 2 +// Small utility to reduce size of the conflict matrix by only considering locals that appear in +// the candidates -struct FilterInformation<'a, 'tcx> { - body: &'a Body<'tcx>, - points: &'a DenseLocationMap, - live: &'a SparseIntervalMatrix<Local, PointIndex>, - candidates: &'a mut Candidates, - write_info: &'a mut WriteInfo, - at: Location, +newtype_index! { + /// Represent a subset of locals which appear in candidates. + struct RelevantLocal {} } -// We first implement some utility functions which we will expose removing candidates according to -// different needs. Throughout the liveness filtering, the `candidates` are only ever accessed -// through these methods, and not directly. -impl Candidates { - /// Collects the candidates for merging. - /// - /// This is responsible for enforcing the first and third bullet point. - fn reset_and_find<'tcx>(&mut self, body: &Body<'tcx>, borrowed: &DenseBitSet<Local>) { - self.c.clear(); - self.reverse.clear(); - let mut visitor = FindAssignments { body, candidates: &mut self.c, borrowed }; - visitor.visit_body(body); - // Deduplicate candidates. - for (_, cands) in self.c.iter_mut() { - cands.sort(); - cands.dedup(); - } - // Generate the reverse map. - for (src, cands) in self.c.iter() { - for dest in cands.iter().copied() { - self.reverse.entry(dest).or_default().push(*src); - } - } - } - - /// Just `Vec::retain`, but the condition is inverted and we add debugging output - fn vec_filter_candidates( - src: Local, - v: &mut Vec<Local>, - mut f: impl FnMut(Local) -> CandidateFilter, - at: Location, - ) { - v.retain(|dest| { - let remove = f(*dest); - if remove == CandidateFilter::Remove { - trace!("eliminating {:?} => {:?} due to conflict at {:?}", src, dest, at); - } - remove == CandidateFilter::Keep - }); - } - - /// `vec_filter_candidates` but for an `Entry` - fn entry_filter_candidates( - mut entry: IndexOccupiedEntry<'_, Local, Vec<Local>>, - p: Local, - f: impl FnMut(Local) -> CandidateFilter, - at: Location, - ) { - let candidates = entry.get_mut(); - Self::vec_filter_candidates(p, candidates, f, at); - if candidates.len() == 0 { - // FIXME(#120456) - is `swap_remove` correct? - entry.swap_remove(); - } - } - - /// For all candidates `(p, q)` or `(q, p)` removes the candidate if `f(q)` says to do so - fn filter_candidates_by( - &mut self, - p: Local, - mut f: impl FnMut(Local) -> CandidateFilter, - at: Location, - ) { - // Cover the cases where `p` appears as a `src` - if let IndexEntry::Occupied(entry) = self.c.entry(p) { - Self::entry_filter_candidates(entry, p, &mut f, at); - } - // And the cases where `p` appears as a `dest` - let Some(srcs) = self.reverse.get_mut(&p) else { - return; - }; - // We use `retain` here to remove the elements from the reverse set if we've removed the - // matching candidate in the forward set. - srcs.retain(|src| { - if f(*src) == CandidateFilter::Keep { - return true; - } - let IndexEntry::Occupied(entry) = self.c.entry(*src) else { - return false; - }; - Self::entry_filter_candidates( - entry, - *src, - |dest| { - if dest == p { CandidateFilter::Remove } else { CandidateFilter::Keep } - }, - at, - ); - false - }); - } +#[derive(Debug)] +struct RelevantLocals { + original: IndexVec<RelevantLocal, Local>, + shrink: IndexVec<Local, Option<RelevantLocal>>, + renames: UnionFind<RelevantLocal>, } -#[derive(Copy, Clone, PartialEq, Eq)] -enum CandidateFilter { - Keep, - Remove, -} +impl RelevantLocals { + #[tracing::instrument(level = "trace", skip(candidates, num_locals), ret)] + fn compute(candidates: &Candidates, num_locals: usize) -> RelevantLocals { + let mut original = IndexVec::with_capacity(candidates.c.len()); + let mut shrink = IndexVec::from_elem_n(None, num_locals); -impl<'a, 'tcx> FilterInformation<'a, 'tcx> { - /// Filters the set of candidates to remove those that conflict. - /// - /// The steps we take are exactly those that are outlined at the top of the file. For each - /// statement/terminator, we collect the set of locals that are written to in that - /// statement/terminator, and then we remove all pairs of candidates that contain one such local - /// and another one that is live. - /// - /// We need to be careful about the ordering of operations within each statement/terminator - /// here. Many statements might write and read from more than one place, and we need to consider - /// them all. The strategy for doing this is as follows: We first gather all the places that are - /// written to within the statement/terminator via `WriteInfo`. Then, we use the liveness - /// analysis from *before* the statement/terminator (in the control flow sense) to eliminate - /// candidates - this is because we want to conservatively treat a pair of locals that is both - /// read and written in the statement/terminator to be conflicting, and the liveness analysis - /// before the statement/terminator will correctly report locals that are read in the - /// statement/terminator to be live. We are additionally conservative by treating all written to - /// locals as also being read from. - fn filter_liveness( - candidates: &mut Candidates, - points: &DenseLocationMap, - live: &SparseIntervalMatrix<Local, PointIndex>, - write_info: &mut WriteInfo, - body: &Body<'tcx>, - ) { - let mut this = FilterInformation { - body, - points, - live, - candidates, - // We don't actually store anything at this scope, we just keep things here to be able - // to reuse the allocation. - write_info, - // Doesn't matter what we put here, will be overwritten before being used - at: Location::START, + // Mark a local as relevant and record it into the maps. + let mut declare = |local| { + shrink.get_or_insert_with(local, || original.push(local)); }; - this.internal_filter_liveness(); - } - fn internal_filter_liveness(&mut self) { - for (block, data) in traversal::preorder(self.body) { - self.at = Location { block, statement_index: data.statements.len() }; - self.write_info.for_terminator(&data.terminator().kind); - self.apply_conflicts(); - - for (i, statement) in data.statements.iter().enumerate().rev() { - self.at = Location { block, statement_index: i }; - self.write_info.for_statement(&statement.kind, self.body); - self.apply_conflicts(); - } + for &(src, dest) in candidates.c.iter() { + declare(src); + declare(dest) } - } - fn apply_conflicts(&mut self) { - let writes = &self.write_info.writes; - for p in writes { - let other_skip = self.write_info.skip_pair.and_then(|(a, b)| { - if a == *p { - Some(b) - } else if b == *p { - Some(a) - } else { - None - } - }); - let at = self.points.point_from_location(self.at); - self.candidates.filter_candidates_by( - *p, - |q| { - if Some(q) == other_skip { - return CandidateFilter::Keep; - } - // It is possible that a local may be live for less than the - // duration of a statement This happens in the case of function - // calls or inline asm. Because of this, we also mark locals as - // conflicting when both of them are written to in the same - // statement. - if self.live.contains(q, at) || writes.contains(&q) { - CandidateFilter::Remove - } else { - CandidateFilter::Keep - } - }, - self.at, - ); - } + let renames = UnionFind::new(original.len()); + RelevantLocals { original, shrink, renames } } -} -/// Describes where a statement/terminator writes to -#[derive(Default, Debug)] -struct WriteInfo { - writes: Vec<Local>, - /// If this pair of locals is a candidate pair, completely skip processing it during this - /// statement. All other candidates are unaffected. - skip_pair: Option<(Local, Local)>, -} - -impl WriteInfo { - fn for_statement<'tcx>(&mut self, statement: &StatementKind<'tcx>, body: &Body<'tcx>) { - self.reset(); - match statement { - StatementKind::Assign(box (lhs, rhs)) => { - self.add_place(*lhs); - match rhs { - Rvalue::Use(op) => { - self.add_operand(op); - self.consider_skipping_for_assign_use(*lhs, op, body); - } - Rvalue::Repeat(op, _) => { - self.add_operand(op); - } - Rvalue::Cast(_, op, _) - | Rvalue::UnaryOp(_, op) - | Rvalue::ShallowInitBox(op, _) => { - self.add_operand(op); - } - Rvalue::BinaryOp(_, ops) => { - for op in [&ops.0, &ops.1] { - self.add_operand(op); - } - } - Rvalue::Aggregate(_, ops) => { - for op in ops { - self.add_operand(op); - } - } - Rvalue::WrapUnsafeBinder(op, _) => { - self.add_operand(op); - } - Rvalue::ThreadLocalRef(_) - | Rvalue::NullaryOp(_, _) - | Rvalue::Ref(_, _, _) - | Rvalue::RawPtr(_, _) - | Rvalue::Len(_) - | Rvalue::Discriminant(_) - | Rvalue::CopyForDeref(_) => {} - } - } - // Retags are technically also reads, but reporting them as a write suffices - StatementKind::SetDiscriminant { place, .. } - | StatementKind::Deinit(place) - | StatementKind::Retag(_, place) => { - self.add_place(**place); - } - StatementKind::Intrinsic(_) - | StatementKind::ConstEvalCounter - | StatementKind::Nop - | StatementKind::Coverage(_) - | StatementKind::StorageLive(_) - | StatementKind::StorageDead(_) - | StatementKind::BackwardIncompatibleDropHint { .. } - | StatementKind::PlaceMention(_) => {} - StatementKind::FakeRead(_) | StatementKind::AscribeUserType(_, _) => { - bug!("{:?} not found in this MIR phase", statement) - } - } + fn find(&mut self, src: Local) -> Option<RelevantLocal> { + let src = self.shrink[src]?; + let src = self.renames.find(src); + Some(src) } - fn consider_skipping_for_assign_use<'tcx>( - &mut self, - lhs: Place<'tcx>, - rhs: &Operand<'tcx>, - body: &Body<'tcx>, - ) { - let Some(rhs) = rhs.place() else { return }; - if let Some(pair) = places_to_candidate_pair(lhs, rhs, body) { - self.skip_pair = Some(pair); - } - } - - fn for_terminator<'tcx>(&mut self, terminator: &TerminatorKind<'tcx>) { - self.reset(); - match terminator { - TerminatorKind::SwitchInt { discr: op, .. } - | TerminatorKind::Assert { cond: op, .. } => { - self.add_operand(op); - } - TerminatorKind::Call { destination, func, args, .. } => { - self.add_place(*destination); - self.add_operand(func); - for arg in args { - self.add_operand(&arg.node); - } - } - TerminatorKind::TailCall { func, args, .. } => { - self.add_operand(func); - for arg in args { - self.add_operand(&arg.node); - } - } - TerminatorKind::InlineAsm { operands, .. } => { - for asm_operand in operands { - match asm_operand { - InlineAsmOperand::In { value, .. } => { - self.add_operand(value); - } - InlineAsmOperand::Out { place, .. } => { - if let Some(place) = place { - self.add_place(*place); - } - } - // Note that the `late` field in `InOut` is about whether the registers used - // for these things overlap, and is of absolutely no interest to us. - InlineAsmOperand::InOut { in_value, out_place, .. } => { - if let Some(place) = out_place { - self.add_place(*place); - } - self.add_operand(in_value); - } - InlineAsmOperand::Const { .. } - | InlineAsmOperand::SymFn { .. } - | InlineAsmOperand::SymStatic { .. } - | InlineAsmOperand::Label { .. } => {} - } - } - } - TerminatorKind::Goto { .. } - | TerminatorKind::UnwindResume - | TerminatorKind::UnwindTerminate(_) - | TerminatorKind::Return - | TerminatorKind::Unreachable { .. } => (), - TerminatorKind::Drop { .. } => { - // `Drop`s create a `&mut` and so are not considered - } - TerminatorKind::Yield { .. } - | TerminatorKind::CoroutineDrop - | TerminatorKind::FalseEdge { .. } - | TerminatorKind::FalseUnwind { .. } => { - bug!("{:?} not found in this MIR phase", terminator) - } - } - } - - fn add_place(&mut self, place: Place<'_>) { - self.writes.push(place.local); - } - - fn add_operand<'tcx>(&mut self, op: &Operand<'tcx>) { - match op { - // FIXME(JakobDegen): In a previous version, the `Move` case was incorrectly treated as - // being a read only. This was unsound, however we cannot add a regression test because - // it is not possible to set this off with current MIR. Once we have that ability, a - // regression test should be added. - Operand::Move(p) => self.add_place(*p), - Operand::Copy(_) | Operand::Constant(_) => (), - } - } - - fn reset(&mut self) { - self.writes.clear(); - self.skip_pair = None; + fn union(&mut self, lhs: RelevantLocal, rhs: RelevantLocal) -> RelevantLocal { + let head = self.renames.unify(lhs, rhs); + // We need to ensure we keep the original local of the RHS, as it may be a required local. + self.original[head] = self.original[rhs]; + head } } ///////////////////////////////////////////////////// // Candidate accumulation -/// If the pair of places is being considered for merging, returns the candidate which would be -/// merged in order to accomplish this. -/// -/// The contract here is in one direction - there is a guarantee that merging the locals that are -/// outputted by this function would result in an assignment between the inputs becoming a -/// self-assignment. However, there is no guarantee that the returned pair is actually suitable for -/// merging - candidate collection must still check this independently. -/// -/// This output is unique for each unordered pair of input places. -fn places_to_candidate_pair<'tcx>( - a: Place<'tcx>, - b: Place<'tcx>, - body: &Body<'tcx>, -) -> Option<(Local, Local)> { - let (mut a, mut b) = if a.projection.len() == 0 && b.projection.len() == 0 { - (a.local, b.local) - } else { - return None; - }; +#[derive(Debug, Default)] +struct Candidates { + /// The set of candidates we are considering in this optimization. + /// + /// Whether a place ends up in the key or the value does not correspond to whether it appears as + /// the lhs or rhs of any assignment. As a matter of fact, the places in here might never appear + /// in an assignment at all. This happens because if we see an assignment like this: + /// + /// ```ignore (syntax-highlighting-only) + /// _1.0 = _2.0 + /// ``` + /// + /// We will still report that we would like to merge `_1` and `_2` in an attempt to allow us to + /// remove that assignment. + c: Vec<(Local, Local)>, +} - // By sorting, we make sure we're input order independent - if a > b { - std::mem::swap(&mut a, &mut b); - } +// We first implement some utility functions which we will expose removing candidates according to +// different needs. Throughout the liveness filtering, the `candidates` are only ever accessed +// through these methods, and not directly. +impl Candidates { + /// Collects the candidates for merging. + /// + /// This is responsible for enforcing the first and third bullet point. + fn find(body: &Body<'_>, borrowed: &DenseBitSet<Local>) -> Candidates { + let mut visitor = FindAssignments { body, candidates: Default::default(), borrowed }; + visitor.visit_body(body); - // We could now return `(a, b)`, but then we miss some candidates in the case where `a` can't be - // used as a `src`. - if is_local_required(a, body) { - std::mem::swap(&mut a, &mut b); + Candidates { c: visitor.candidates } } - // We could check `is_local_required` again here, but there's no need - after all, we make no - // promise that the candidate pair is actually valid - Some((a, b)) } struct FindAssignments<'a, 'tcx> { body: &'a Body<'tcx>, - candidates: &'a mut FxIndexMap<Local, Vec<Local>>, + candidates: Vec<(Local, Local)>, borrowed: &'a DenseBitSet<Local>, } @@ -753,11 +410,9 @@ impl<'tcx> Visitor<'tcx> for FindAssignments<'_, 'tcx> { lhs, Rvalue::CopyForDeref(rhs) | Rvalue::Use(Operand::Copy(rhs) | Operand::Move(rhs)), )) = &statement.kind + && let Some(src) = lhs.as_local() + && let Some(dest) = rhs.as_local() { - let Some((src, dest)) = places_to_candidate_pair(*lhs, *rhs, self.body) else { - return; - }; - // As described at the top of the file, we do not go near things that have // their address taken. if self.borrowed.contains(src) || self.borrowed.contains(dest) { @@ -774,13 +429,8 @@ impl<'tcx> Visitor<'tcx> for FindAssignments<'_, 'tcx> { return; } - // Also, we need to make sure that MIR actually allows the `src` to be removed - if is_local_required(src, self.body) { - return; - } - // We may insert duplicates here, but that's fine - self.candidates.entry(src).or_default().push(dest); + self.candidates.push((src, dest)); } } } @@ -803,22 +453,162 @@ fn dest_prop_mir_dump<'tcx>( tcx: TyCtxt<'tcx>, body: &Body<'tcx>, points: &DenseLocationMap, - live: &SparseIntervalMatrix<Local, PointIndex>, - round: usize, + live: &SparseIntervalMatrix<RelevantLocal, TwoStepIndex>, + relevant: &RelevantLocals, ) { let locals_live_at = |location| { - let location = points.point_from_location(location); - live.rows().filter(|&r| live.contains(r, location)).collect::<Vec<_>>() + live.rows() + .filter(|&r| live.contains(r, location)) + .map(|rl| relevant.original[rl]) + .collect::<Vec<_>>() }; if let Some(dumper) = MirDumper::new(tcx, "DestinationPropagation-dataflow", body) { let extra_data = &|pass_where, w: &mut dyn std::io::Write| { if let PassWhere::BeforeLocation(loc) = pass_where { - writeln!(w, " // live: {:?}", locals_live_at(loc))?; + let location = TwoStepIndex::new(points, loc, Effect::Before); + let live = locals_live_at(location); + writeln!(w, " // before: {:?} => {:?}", location, live)?; + } + if let PassWhere::AfterLocation(loc) = pass_where { + let location = TwoStepIndex::new(points, loc, Effect::After); + let live = locals_live_at(location); + writeln!(w, " // after: {:?} => {:?}", location, live)?; } Ok(()) }; - dumper.set_disambiguator(&round).set_extra_data(extra_data).dump_mir(body) + dumper.set_extra_data(extra_data).dump_mir(body) } } + +#[derive(Copy, Clone, Debug)] +enum Effect { + Before, + After, +} + +rustc_index::newtype_index! { + /// A reversed `PointIndex` but with the lower bit encoding early/late inside the statement. + /// The reversed order allows to use the more efficient `IntervalSet::append` method while we + /// iterate on the statements in reverse order. + #[orderable] + #[debug_format = "TwoStepIndex({})"] + struct TwoStepIndex {} +} + +impl TwoStepIndex { + fn new(elements: &DenseLocationMap, location: Location, effect: Effect) -> TwoStepIndex { + let point = elements.point_from_location(location); + let effect = match effect { + Effect::Before => 0, + Effect::After => 1, + }; + let max_index = 2 * elements.num_points() as u32 - 1; + let index = 2 * point.as_u32() + (effect as u32); + // Reverse the indexing to use more efficient `IntervalSet::append`. + TwoStepIndex::from_u32(max_index - index) + } +} + +struct VisitPlacesWith<F>(F); + +impl<'tcx, F> Visitor<'tcx> for VisitPlacesWith<F> +where + F: FnMut(Place<'tcx>, PlaceContext), +{ + fn visit_local(&mut self, local: Local, ctxt: PlaceContext, _: Location) { + (self.0)(local.into(), ctxt); + } + + fn visit_place(&mut self, place: &Place<'tcx>, ctxt: PlaceContext, location: Location) { + (self.0)(*place, ctxt); + self.visit_projection(place.as_ref(), ctxt, location); + } +} + +/// Add points depending on the result of the given dataflow analysis. +fn save_as_intervals<'tcx>( + elements: &DenseLocationMap, + body: &Body<'tcx>, + relevant: &RelevantLocals, + results: Results<DenseBitSet<Local>>, +) -> SparseIntervalMatrix<RelevantLocal, TwoStepIndex> { + let mut values = SparseIntervalMatrix::new(2 * elements.num_points()); + let mut state = MaybeLiveLocals.bottom_value(body); + let reachable_blocks = traversal::reachable_as_bitset(body); + + let two_step_loc = |location, effect| TwoStepIndex::new(elements, location, effect); + let append_at = + |values: &mut SparseIntervalMatrix<_, _>, state: &DenseBitSet<Local>, twostep| { + for (relevant, &original) in relevant.original.iter_enumerated() { + if state.contains(original) { + values.append(relevant, twostep); + } + } + }; + + // Iterate blocks in decreasing order, to visit locations in decreasing order. This + // allows to use the more efficient `append` method to interval sets. + for block in body.basic_blocks.indices().rev() { + if !reachable_blocks.contains(block) { + continue; + } + + state.clone_from(&results[block]); + + let block_data = &body.basic_blocks[block]; + let loc = Location { block, statement_index: block_data.statements.len() }; + + let term = block_data.terminator(); + let mut twostep = two_step_loc(loc, Effect::After); + append_at(&mut values, &state, twostep); + // Ensure we have a non-zero live range even for dead stores. This is done by marking all + // the written-to locals as live in the second half of the statement. + // We also ensure that operands read by terminators conflict with writes by that terminator. + // For instance a function call may read args after having written to the destination. + VisitPlacesWith(|place, ctxt| match DefUse::for_place(place, ctxt) { + DefUse::Def | DefUse::Use | DefUse::PartialWrite => { + if let Some(relevant) = relevant.shrink[place.local] { + values.insert(relevant, twostep); + } + } + DefUse::NonUse => {} + }) + .visit_terminator(term, loc); + + twostep = TwoStepIndex::from_u32(twostep.as_u32() + 1); + debug_assert_eq!(twostep, two_step_loc(loc, Effect::Before)); + MaybeLiveLocals.apply_early_terminator_effect(&mut state, term, loc); + MaybeLiveLocals.apply_primary_terminator_effect(&mut state, term, loc); + append_at(&mut values, &state, twostep); + + for (statement_index, stmt) in block_data.statements.iter().enumerate().rev() { + let loc = Location { block, statement_index }; + twostep = TwoStepIndex::from_u32(twostep.as_u32() + 1); + debug_assert_eq!(twostep, two_step_loc(loc, Effect::After)); + append_at(&mut values, &state, twostep); + // Ensure we have a non-zero live range even for dead stores. This is done by marking + // all the written-to locals as live in the second half of the statement. + VisitPlacesWith(|place, ctxt| match DefUse::for_place(place, ctxt) { + DefUse::Def | DefUse::PartialWrite => { + if let Some(relevant) = relevant.shrink[place.local] { + values.insert(relevant, twostep); + } + } + DefUse::Use | DefUse::NonUse => {} + }) + .visit_statement(stmt, loc); + + twostep = TwoStepIndex::from_u32(twostep.as_u32() + 1); + debug_assert_eq!(twostep, two_step_loc(loc, Effect::Before)); + MaybeLiveLocals.apply_early_statement_effect(&mut state, stmt, loc); + MaybeLiveLocals.apply_primary_statement_effect(&mut state, stmt, loc); + // ... but reads from operands are marked as live here so they do not conflict with + // the all the writes we manually marked as live in the second half of the statement. + append_at(&mut values, &state, twostep); + } + } + + values +} diff --git a/compiler/rustc_mir_transform/src/errors.rs b/compiler/rustc_mir_transform/src/errors.rs index ad9635aae33..775f5f9a7cf 100644 --- a/compiler/rustc_mir_transform/src/errors.rs +++ b/compiler/rustc_mir_transform/src/errors.rs @@ -2,6 +2,7 @@ use rustc_errors::codes::*; use rustc_errors::{Diag, LintDiagnostic}; use rustc_macros::{Diagnostic, LintDiagnostic, Subdiagnostic}; use rustc_middle::mir::AssertKind; +use rustc_middle::query::Key; use rustc_middle::ty::TyCtxt; use rustc_session::lint::{self, Lint}; use rustc_span::def_id::DefId; @@ -9,6 +10,46 @@ use rustc_span::{Ident, Span, Symbol}; use crate::fluent_generated as fluent; +/// Emit diagnostic for calls to `#[inline(always)]`-annotated functions with a +/// `#[target_feature]` attribute where the caller enables a different set of target features. +pub(crate) fn emit_inline_always_target_feature_diagnostic<'a, 'tcx>( + tcx: TyCtxt<'tcx>, + call_span: Span, + callee_def_id: DefId, + caller_def_id: DefId, + callee_only: &[&'a str], +) { + let callee = tcx.def_path_str(callee_def_id); + let caller = tcx.def_path_str(caller_def_id); + + tcx.node_span_lint( + lint::builtin::INLINE_ALWAYS_MISMATCHING_TARGET_FEATURES, + tcx.local_def_id_to_hir_id(caller_def_id.as_local().unwrap()), + call_span, + |lint| { + lint.primary_message(format!( + "call to `#[inline(always)]`-annotated `{callee}` \ + requires the same target features to be inlined" + )); + lint.note("function will not be inlined"); + + lint.note(format!( + "the following target features are on `{callee}` but missing from `{caller}`: {}", + callee_only.join(", ") + )); + lint.span_note(callee_def_id.default_span(tcx), format!("`{callee}` is defined here")); + + let feats = callee_only.join(","); + lint.span_suggestion( + tcx.def_span(caller_def_id).shrink_to_lo(), + format!("add `#[target_feature]` attribute to `{caller}`"), + format!("#[target_feature(enable = \"{feats}\")]\n"), + lint::Applicability::MaybeIncorrect, + ); + }, + ); +} + #[derive(LintDiagnostic)] #[diag(mir_transform_unconditional_recursion)] #[help] diff --git a/compiler/rustc_mir_transform/src/gvn.rs b/compiler/rustc_mir_transform/src/gvn.rs index 5a13394543b..f867c130efb 100644 --- a/compiler/rustc_mir_transform/src/gvn.rs +++ b/compiler/rustc_mir_transform/src/gvn.rs @@ -87,6 +87,7 @@ use std::borrow::Cow; use either::Either; +use itertools::Itertools as _; use rustc_abi::{self as abi, BackendRepr, FIRST_VARIANT, FieldIdx, Primitive, Size, VariantIdx}; use rustc_const_eval::const_eval::DummyMachine; use rustc_const_eval::interpret::{ @@ -895,18 +896,13 @@ impl<'body, 'tcx> VnState<'body, 'tcx> { fn simplify_aggregate_to_copy( &mut self, - lhs: &Place<'tcx>, - rvalue: &mut Rvalue<'tcx>, - location: Location, - fields: &[VnIndex], + ty: Ty<'tcx>, variant_index: VariantIdx, + fields: &[VnIndex], ) -> Option<VnIndex> { - let Some(&first_field) = fields.first() else { - return None; - }; - let Value::Projection(copy_from_value, _) = *self.get(first_field) else { - return None; - }; + let Some(&first_field) = fields.first() else { return None }; + let Value::Projection(copy_from_value, _) = *self.get(first_field) else { return None }; + // All fields must correspond one-to-one and come from the same aggregate value. if fields.iter().enumerate().any(|(index, &v)| { if let Value::Projection(pointer, ProjectionElem::Field(from_index, _)) = *self.get(v) @@ -933,21 +929,8 @@ impl<'body, 'tcx> VnState<'body, 'tcx> { } } - // Allow introducing places with non-constant offsets, as those are still better than - // reconstructing an aggregate. - if self.ty(copy_from_local_value) == rvalue.ty(self.local_decls, self.tcx) - && let Some(place) = self.try_as_place(copy_from_local_value, location, true) - { - // Avoid creating `*a = copy (*b)`, as they might be aliases resulting in overlapping assignments. - // FIXME: This also avoids any kind of projection, not just derefs. We can add allowed projections. - if lhs.as_local().is_some() { - self.reused_locals.insert(place.local); - *rvalue = Rvalue::Use(Operand::Copy(place)); - } - return Some(copy_from_local_value); - } - - None + // Both must be variants of the same type. + if self.ty(copy_from_local_value) == ty { Some(copy_from_local_value) } else { None } } fn simplify_aggregate( @@ -1023,20 +1006,27 @@ impl<'body, 'tcx> VnState<'body, 'tcx> { } }; - if ty.is_array() && fields.len() > 4 { - let first = fields[0]; - if fields.iter().all(|&v| v == first) { - let len = ty::Const::from_target_usize(self.tcx, fields.len().try_into().unwrap()); - if let Some(op) = self.try_as_operand(first, location) { - *rvalue = Rvalue::Repeat(op, len); - } - return Some(self.insert(ty, Value::Repeat(first, len))); + if ty.is_array() + && fields.len() > 4 + && let Ok(&first) = fields.iter().all_equal_value() + { + let len = ty::Const::from_target_usize(self.tcx, fields.len().try_into().unwrap()); + if let Some(op) = self.try_as_operand(first, location) { + *rvalue = Rvalue::Repeat(op, len); } + return Some(self.insert(ty, Value::Repeat(first, len))); } - if let Some(value) = - self.simplify_aggregate_to_copy(lhs, rvalue, location, &fields, variant_index) - { + if let Some(value) = self.simplify_aggregate_to_copy(ty, variant_index, &fields) { + // Allow introducing places with non-constant offsets, as those are still better than + // reconstructing an aggregate. But avoid creating `*a = copy (*b)`, as they might be + // aliases resulting in overlapping assignments. + let allow_complex_projection = + lhs.projection[..].iter().all(PlaceElem::is_stable_offset); + if let Some(place) = self.try_as_place(value, location, allow_complex_projection) { + self.reused_locals.insert(place.local); + *rvalue = Rvalue::Use(Operand::Copy(place)); + } return Some(value); } diff --git a/compiler/rustc_mir_transform/src/lib.rs b/compiler/rustc_mir_transform/src/lib.rs index 8f319e64916..1663dfa744f 100644 --- a/compiler/rustc_mir_transform/src/lib.rs +++ b/compiler/rustc_mir_transform/src/lib.rs @@ -117,6 +117,7 @@ declare_passes! { mod add_subtyping_projections : Subtyper; mod check_inline : CheckForceInline; mod check_call_recursion : CheckCallRecursion, CheckDropRecursion; + mod check_inline_always_target_features: CheckInlineAlwaysTargetFeature; mod check_alignment : CheckAlignment; mod check_enums : CheckEnums; mod check_const_item_mutation : CheckConstItemMutation; @@ -384,6 +385,9 @@ fn mir_built(tcx: TyCtxt<'_>, def: LocalDefId) -> &Steal<Body<'_>> { // MIR-level lints. &Lint(check_inline::CheckForceInline), &Lint(check_call_recursion::CheckCallRecursion), + // Check callee's target features match callers target features when + // using `#[inline(always)]` + &Lint(check_inline_always_target_features::CheckInlineAlwaysTargetFeature), &Lint(check_packed_ref::CheckPackedRef), &Lint(check_const_item_mutation::CheckConstItemMutation), &Lint(function_item_references::FunctionItemReferences), diff --git a/compiler/rustc_mir_transform/src/lint_tail_expr_drop_order.rs b/compiler/rustc_mir_transform/src/lint_tail_expr_drop_order.rs index 794984d2f3e..61c9bbe3123 100644 --- a/compiler/rustc_mir_transform/src/lint_tail_expr_drop_order.rs +++ b/compiler/rustc_mir_transform/src/lint_tail_expr_drop_order.rs @@ -2,6 +2,7 @@ use std::cell::RefCell; use std::collections::hash_map; use std::rc::Rc; +use itertools::Itertools as _; use rustc_data_structures::fx::{FxHashMap, FxHashSet, FxIndexMap}; use rustc_data_structures::unord::{UnordMap, UnordSet}; use rustc_errors::Subdiagnostic; @@ -339,9 +340,9 @@ pub(crate) fn run_lint<'tcx>(tcx: TyCtxt<'tcx>, def_id: LocalDefId, body: &Body< // Suppose that all BIDs point into the same local, // we can remove the this local from the observed drops, // so that we can focus our diagnosis more on the others. - if candidates.iter().all(|&(_, place)| candidates[0].1.local == place.local) { + if let Ok(local) = candidates.iter().map(|&(_, place)| place.local).all_equal_value() { for path_idx in all_locals_dropped.iter() { - if move_data.move_paths[path_idx].place.local == candidates[0].1.local { + if move_data.move_paths[path_idx].place.local == local { to_exclude.insert(path_idx); } } diff --git a/compiler/rustc_mir_transform/src/simplify.rs b/compiler/rustc_mir_transform/src/simplify.rs index 468ef742dfb..75917d23883 100644 --- a/compiler/rustc_mir_transform/src/simplify.rs +++ b/compiler/rustc_mir_transform/src/simplify.rs @@ -34,6 +34,7 @@ //! The normal logic that a program with UB can be changed to do anything does not apply to //! pre-"runtime" MIR! +use itertools::Itertools as _; use rustc_index::{Idx, IndexSlice, IndexVec}; use rustc_middle::mir::visit::{MutVisitor, MutatingUseContext, PlaceContext, Visitor}; use rustc_middle::mir::*; @@ -288,20 +289,13 @@ impl<'a, 'tcx> CfgSimplifier<'a, 'tcx> { return false; }; - let first_succ = { - if let Some(first_succ) = terminator.successors().next() { - if terminator.successors().all(|s| s == first_succ) { - let count = terminator.successors().count(); - self.pred_count[first_succ] -= (count - 1) as u32; - first_succ - } else { - return false; - } - } else { - return false; - } + let Ok(first_succ) = terminator.successors().all_equal_value() else { + return false; }; + let count = terminator.successors().count(); + self.pred_count[first_succ] -= (count - 1) as u32; + debug!("simplifying branch {:?}", terminator); terminator.kind = TerminatorKind::Goto { target: first_succ }; true diff --git a/compiler/rustc_mir_transform/src/ssa.rs b/compiler/rustc_mir_transform/src/ssa.rs index cd9a7f4a39d..73c249a3c8c 100644 --- a/compiler/rustc_mir_transform/src/ssa.rs +++ b/compiler/rustc_mir_transform/src/ssa.rs @@ -225,6 +225,9 @@ impl SsaVisitor<'_, '_> { impl<'tcx> Visitor<'tcx> for SsaVisitor<'_, 'tcx> { fn visit_local(&mut self, local: Local, ctxt: PlaceContext, loc: Location) { + if ctxt.may_observe_address() { + self.borrowed_locals.insert(local); + } match ctxt { PlaceContext::MutatingUse(MutatingUseContext::Projection) | PlaceContext::NonMutatingUse(NonMutatingUseContext::Projection) => bug!(), @@ -237,7 +240,6 @@ impl<'tcx> Visitor<'tcx> for SsaVisitor<'_, 'tcx> { PlaceContext::NonMutatingUse( NonMutatingUseContext::SharedBorrow | NonMutatingUseContext::FakeBorrow, ) => { - self.borrowed_locals.insert(local); self.check_dominates(local, loc); self.direct_uses[local] += 1; } diff --git a/compiler/rustc_next_trait_solver/src/solve/eval_ctxt/mod.rs b/compiler/rustc_next_trait_solver/src/solve/eval_ctxt/mod.rs index 31106a74527..443aebbdb4d 100644 --- a/compiler/rustc_next_trait_solver/src/solve/eval_ctxt/mod.rs +++ b/compiler/rustc_next_trait_solver/src/solve/eval_ctxt/mod.rs @@ -1275,7 +1275,7 @@ pub fn evaluate_root_goal_for_proof_tree_raw_provider< >( cx: I, canonical_goal: CanonicalInput<I>, -) -> (QueryResult<I>, I::ProbeRef) { +) -> (QueryResult<I>, I::Probe) { let mut inspect = inspect::ProofTreeBuilder::new(); let canonical_result = SearchGraph::<D>::evaluate_root_goal_for_proof_tree( cx, @@ -1284,7 +1284,7 @@ pub fn evaluate_root_goal_for_proof_tree_raw_provider< &mut inspect, ); let final_revision = inspect.unwrap(); - (canonical_result, cx.mk_probe_ref(final_revision)) + (canonical_result, cx.mk_probe(final_revision)) } /// Evaluate a goal to build a proof tree. diff --git a/compiler/rustc_parse/messages.ftl b/compiler/rustc_parse/messages.ftl index 4ca2f57bd87..77dd313d9b8 100644 --- a/compiler/rustc_parse/messages.ftl +++ b/compiler/rustc_parse/messages.ftl @@ -58,7 +58,7 @@ parse_async_use_order_incorrect = the order of `use` and `async` is incorrect parse_at_dot_dot_in_struct_pattern = `@ ..` is not supported in struct patterns .suggestion = bind to each field separately or, if you don't need them, just remove `{$ident} @` -parse_at_in_struct_pattern = Unexpected `@` in struct pattern +parse_at_in_struct_pattern = unexpected `@` in struct pattern .note = struct patterns use `field: pattern` syntax to bind to fields .help = consider replacing `new_name @ field_name` with `field_name: new_name` if that is what you intended diff --git a/compiler/rustc_parse/src/lexer/mod.rs b/compiler/rustc_parse/src/lexer/mod.rs index 9792240a548..f5f081efc49 100644 --- a/compiler/rustc_parse/src/lexer/mod.rs +++ b/compiler/rustc_parse/src/lexer/mod.rs @@ -6,7 +6,7 @@ use rustc_ast::util::unicode::{TEXT_FLOW_CONTROL_CHARS, contains_text_flow_contr use rustc_errors::codes::*; use rustc_errors::{Applicability, Diag, DiagCtxtHandle, StashKey}; use rustc_lexer::{ - Base, Cursor, DocStyle, FrontmatterAllowed, LiteralKind, RawStrError, is_whitespace, + Base, Cursor, DocStyle, FrontmatterAllowed, LiteralKind, RawStrError, is_horizontal_whitespace, }; use rustc_literal_escaper::{EscapeError, Mode, check_for_errors}; use rustc_session::lint::BuiltinLintDiag; @@ -44,19 +44,44 @@ pub(crate) struct UnmatchedDelim { pub candidate_span: Option<Span>, } +/// Which tokens should be stripped before lexing the tokens. +pub(crate) enum StripTokens { + /// Strip both shebang and frontmatter. + ShebangAndFrontmatter, + /// Strip the shebang but not frontmatter. + /// + /// That means that char sequences looking like frontmatter are simply + /// interpreted as regular Rust lexemes. + Shebang, + /// Strip nothing. + /// + /// In other words, char sequences looking like a shebang or frontmatter + /// are simply interpreted as regular Rust lexemes. + Nothing, +} + pub(crate) fn lex_token_trees<'psess, 'src>( psess: &'psess ParseSess, mut src: &'src str, mut start_pos: BytePos, override_span: Option<Span>, - frontmatter_allowed: FrontmatterAllowed, + strip_tokens: StripTokens, ) -> Result<TokenStream, Vec<Diag<'psess>>> { - // Skip `#!`, if present. - if let Some(shebang_len) = rustc_lexer::strip_shebang(src) { - src = &src[shebang_len..]; - start_pos = start_pos + BytePos::from_usize(shebang_len); + match strip_tokens { + StripTokens::Shebang | StripTokens::ShebangAndFrontmatter => { + if let Some(shebang_len) = rustc_lexer::strip_shebang(src) { + src = &src[shebang_len..]; + start_pos = start_pos + BytePos::from_usize(shebang_len); + } + } + StripTokens::Nothing => {} } + let frontmatter_allowed = match strip_tokens { + StripTokens::ShebangAndFrontmatter => FrontmatterAllowed::Yes, + StripTokens::Shebang | StripTokens::Nothing => FrontmatterAllowed::No, + }; + let cursor = Cursor::new(src, frontmatter_allowed); let mut lexer = Lexer { psess, @@ -597,7 +622,7 @@ impl<'psess, 'src> Lexer<'psess, 'src> { let last_line_start = within.rfind('\n').map_or(0, |i| i + 1); let last_line = &within[last_line_start..]; - let last_line_trimmed = last_line.trim_start_matches(is_whitespace); + let last_line_trimmed = last_line.trim_start_matches(is_horizontal_whitespace); let last_line_start_pos = frontmatter_opening_end_pos + BytePos(last_line_start as u32); let frontmatter_span = self.mk_sp(frontmatter_opening_pos, self.pos); @@ -640,7 +665,7 @@ impl<'psess, 'src> Lexer<'psess, 'src> { }); } - if !rest.trim_matches(is_whitespace).is_empty() { + if !rest.trim_matches(is_horizontal_whitespace).is_empty() { let span = self.mk_sp(last_line_start_pos, self.pos); self.dcx().emit_err(errors::FrontmatterExtraCharactersAfterClose { span }); } diff --git a/compiler/rustc_parse/src/lib.rs b/compiler/rustc_parse/src/lib.rs index 197333d942d..d8792d7af4c 100644 --- a/compiler/rustc_parse/src/lib.rs +++ b/compiler/rustc_parse/src/lib.rs @@ -21,7 +21,6 @@ use rustc_ast::tokenstream::{DelimSpan, TokenStream}; use rustc_ast::{AttrItem, Attribute, MetaItemInner, token}; use rustc_ast_pretty::pprust; use rustc_errors::{Diag, EmissionGuarantee, FatalError, PResult, pluralize}; -use rustc_lexer::FrontmatterAllowed; use rustc_session::parse::ParseSess; use rustc_span::source_map::SourceMap; use rustc_span::{FileName, SourceFile, Span}; @@ -34,6 +33,8 @@ pub mod parser; use parser::Parser; use rustc_ast::token::Delimiter; +use crate::lexer::StripTokens; + pub mod lexer; mod errors; @@ -62,7 +63,20 @@ pub fn new_parser_from_source_str( source: String, ) -> Result<Parser<'_>, Vec<Diag<'_>>> { let source_file = psess.source_map().new_source_file(name, source); - new_parser_from_source_file(psess, source_file) + new_parser_from_source_file(psess, source_file, StripTokens::ShebangAndFrontmatter) +} + +/// Creates a new parser from a simple (no shebang, no frontmatter) source string. +/// +/// On failure, the errors must be consumed via `unwrap_or_emit_fatal`, `emit`, `cancel`, +/// etc., otherwise a panic will occur when they are dropped. +pub fn new_parser_from_simple_source_str( + psess: &ParseSess, + name: FileName, + source: String, +) -> Result<Parser<'_>, Vec<Diag<'_>>> { + let source_file = psess.source_map().new_source_file(name, source); + new_parser_from_source_file(psess, source_file, StripTokens::Nothing) } /// Creates a new parser from a filename. On failure, the errors must be consumed via @@ -96,7 +110,7 @@ pub fn new_parser_from_file<'a>( } err.emit(); }); - new_parser_from_source_file(psess, source_file) + new_parser_from_source_file(psess, source_file, StripTokens::ShebangAndFrontmatter) } pub fn utf8_error<E: EmissionGuarantee>( @@ -147,9 +161,10 @@ pub fn utf8_error<E: EmissionGuarantee>( fn new_parser_from_source_file( psess: &ParseSess, source_file: Arc<SourceFile>, + strip_tokens: StripTokens, ) -> Result<Parser<'_>, Vec<Diag<'_>>> { let end_pos = source_file.end_position(); - let stream = source_file_to_stream(psess, source_file, None, FrontmatterAllowed::Yes)?; + let stream = source_file_to_stream(psess, source_file, None, strip_tokens)?; let mut parser = Parser::new(psess, stream, None); if parser.token == token::Eof { parser.token.span = Span::new(end_pos, end_pos, parser.token.span.ctxt(), None); @@ -165,8 +180,8 @@ pub fn source_str_to_stream( ) -> Result<TokenStream, Vec<Diag<'_>>> { let source_file = psess.source_map().new_source_file(name, source); // used mainly for `proc_macro` and the likes, not for our parsing purposes, so don't parse - // frontmatters as frontmatters. - source_file_to_stream(psess, source_file, override_span, FrontmatterAllowed::No) + // frontmatters as frontmatters, but for compatibility reason still strip the shebang + source_file_to_stream(psess, source_file, override_span, StripTokens::Shebang) } /// Given a source file, produces a sequence of token trees. Returns any buffered errors from @@ -175,7 +190,7 @@ fn source_file_to_stream<'psess>( psess: &'psess ParseSess, source_file: Arc<SourceFile>, override_span: Option<Span>, - frontmatter_allowed: FrontmatterAllowed, + strip_tokens: StripTokens, ) -> Result<TokenStream, Vec<Diag<'psess>>> { let src = source_file.src.as_ref().unwrap_or_else(|| { psess.dcx().bug(format!( @@ -184,13 +199,7 @@ fn source_file_to_stream<'psess>( )); }); - lexer::lex_token_trees( - psess, - src.as_str(), - source_file.start_pos, - override_span, - frontmatter_allowed, - ) + lexer::lex_token_trees(psess, src.as_str(), source_file.start_pos, override_span, strip_tokens) } /// Runs the given subparser `f` on the tokens of the given `attr`'s item. diff --git a/compiler/rustc_parse/src/parser/expr.rs b/compiler/rustc_parse/src/parser/expr.rs index 9b09cbba7af..4c02547357e 100644 --- a/compiler/rustc_parse/src/parser/expr.rs +++ b/compiler/rustc_parse/src/parser/expr.rs @@ -2742,7 +2742,7 @@ impl<'a> Parser<'a> { /// The specified `edition` in `let_chains_policy` should be that of the whole `if` construct, /// i.e. the same span we use to later decide whether the drop behaviour should be that of /// edition `..=2021` or that of `2024..`. - // Public because it is used in rustfmt forks such as https://github.com/tucant/rustfmt/blob/30c83df9e1db10007bdd16dafce8a86b404329b2/src/parse/macros/html.rs#L57 for custom if expressions. + // Public to use it for custom `if` expressions in rustfmt forks like https://github.com/tucant/rustfmt pub fn parse_expr_cond( &mut self, let_chains_policy: LetChainsPolicy, diff --git a/compiler/rustc_parse/src/parser/mod.rs b/compiler/rustc_parse/src/parser/mod.rs index 15598f32429..3bbca622975 100644 --- a/compiler/rustc_parse/src/parser/mod.rs +++ b/compiler/rustc_parse/src/parser/mod.rs @@ -22,6 +22,8 @@ use std::{fmt, mem, slice}; use attr_wrapper::{AttrWrapper, UsePreAttrPos}; pub use diagnostics::AttemptLocalParseRecovery; pub(crate) use expr::ForbiddenLetReason; +// Public to use it for custom `if` expressions in rustfmt forks like https://github.com/tucant/rustfmt +pub use expr::LetChainsPolicy; pub(crate) use item::{FnContext, FnParseMode}; pub use pat::{CommaRecoveryMode, RecoverColon, RecoverComma}; pub use path::PathStyle; diff --git a/compiler/rustc_passes/messages.ftl b/compiler/rustc_passes/messages.ftl index 498d9a405fa..afd08319738 100644 --- a/compiler/rustc_passes/messages.ftl +++ b/compiler/rustc_passes/messages.ftl @@ -687,6 +687,7 @@ passes_unused_var_maybe_capture_ref = unused variable: `{$name}` passes_unused_var_remove_field = unused variable: `{$name}` passes_unused_var_remove_field_suggestion = try removing the field +passes_unused_var_typo = you might have meant to pattern match on the similarly named {$kind} `{$item_name}` passes_unused_variable_args_in_macro = `{$name}` is captured in macro and introduced a unused variable diff --git a/compiler/rustc_passes/src/errors.rs b/compiler/rustc_passes/src/errors.rs index 680e2a26d84..23dcabef1a1 100644 --- a/compiler/rustc_passes/src/errors.rs +++ b/compiler/rustc_passes/src/errors.rs @@ -1367,6 +1367,22 @@ pub(crate) struct UnusedVarRemoveFieldSugg { #[note] pub(crate) struct UnusedVarAssignedOnly { pub name: String, + #[subdiagnostic] + pub typo: Option<PatternTypo>, +} + +#[derive(Subdiagnostic)] +#[multipart_suggestion( + passes_unused_var_typo, + style = "verbose", + applicability = "machine-applicable" +)] +pub(crate) struct PatternTypo { + #[suggestion_part(code = "{code}")] + pub span: Span, + pub code: String, + pub item_name: String, + pub kind: String, } #[derive(LintDiagnostic)] @@ -1434,6 +1450,8 @@ pub(crate) struct UnusedVariableTryPrefix { #[subdiagnostic] pub sugg: UnusedVariableSugg, pub name: String, + #[subdiagnostic] + pub typo: Option<PatternTypo>, } #[derive(Subdiagnostic)] diff --git a/compiler/rustc_passes/src/liveness.rs b/compiler/rustc_passes/src/liveness.rs index ae16a51bc69..1b2ffb5b3db 100644 --- a/compiler/rustc_passes/src/liveness.rs +++ b/compiler/rustc_passes/src/liveness.rs @@ -95,8 +95,10 @@ use rustc_hir::{Expr, HirId, HirIdMap, HirIdSet, find_attr}; use rustc_index::IndexVec; use rustc_middle::query::Providers; use rustc_middle::span_bug; +use rustc_middle::ty::print::with_no_trimmed_paths; use rustc_middle::ty::{self, RootVariableMinCaptureList, Ty, TyCtxt}; use rustc_session::lint; +use rustc_span::edit_distance::find_best_match_for_name; use rustc_span::{BytePos, Span, Symbol}; use tracing::{debug, instrument}; @@ -1688,6 +1690,51 @@ impl<'tcx> Liveness<'_, 'tcx> { let is_assigned = if ln == self.exit_ln { false } else { self.assigned_on_exit(ln, var) }; + let mut typo = None; + for (hir_id, _, span) in &hir_ids_and_spans { + let ty = self.typeck_results.node_type(*hir_id); + if let ty::Adt(adt, _) = ty.peel_refs().kind() { + let name = Symbol::intern(&name); + let adt_def = self.ir.tcx.adt_def(adt.did()); + let variant_names: Vec<_> = adt_def + .variants() + .iter() + .filter(|v| matches!(v.ctor, Some((CtorKind::Const, _)))) + .map(|v| v.name) + .collect(); + if let Some(name) = find_best_match_for_name(&variant_names, name, None) + && let Some(variant) = adt_def.variants().iter().find(|v| { + v.name == name && matches!(v.ctor, Some((CtorKind::Const, _))) + }) + { + typo = Some(errors::PatternTypo { + span: *span, + code: with_no_trimmed_paths!(self.ir.tcx.def_path_str(variant.def_id)), + kind: self.ir.tcx.def_descr(variant.def_id).to_string(), + item_name: variant.name.to_string(), + }); + } + } + } + if typo.is_none() { + for (hir_id, _, span) in &hir_ids_and_spans { + let ty = self.typeck_results.node_type(*hir_id); + // Look for consts of the same type with similar names as well, not just unit + // structs and variants. + for def_id in self.ir.tcx.hir_body_owners() { + if let DefKind::Const = self.ir.tcx.def_kind(def_id) + && self.ir.tcx.type_of(def_id).instantiate_identity() == ty + { + typo = Some(errors::PatternTypo { + span: *span, + code: with_no_trimmed_paths!(self.ir.tcx.def_path_str(def_id)), + kind: "constant".to_string(), + item_name: self.ir.tcx.item_name(def_id).to_string(), + }); + } + } + } + } if is_assigned { self.ir.tcx.emit_node_span_lint( lint::builtin::UNUSED_VARIABLES, @@ -1696,7 +1743,7 @@ impl<'tcx> Liveness<'_, 'tcx> { .into_iter() .map(|(_, _, ident_span)| ident_span) .collect::<Vec<_>>(), - errors::UnusedVarAssignedOnly { name }, + errors::UnusedVarAssignedOnly { name, typo }, ) } else if can_remove { let spans = hir_ids_and_spans @@ -1788,6 +1835,7 @@ impl<'tcx> Liveness<'_, 'tcx> { name, sugg, string_interp: suggestions, + typo, }, ); } diff --git a/compiler/rustc_query_impl/src/lib.rs b/compiler/rustc_query_impl/src/lib.rs index 306d4dbc4b4..e499e08c82b 100644 --- a/compiler/rustc_query_impl/src/lib.rs +++ b/compiler/rustc_query_impl/src/lib.rs @@ -58,7 +58,7 @@ impl<'tcx, C: QueryCache, const ANON: bool, const DEPTH_LIMIT: bool, const FEEDA for DynamicConfig<'tcx, C, ANON, DEPTH_LIMIT, FEEDABLE> { fn clone(&self) -> Self { - DynamicConfig { dynamic: self.dynamic } + *self } } diff --git a/compiler/rustc_query_system/messages.ftl b/compiler/rustc_query_system/messages.ftl index f48dc60afa0..f686608034c 100644 --- a/compiler/rustc_query_system/messages.ftl +++ b/compiler/rustc_query_system/messages.ftl @@ -18,8 +18,8 @@ query_system_cycle_usage = cycle used when {$usage} query_system_increment_compilation = internal compiler error: encountered incremental compilation error with {$dep_node} .help = This is a known issue with the compiler. Run {$run_cmd} to allow your project to compile -query_system_increment_compilation_note1 = Please follow the instructions below to create a bug report with the provided information -query_system_increment_compilation_note2 = See <https://github.com/rust-lang/rust/issues/84970> for more information +query_system_increment_compilation_note1 = please follow the instructions below to create a bug report with the provided information +query_system_increment_compilation_note2 = see <https://github.com/rust-lang/rust/issues/84970> for more information query_system_overflow_note = query depth increased by {$depth} when {$desc} diff --git a/compiler/rustc_resolve/messages.ftl b/compiler/rustc_resolve/messages.ftl index 47280a93677..0e566e20a12 100644 --- a/compiler/rustc_resolve/messages.ftl +++ b/compiler/rustc_resolve/messages.ftl @@ -470,6 +470,8 @@ resolve_variable_bound_with_different_mode = .label = bound in different ways .first_binding_span = first binding +resolve_variable_is_a_typo = you might have meant to use the similarly named previously used binding `{$typo}` + resolve_variable_is_not_bound_in_all_patterns = variable `{$name}` is not bound in all patterns diff --git a/compiler/rustc_resolve/src/diagnostics.rs b/compiler/rustc_resolve/src/diagnostics.rs index 5dfc4292a38..689e713ef46 100644 --- a/compiler/rustc_resolve/src/diagnostics.rs +++ b/compiler/rustc_resolve/src/diagnostics.rs @@ -1,3 +1,4 @@ +use itertools::Itertools as _; use rustc_ast::visit::{self, Visitor}; use rustc_ast::{ self as ast, CRATE_NODE_ID, Crate, ItemKind, ModKind, NodeId, Path, join_path_idents, @@ -661,8 +662,12 @@ impl<'ra, 'tcx> Resolver<'ra, 'tcx> { ResolutionError::VariableNotBoundInPattern(binding_error, parent_scope) => { let BindingError { name, target, origin, could_be_path } = binding_error; - let target_sp = target.iter().copied().collect::<Vec<_>>(); - let origin_sp = origin.iter().copied().collect::<Vec<_>>(); + let mut target_sp = target.iter().map(|pat| pat.span).collect::<Vec<_>>(); + target_sp.sort(); + target_sp.dedup(); + let mut origin_sp = origin.iter().map(|(span, _)| *span).collect::<Vec<_>>(); + origin_sp.sort(); + origin_sp.dedup(); let msp = MultiSpan::from_spans(target_sp.clone()); let mut err = self @@ -671,8 +676,35 @@ impl<'ra, 'tcx> Resolver<'ra, 'tcx> { for sp in target_sp { err.subdiagnostic(errors::PatternDoesntBindName { span: sp, name }); } - for sp in origin_sp { - err.subdiagnostic(errors::VariableNotInAllPatterns { span: sp }); + for sp in &origin_sp { + err.subdiagnostic(errors::VariableNotInAllPatterns { span: *sp }); + } + let mut suggested_typo = false; + if !target.iter().all(|pat| matches!(pat.kind, ast::PatKind::Ident(..))) + && !origin.iter().all(|(_, pat)| matches!(pat.kind, ast::PatKind::Ident(..))) + { + // The check above is so that when we encounter `match foo { (a | b) => {} }`, + // we don't suggest `(a | a) => {}`, which would never be what the user wants. + let mut target_visitor = BindingVisitor::default(); + for pat in &target { + target_visitor.visit_pat(pat); + } + target_visitor.identifiers.sort(); + target_visitor.identifiers.dedup(); + let mut origin_visitor = BindingVisitor::default(); + for (_, pat) in &origin { + origin_visitor.visit_pat(pat); + } + origin_visitor.identifiers.sort(); + origin_visitor.identifiers.dedup(); + // Find if the binding could have been a typo + if let Some(typo) = + find_best_match_for_name(&target_visitor.identifiers, name.name, None) + && !origin_visitor.identifiers.contains(&typo) + { + err.subdiagnostic(errors::PatternBindingTypo { spans: origin_sp, typo }); + suggested_typo = true; + } } if could_be_path { let import_suggestions = self.lookup_import_candidates( @@ -693,10 +725,86 @@ impl<'ra, 'tcx> Resolver<'ra, 'tcx> { }, ); - if import_suggestions.is_empty() { + if import_suggestions.is_empty() && !suggested_typo { + let kinds = [ + DefKind::Ctor(CtorOf::Variant, CtorKind::Const), + DefKind::Ctor(CtorOf::Struct, CtorKind::Const), + DefKind::Const, + DefKind::AssocConst, + ]; + let mut local_names = vec![]; + self.add_module_candidates( + parent_scope.module, + &mut local_names, + &|res| matches!(res, Res::Def(_, _)), + None, + ); + let local_names: FxHashSet<_> = local_names + .into_iter() + .filter_map(|s| match s.res { + Res::Def(_, def_id) => Some(def_id), + _ => None, + }) + .collect(); + + let mut local_suggestions = vec![]; + let mut suggestions = vec![]; + for kind in kinds { + if let Some(suggestion) = self.early_lookup_typo_candidate( + ScopeSet::All(Namespace::ValueNS), + &parent_scope, + name, + &|res: Res| match res { + Res::Def(k, _) => k == kind, + _ => false, + }, + ) && let Res::Def(kind, mut def_id) = suggestion.res + { + if let DefKind::Ctor(_, _) = kind { + def_id = self.tcx.parent(def_id); + } + let kind = kind.descr(def_id); + if local_names.contains(&def_id) { + // The item is available in the current scope. Very likely to + // be a typo. Don't use the full path. + local_suggestions.push(( + suggestion.candidate, + suggestion.candidate.to_string(), + kind, + )); + } else { + suggestions.push(( + suggestion.candidate, + self.def_path_str(def_id), + kind, + )); + } + } + } + let suggestions = if !local_suggestions.is_empty() { + // There is at least one item available in the current scope that is a + // likely typo. We only show those. + local_suggestions + } else { + suggestions + }; + for (name, sugg, kind) in suggestions { + err.span_suggestion_verbose( + span, + format!( + "you might have meant to use the similarly named {kind} `{name}`", + ), + sugg, + Applicability::MaybeIncorrect, + ); + suggested_typo = true; + } + } + if import_suggestions.is_empty() && !suggested_typo { let help_msg = format!( - "if you meant to match on a variant or a `const` item, consider \ - making the path in the pattern qualified: `path::to::ModOrType::{name}`", + "if you meant to match on a unit struct, unit variant or a `const` \ + item, consider making the path in the pattern qualified: \ + `path::to::ModOrType::{name}`", ); err.span_help(span, help_msg); } @@ -1016,6 +1124,39 @@ impl<'ra, 'tcx> Resolver<'ra, 'tcx> { .emit() } + fn def_path_str(&self, mut def_id: DefId) -> String { + // We can't use `def_path_str` in resolve. + let mut path = vec![def_id]; + while let Some(parent) = self.tcx.opt_parent(def_id) { + def_id = parent; + path.push(def_id); + if def_id.is_top_level_module() { + break; + } + } + // We will only suggest importing directly if it is accessible through that path. + path.into_iter() + .rev() + .map(|def_id| { + self.tcx + .opt_item_name(def_id) + .map(|name| { + match ( + def_id.is_top_level_module(), + def_id.is_local(), + self.tcx.sess.edition(), + ) { + (true, true, Edition::Edition2015) => String::new(), + (true, true, _) => kw::Crate.to_string(), + (true, false, _) | (false, _, _) => name.to_string(), + } + }) + .unwrap_or_else(|| "_".to_string()) + }) + .collect::<Vec<String>>() + .join("::") + } + pub(crate) fn add_scope_set_candidates( &mut self, suggestions: &mut Vec<TypoSuggestion>, @@ -1846,7 +1987,7 @@ impl<'ra, 'tcx> Resolver<'ra, 'tcx> { let extern_prelude_ambiguity = || { self.extern_prelude.get(&Macros20NormalizedIdent::new(ident)).is_some_and(|entry| { entry.item_binding.map(|(b, _)| b) == Some(b1) - && entry.flag_binding.as_ref().and_then(|pb| pb.get().binding()) == Some(b2) + && entry.flag_binding.as_ref().and_then(|pb| pb.get().0.binding()) == Some(b2) }) }; let (b1, b2, misc1, misc2, swapped) = if b2.span.is_dummy() && !b1.span.is_dummy() { @@ -3329,16 +3470,11 @@ fn show_candidates( err.note(note.to_string()); } } else { - let (_, descr_first, _, _, _) = &inaccessible_path_strings[0]; - let descr = if inaccessible_path_strings + let descr = inaccessible_path_strings .iter() - .skip(1) - .all(|(_, descr, _, _, _)| descr == descr_first) - { - descr_first - } else { - "item" - }; + .map(|&(_, descr, _, _, _)| descr) + .all_equal_value() + .unwrap_or("item"); let plural_descr = if descr.ends_with('s') { format!("{descr}es") } else { format!("{descr}s") }; @@ -3396,7 +3532,7 @@ impl UsePlacementFinder { } } -impl<'tcx> visit::Visitor<'tcx> for UsePlacementFinder { +impl<'tcx> Visitor<'tcx> for UsePlacementFinder { fn visit_crate(&mut self, c: &Crate) { if self.target_module == CRATE_NODE_ID { let inject = c.spans.inject_use_span; @@ -3424,6 +3560,22 @@ impl<'tcx> visit::Visitor<'tcx> for UsePlacementFinder { } } +#[derive(Default)] +struct BindingVisitor { + identifiers: Vec<Symbol>, + spans: FxHashMap<Symbol, Vec<Span>>, +} + +impl<'tcx> Visitor<'tcx> for BindingVisitor { + fn visit_pat(&mut self, pat: &ast::Pat) { + if let ast::PatKind::Ident(_, ident, _) = pat.kind { + self.identifiers.push(ident.name); + self.spans.entry(ident.name).or_default().push(ident.span); + } + visit::walk_pat(self, pat); + } +} + fn search_for_any_use_in_items(items: &[Box<ast::Item>]) -> Option<Span> { for item in items { if let ItemKind::Use(..) = item.kind diff --git a/compiler/rustc_resolve/src/errors.rs b/compiler/rustc_resolve/src/errors.rs index 63d6fa23a14..72be94e58db 100644 --- a/compiler/rustc_resolve/src/errors.rs +++ b/compiler/rustc_resolve/src/errors.rs @@ -986,6 +986,18 @@ pub(crate) struct VariableNotInAllPatterns { pub(crate) span: Span, } +#[derive(Subdiagnostic)] +#[multipart_suggestion( + resolve_variable_is_a_typo, + applicability = "maybe-incorrect", + style = "verbose" +)] +pub(crate) struct PatternBindingTypo { + #[suggestion_part(code = "{typo}")] + pub(crate) spans: Vec<Span>, + pub(crate) typo: Symbol, +} + #[derive(Diagnostic)] #[diag(resolve_name_defined_multiple_time)] #[note] diff --git a/compiler/rustc_resolve/src/ident.rs b/compiler/rustc_resolve/src/ident.rs index dae42645bec..bc06a227571 100644 --- a/compiler/rustc_resolve/src/ident.rs +++ b/compiler/rustc_resolve/src/ident.rs @@ -422,6 +422,8 @@ impl<'ra, 'tcx> Resolver<'ra, 'tcx> { // to detect potential ambiguities. let mut innermost_result: Option<(NameBinding<'_>, Flags)> = None; let mut determinacy = Determinacy::Determined; + let mut extern_prelude_item_binding = None; + let mut extern_prelude_flag_binding = None; // Shadowed bindings don't need to be marked as used or non-speculatively loaded. macro finalize_scope() { if innermost_result.is_none() { finalize } else { None } @@ -558,7 +560,10 @@ impl<'ra, 'tcx> Resolver<'ra, 'tcx> { Scope::ExternPreludeItems => { // FIXME: use `finalize_scope` here. match this.reborrow().extern_prelude_get_item(ident, finalize.is_some()) { - Some(binding) => Ok((binding, Flags::empty())), + Some(binding) => { + extern_prelude_item_binding = Some(binding); + Ok((binding, Flags::empty())) + } None => Err(Determinacy::determined( this.graph_root.unexpanded_invocations.borrow().is_empty(), )), @@ -566,7 +571,10 @@ impl<'ra, 'tcx> Resolver<'ra, 'tcx> { } Scope::ExternPreludeFlags => { match this.extern_prelude_get_flag(ident, finalize_scope!().is_some()) { - Some(binding) => Ok((binding, Flags::empty())), + Some(binding) => { + extern_prelude_flag_binding = Some(binding); + Ok((binding, Flags::empty())) + } None => Err(Determinacy::Determined), } } @@ -686,7 +694,16 @@ impl<'ra, 'tcx> Resolver<'ra, 'tcx> { } else { None }; - if let Some(kind) = ambiguity_error_kind { + // Skip ambiguity errors for extern flag bindings "overridden" + // by extern item bindings. + // FIXME: Remove with lang team approval. + let issue_145575_hack = Some(binding) + == extern_prelude_flag_binding + && extern_prelude_item_binding.is_some() + && extern_prelude_item_binding != Some(innermost_binding); + if let Some(kind) = ambiguity_error_kind + && !issue_145575_hack + { let misc = |f: Flags| { if f.contains(Flags::MISC_SUGGEST_CRATE) { AmbiguityErrorMisc::SuggestCrate diff --git a/compiler/rustc_resolve/src/late.rs b/compiler/rustc_resolve/src/late.rs index 6a12fafcb1e..521fef2b9d4 100644 --- a/compiler/rustc_resolve/src/late.rs +++ b/compiler/rustc_resolve/src/late.rs @@ -8,7 +8,6 @@ use std::assert_matches::debug_assert_matches; use std::borrow::Cow; -use std::collections::BTreeSet; use std::collections::hash_map::Entry; use std::mem::{replace, swap, take}; @@ -3682,31 +3681,30 @@ impl<'a, 'ast, 'ra, 'tcx> LateResolutionVisitor<'a, 'ast, 'ra, 'tcx> { // 2) Record any missing bindings or binding mode inconsistencies. for (map_outer, pat_outer) in not_never_pats.iter() { // Check against all arms except for the same pattern which is always self-consistent. - let inners = not_never_pats - .iter() - .filter(|(_, pat)| pat.id != pat_outer.id) - .flat_map(|(map, _)| map); - - for (&name, binding_inner) in inners { - match map_outer.get(&name) { - None => { - // The inner binding is missing in the outer. - let binding_error = - missing_vars.entry(name).or_insert_with(|| BindingError { - name, - origin: BTreeSet::new(), - target: BTreeSet::new(), - could_be_path: name.as_str().starts_with(char::is_uppercase), - }); - binding_error.origin.insert(binding_inner.span); - binding_error.target.insert(pat_outer.span); - } - Some(binding_outer) => { - if binding_outer.annotation != binding_inner.annotation { - // The binding modes in the outer and inner bindings differ. - inconsistent_vars - .entry(name) - .or_insert((binding_inner.span, binding_outer.span)); + let inners = not_never_pats.iter().filter(|(_, pat)| pat.id != pat_outer.id); + + for (map, pat) in inners { + for (&name, binding_inner) in map { + match map_outer.get(&name) { + None => { + // The inner binding is missing in the outer. + let binding_error = + missing_vars.entry(name).or_insert_with(|| BindingError { + name, + origin: Default::default(), + target: Default::default(), + could_be_path: name.as_str().starts_with(char::is_uppercase), + }); + binding_error.origin.push((binding_inner.span, (***pat).clone())); + binding_error.target.push((***pat_outer).clone()); + } + Some(binding_outer) => { + if binding_outer.annotation != binding_inner.annotation { + // The binding modes in the outer and inner bindings differ. + inconsistent_vars + .entry(name) + .or_insert((binding_inner.span, binding_outer.span)); + } } } } @@ -3719,7 +3717,7 @@ impl<'a, 'ast, 'ra, 'tcx> LateResolutionVisitor<'a, 'ast, 'ra, 'tcx> { v.could_be_path = false; } self.report_error( - *v.origin.iter().next().unwrap(), + v.origin.iter().next().unwrap().0, ResolutionError::VariableNotBoundInPattern(v, self.parent_scope), ); } diff --git a/compiler/rustc_resolve/src/late/diagnostics.rs b/compiler/rustc_resolve/src/late/diagnostics.rs index 80b2095d8cc..9e3c0938836 100644 --- a/compiler/rustc_resolve/src/late/diagnostics.rs +++ b/compiler/rustc_resolve/src/late/diagnostics.rs @@ -3099,7 +3099,7 @@ impl<'ast, 'ra, 'tcx> LateResolutionVisitor<'_, 'ast, 'ra, 'tcx> { |err, _, span, message, suggestion, span_suggs| { err.multipart_suggestion_verbose( message, - std::iter::once((span, suggestion)).chain(span_suggs.clone()).collect(), + std::iter::once((span, suggestion)).chain(span_suggs).collect(), Applicability::MaybeIncorrect, ); true diff --git a/compiler/rustc_resolve/src/lib.rs b/compiler/rustc_resolve/src/lib.rs index 9674c0356c2..8b185ce7ef2 100644 --- a/compiler/rustc_resolve/src/lib.rs +++ b/compiler/rustc_resolve/src/lib.rs @@ -230,8 +230,8 @@ enum Used { #[derive(Debug)] struct BindingError { name: Ident, - origin: BTreeSet<Span>, - target: BTreeSet<Span>, + origin: Vec<(Span, ast::Pat)>, + target: Vec<ast::Pat>, could_be_path: bool, } @@ -1031,7 +1031,7 @@ struct ExternPreludeEntry<'ra> { /// `flag_binding` is `None`, or when `extern crate` introducing `item_binding` used renaming. item_binding: Option<(NameBinding<'ra>, /* introduced by item */ bool)>, /// Binding from an `--extern` flag, lazily populated on first use. - flag_binding: Option<Cell<PendingBinding<'ra>>>, + flag_binding: Option<Cell<(PendingBinding<'ra>, /* finalized */ bool)>>, } impl ExternPreludeEntry<'_> { @@ -1042,7 +1042,7 @@ impl ExternPreludeEntry<'_> { fn flag() -> Self { ExternPreludeEntry { item_binding: None, - flag_binding: Some(Cell::new(PendingBinding::Pending)), + flag_binding: Some(Cell::new((PendingBinding::Pending, false))), } } } @@ -2245,14 +2245,16 @@ impl<'ra, 'tcx> Resolver<'ra, 'tcx> { fn extern_prelude_get_flag(&self, ident: Ident, finalize: bool) -> Option<NameBinding<'ra>> { let entry = self.extern_prelude.get(&Macros20NormalizedIdent::new(ident)); entry.and_then(|entry| entry.flag_binding.as_ref()).and_then(|flag_binding| { - let binding = match flag_binding.get() { + let (pending_binding, finalized) = flag_binding.get(); + let binding = match pending_binding { PendingBinding::Ready(binding) => { - if finalize { + if finalize && !finalized { self.cstore_mut().process_path_extern(self.tcx, ident.name, ident.span); } binding } PendingBinding::Pending => { + debug_assert!(!finalized); let crate_id = if finalize { self.cstore_mut().process_path_extern(self.tcx, ident.name, ident.span) } else { @@ -2264,7 +2266,7 @@ impl<'ra, 'tcx> Resolver<'ra, 'tcx> { }) } }; - flag_binding.set(PendingBinding::Ready(binding)); + flag_binding.set((PendingBinding::Ready(binding), finalize || finalized)); binding.or_else(|| finalize.then_some(self.dummy_binding)) }) } diff --git a/compiler/rustc_session/src/options.rs b/compiler/rustc_session/src/options.rs index 6d5be2d92cd..69facde6936 100644 --- a/compiler/rustc_session/src/options.rs +++ b/compiler/rustc_session/src/options.rs @@ -83,10 +83,77 @@ pub struct TargetModifier { pub value_name: String, } +mod target_modifier_consistency_check { + use super::*; + pub(super) fn sanitizer(l: &TargetModifier, r: Option<&TargetModifier>) -> bool { + let mut lparsed: SanitizerSet = Default::default(); + let lval = if l.value_name.is_empty() { None } else { Some(l.value_name.as_str()) }; + parse::parse_sanitizers(&mut lparsed, lval); + + let mut rparsed: SanitizerSet = Default::default(); + let rval = r.filter(|v| !v.value_name.is_empty()).map(|v| v.value_name.as_str()); + parse::parse_sanitizers(&mut rparsed, rval); + + // Some sanitizers need to be target modifiers, and some do not. + // For now, we should mark all sanitizers as target modifiers except for these: + // AddressSanitizer, LeakSanitizer + let tmod_sanitizers = SanitizerSet::MEMORY + | SanitizerSet::THREAD + | SanitizerSet::HWADDRESS + | SanitizerSet::CFI + | SanitizerSet::MEMTAG + | SanitizerSet::SHADOWCALLSTACK + | SanitizerSet::KCFI + | SanitizerSet::KERNELADDRESS + | SanitizerSet::SAFESTACK + | SanitizerSet::DATAFLOW; + + lparsed & tmod_sanitizers == rparsed & tmod_sanitizers + } + pub(super) fn sanitizer_cfi_normalize_integers( + opts: &Options, + l: &TargetModifier, + r: Option<&TargetModifier>, + ) -> bool { + // For kCFI, the helper flag -Zsanitizer-cfi-normalize-integers should also be a target modifier + if opts.unstable_opts.sanitizer.contains(SanitizerSet::KCFI) { + if let Some(r) = r { + return l.extend().tech_value == r.extend().tech_value; + } else { + return false; + } + } + true + } +} + impl TargetModifier { pub fn extend(&self) -> ExtendedTargetModifierInfo { self.opt.reparse(&self.value_name) } + // Custom consistency check for target modifiers (or default `l.tech_value == r.tech_value`) + // When other is None, consistency with default value is checked + pub fn consistent(&self, opts: &Options, other: Option<&TargetModifier>) -> bool { + assert!(other.is_none() || self.opt == other.unwrap().opt); + match self.opt { + OptionsTargetModifiers::UnstableOptions(unstable) => match unstable { + UnstableOptionsTargetModifiers::sanitizer => { + return target_modifier_consistency_check::sanitizer(self, other); + } + UnstableOptionsTargetModifiers::sanitizer_cfi_normalize_integers => { + return target_modifier_consistency_check::sanitizer_cfi_normalize_integers( + opts, self, other, + ); + } + _ => {} + }, + _ => {} + }; + match other { + Some(other) => self.extend().tech_value == other.extend().tech_value, + None => false, + } + } } fn tmod_push_impl( @@ -2504,13 +2571,13 @@ written to standard error output)"), retpoline_external_thunk: bool = (false, parse_bool, [TRACKED TARGET_MODIFIER], "enables retpoline-external-thunk, retpoline-indirect-branches and retpoline-indirect-calls \ target features (default: no)"), - sanitizer: SanitizerSet = (SanitizerSet::empty(), parse_sanitizers, [TRACKED], + sanitizer: SanitizerSet = (SanitizerSet::empty(), parse_sanitizers, [TRACKED TARGET_MODIFIER], "use a sanitizer"), sanitizer_cfi_canonical_jump_tables: Option<bool> = (Some(true), parse_opt_bool, [TRACKED], "enable canonical jump tables (default: yes)"), sanitizer_cfi_generalize_pointers: Option<bool> = (None, parse_opt_bool, [TRACKED], "enable generalizing pointer types (default: no)"), - sanitizer_cfi_normalize_integers: Option<bool> = (None, parse_opt_bool, [TRACKED], + sanitizer_cfi_normalize_integers: Option<bool> = (None, parse_opt_bool, [TRACKED TARGET_MODIFIER], "enable normalizing integer types (default: no)"), sanitizer_dataflow_abilist: Vec<String> = (Vec::new(), parse_comma_list, [TRACKED], "additional ABI list files that control how shadow parameters are passed (comma separated)"), diff --git a/compiler/rustc_span/src/symbol.rs b/compiler/rustc_span/src/symbol.rs index 77260d07c99..e5108d8b7e9 100644 --- a/compiler/rustc_span/src/symbol.rs +++ b/compiler/rustc_span/src/symbol.rs @@ -2162,6 +2162,7 @@ symbols! { target_family, target_feature, target_feature_11, + target_feature_inline_always, target_has_atomic, target_has_atomic_equal_alignment, target_has_atomic_load_store, @@ -2280,6 +2281,8 @@ symbols! { unboxed_closures, unchecked_add, unchecked_div, + unchecked_funnel_shl, + unchecked_funnel_shr, unchecked_mul, unchecked_rem, unchecked_shl, diff --git a/compiler/rustc_symbol_mangling/src/v0.rs b/compiler/rustc_symbol_mangling/src/v0.rs index 0cbd48ba08c..0655c2d5e81 100644 --- a/compiler/rustc_symbol_mangling/src/v0.rs +++ b/compiler/rustc_symbol_mangling/src/v0.rs @@ -82,9 +82,13 @@ pub(super) fn mangle<'tcx>( } pub fn mangle_internal_symbol<'tcx>(tcx: TyCtxt<'tcx>, item_name: &str) -> String { - if item_name == "rust_eh_personality" { + match item_name { // rust_eh_personality must not be renamed as LLVM hard-codes the name - return "rust_eh_personality".to_owned(); + "rust_eh_personality" => return item_name.to_owned(), + // Apple availability symbols need to not be mangled to be usable by + // C/Objective-C code. + "__isPlatformVersionAtLeast" | "__isOSVersionAtLeast" => return item_name.to_owned(), + _ => {} } let prefix = "_R"; diff --git a/compiler/rustc_target/src/callconv/loongarch.rs b/compiler/rustc_target/src/callconv/loongarch.rs index d567ad401bb..9213d73e24e 100644 --- a/compiler/rustc_target/src/callconv/loongarch.rs +++ b/compiler/rustc_target/src/callconv/loongarch.rs @@ -8,16 +8,16 @@ use crate::spec::HasTargetSpec; #[derive(Copy, Clone)] enum RegPassKind { - Float(Reg), - Integer(Reg), + Float { offset_from_start: Size, ty: Reg }, + Integer { offset_from_start: Size, ty: Reg }, Unknown, } #[derive(Copy, Clone)] enum FloatConv { - FloatPair(Reg, Reg), + FloatPair { first_ty: Reg, second_ty_offset_from_start: Size, second_ty: Reg }, Float(Reg), - MixedPair(Reg, Reg), + MixedPair { first_ty: Reg, second_ty_offset_from_start: Size, second_ty: Reg }, } #[derive(Copy, Clone)] @@ -37,6 +37,7 @@ fn should_use_fp_conv_helper<'a, Ty, C>( flen: u64, field1_kind: &mut RegPassKind, field2_kind: &mut RegPassKind, + offset_from_start: Size, ) -> Result<(), CannotUseFpConv> where Ty: TyAbiInterface<'a, C> + Copy, @@ -49,16 +50,16 @@ where } match (*field1_kind, *field2_kind) { (RegPassKind::Unknown, _) => { - *field1_kind = RegPassKind::Integer(Reg { - kind: RegKind::Integer, - size: arg_layout.size, - }); + *field1_kind = RegPassKind::Integer { + offset_from_start, + ty: Reg { kind: RegKind::Integer, size: arg_layout.size }, + }; } - (RegPassKind::Float(_), RegPassKind::Unknown) => { - *field2_kind = RegPassKind::Integer(Reg { - kind: RegKind::Integer, - size: arg_layout.size, - }); + (RegPassKind::Float { .. }, RegPassKind::Unknown) => { + *field2_kind = RegPassKind::Integer { + offset_from_start, + ty: Reg { kind: RegKind::Integer, size: arg_layout.size }, + }; } _ => return Err(CannotUseFpConv), } @@ -69,12 +70,16 @@ where } match (*field1_kind, *field2_kind) { (RegPassKind::Unknown, _) => { - *field1_kind = - RegPassKind::Float(Reg { kind: RegKind::Float, size: arg_layout.size }); + *field1_kind = RegPassKind::Float { + offset_from_start, + ty: Reg { kind: RegKind::Float, size: arg_layout.size }, + }; } (_, RegPassKind::Unknown) => { - *field2_kind = - RegPassKind::Float(Reg { kind: RegKind::Float, size: arg_layout.size }); + *field2_kind = RegPassKind::Float { + offset_from_start, + ty: Reg { kind: RegKind::Float, size: arg_layout.size }, + }; } _ => return Err(CannotUseFpConv), } @@ -96,13 +101,14 @@ where flen, field1_kind, field2_kind, + offset_from_start, ); } return Err(CannotUseFpConv); } } FieldsShape::Array { count, .. } => { - for _ in 0..count { + for i in 0..count { let elem_layout = arg_layout.field(cx, 0); should_use_fp_conv_helper( cx, @@ -111,6 +117,7 @@ where flen, field1_kind, field2_kind, + offset_from_start + elem_layout.size * i, )?; } } @@ -121,7 +128,15 @@ where } for i in arg_layout.fields.index_by_increasing_offset() { let field = arg_layout.field(cx, i); - should_use_fp_conv_helper(cx, &field, xlen, flen, field1_kind, field2_kind)?; + should_use_fp_conv_helper( + cx, + &field, + xlen, + flen, + field1_kind, + field2_kind, + offset_from_start + arg_layout.fields.offset(i), + )?; } } }, @@ -140,14 +155,52 @@ where { let mut field1_kind = RegPassKind::Unknown; let mut field2_kind = RegPassKind::Unknown; - if should_use_fp_conv_helper(cx, arg, xlen, flen, &mut field1_kind, &mut field2_kind).is_err() { + if should_use_fp_conv_helper( + cx, + arg, + xlen, + flen, + &mut field1_kind, + &mut field2_kind, + Size::ZERO, + ) + .is_err() + { return None; } match (field1_kind, field2_kind) { - (RegPassKind::Integer(l), RegPassKind::Float(r)) => Some(FloatConv::MixedPair(l, r)), - (RegPassKind::Float(l), RegPassKind::Integer(r)) => Some(FloatConv::MixedPair(l, r)), - (RegPassKind::Float(l), RegPassKind::Float(r)) => Some(FloatConv::FloatPair(l, r)), - (RegPassKind::Float(f), RegPassKind::Unknown) => Some(FloatConv::Float(f)), + ( + RegPassKind::Integer { offset_from_start, .. } + | RegPassKind::Float { offset_from_start, .. }, + _, + ) if offset_from_start != Size::ZERO => { + panic!("type {:?} has a first field with non-zero offset {offset_from_start:?}", arg.ty) + } + ( + RegPassKind::Integer { ty: first_ty, .. }, + RegPassKind::Float { offset_from_start, ty: second_ty }, + ) => Some(FloatConv::MixedPair { + first_ty, + second_ty_offset_from_start: offset_from_start, + second_ty, + }), + ( + RegPassKind::Float { ty: first_ty, .. }, + RegPassKind::Integer { offset_from_start, ty: second_ty }, + ) => Some(FloatConv::MixedPair { + first_ty, + second_ty_offset_from_start: offset_from_start, + second_ty, + }), + ( + RegPassKind::Float { ty: first_ty, .. }, + RegPassKind::Float { offset_from_start, ty: second_ty }, + ) => Some(FloatConv::FloatPair { + first_ty, + second_ty_offset_from_start: offset_from_start, + second_ty, + }), + (RegPassKind::Float { ty, .. }, RegPassKind::Unknown) => Some(FloatConv::Float(ty)), _ => None, } } @@ -165,11 +218,19 @@ where FloatConv::Float(f) => { arg.cast_to(f); } - FloatConv::FloatPair(l, r) => { - arg.cast_to(CastTarget::pair(l, r)); + FloatConv::FloatPair { first_ty, second_ty_offset_from_start, second_ty } => { + arg.cast_to(CastTarget::offset_pair( + first_ty, + second_ty_offset_from_start, + second_ty, + )); } - FloatConv::MixedPair(l, r) => { - arg.cast_to(CastTarget::pair(l, r)); + FloatConv::MixedPair { first_ty, second_ty_offset_from_start, second_ty } => { + arg.cast_to(CastTarget::offset_pair( + first_ty, + second_ty_offset_from_start, + second_ty, + )); } } return false; @@ -233,15 +294,27 @@ fn classify_arg<'a, Ty, C>( arg.cast_to(f); return; } - Some(FloatConv::FloatPair(l, r)) if *avail_fprs >= 2 => { + Some(FloatConv::FloatPair { first_ty, second_ty_offset_from_start, second_ty }) + if *avail_fprs >= 2 => + { *avail_fprs -= 2; - arg.cast_to(CastTarget::pair(l, r)); + arg.cast_to(CastTarget::offset_pair( + first_ty, + second_ty_offset_from_start, + second_ty, + )); return; } - Some(FloatConv::MixedPair(l, r)) if *avail_fprs >= 1 && *avail_gprs >= 1 => { + Some(FloatConv::MixedPair { first_ty, second_ty_offset_from_start, second_ty }) + if *avail_fprs >= 1 && *avail_gprs >= 1 => + { *avail_gprs -= 1; *avail_fprs -= 1; - arg.cast_to(CastTarget::pair(l, r)); + arg.cast_to(CastTarget::offset_pair( + first_ty, + second_ty_offset_from_start, + second_ty, + )); return; } _ => (), diff --git a/compiler/rustc_target/src/spec/targets/aarch64_pc_windows_msvc.rs b/compiler/rustc_target/src/spec/targets/aarch64_pc_windows_msvc.rs index f1b6fa123de..cd55576ef81 100644 --- a/compiler/rustc_target/src/spec/targets/aarch64_pc_windows_msvc.rs +++ b/compiler/rustc_target/src/spec/targets/aarch64_pc_windows_msvc.rs @@ -15,7 +15,7 @@ pub(crate) fn target() -> Target { llvm_target: "aarch64-pc-windows-msvc".into(), metadata: TargetMetadata { description: Some("ARM64 Windows MSVC".into()), - tier: Some(2), + tier: Some(1), host_tools: Some(true), std: Some(true), }, diff --git a/compiler/rustc_target/src/spec/targets/loongarch64_unknown_none.rs b/compiler/rustc_target/src/spec/targets/loongarch64_unknown_none.rs index 91e3064aaed..b6a08958284 100644 --- a/compiler/rustc_target/src/spec/targets/loongarch64_unknown_none.rs +++ b/compiler/rustc_target/src/spec/targets/loongarch64_unknown_none.rs @@ -17,7 +17,7 @@ pub(crate) fn target() -> Target { arch: "loongarch64".into(), options: TargetOptions { cpu: "generic".into(), - features: "+f,+d".into(), + features: "+f,+d,-lsx".into(), linker_flavor: LinkerFlavor::Gnu(Cc::No, Lld::Yes), linker: Some("rust-lld".into()), llvm_abiname: "lp64d".into(), diff --git a/compiler/rustc_target/src/spec/targets/nvptx64_nvidia_cuda.rs b/compiler/rustc_target/src/spec/targets/nvptx64_nvidia_cuda.rs index 598f0f19f0d..cada0dd640a 100644 --- a/compiler/rustc_target/src/spec/targets/nvptx64_nvidia_cuda.rs +++ b/compiler/rustc_target/src/spec/targets/nvptx64_nvidia_cuda.rs @@ -6,7 +6,7 @@ use crate::spec::{ pub(crate) fn target() -> Target { Target { arch: "nvptx64".into(), - data_layout: "e-p6:32:32-i64:64-i128:128-v16:16-v32:32-n16:32:64".into(), + data_layout: "e-p6:32:32-i64:64-i128:128-i256:256-v16:16-v32:32-n16:32:64".into(), llvm_target: "nvptx64-nvidia-cuda".into(), metadata: TargetMetadata { description: Some("--emit=asm generates PTX code that runs on NVIDIA GPUs".into()), diff --git a/compiler/rustc_trait_selection/src/error_reporting/infer/mod.rs b/compiler/rustc_trait_selection/src/error_reporting/infer/mod.rs index 9447612d026..812e20e4338 100644 --- a/compiler/rustc_trait_selection/src/error_reporting/infer/mod.rs +++ b/compiler/rustc_trait_selection/src/error_reporting/infer/mod.rs @@ -318,7 +318,7 @@ impl<'a, 'tcx> TypeErrCtxt<'a, 'tcx> { let (expected, found) = if expected_str == found_str { (join_path_syms(&expected_abs), join_path_syms(&found_abs)) } else { - (expected_str.clone(), found_str.clone()) + (expected_str, found_str) }; // We've displayed "expected `a::b`, found `a::b`". We add context to diff --git a/compiler/rustc_trait_selection/src/error_reporting/traits/suggestions.rs b/compiler/rustc_trait_selection/src/error_reporting/traits/suggestions.rs index 40a27c9aebe..f6dbbeb51ca 100644 --- a/compiler/rustc_trait_selection/src/error_reporting/traits/suggestions.rs +++ b/compiler/rustc_trait_selection/src/error_reporting/traits/suggestions.rs @@ -32,8 +32,8 @@ use rustc_middle::ty::print::{ }; use rustc_middle::ty::{ self, AdtKind, GenericArgs, InferTy, IsSuggestable, Ty, TyCtxt, TypeFoldable, TypeFolder, - TypeSuperFoldable, TypeVisitableExt, TypeckResults, Upcast, suggest_arbitrary_trait_bound, - suggest_constraining_type_param, + TypeSuperFoldable, TypeSuperVisitable, TypeVisitableExt, TypeVisitor, TypeckResults, Upcast, + suggest_arbitrary_trait_bound, suggest_constraining_type_param, }; use rustc_middle::{bug, span_bug}; use rustc_span::def_id::LocalDefId; @@ -263,6 +263,9 @@ impl<'a, 'tcx> TypeErrCtxt<'a, 'tcx> { _ => (false, None), }; + let mut finder = ParamFinder { .. }; + finder.visit_binder(&trait_pred); + // FIXME: Add check for trait bound that is already present, particularly `?Sized` so we // don't suggest `T: Sized + ?Sized`. loop { @@ -411,6 +414,26 @@ impl<'a, 'tcx> TypeErrCtxt<'a, 'tcx> { } } + hir::Node::TraitItem(hir::TraitItem { + generics, + kind: hir::TraitItemKind::Fn(..), + .. + }) + | hir::Node::ImplItem(hir::ImplItem { + generics, + trait_item_def_id: None, + kind: hir::ImplItemKind::Fn(..), + .. + }) if finder.can_suggest_bound(generics) => { + // Missing generic type parameter bound. + suggest_arbitrary_trait_bound( + self.tcx, + generics, + err, + trait_pred, + associated_ty, + ); + } hir::Node::Item(hir::Item { kind: hir::ItemKind::Struct(_, generics, _) @@ -423,7 +446,7 @@ impl<'a, 'tcx> TypeErrCtxt<'a, 'tcx> { | hir::ItemKind::Const(_, generics, _, _) | hir::ItemKind::TraitAlias(_, generics, _), .. - }) if !param_ty => { + }) if finder.can_suggest_bound(generics) => { // Missing generic type parameter bound. if suggest_arbitrary_trait_bound( self.tcx, @@ -5068,8 +5091,7 @@ impl<'a, 'tcx> TypeErrCtxt<'a, 'tcx> { // Suggesting `T: ?Sized` is only valid in an ADT if `T` is only used in a // borrow. `struct S<'a, T: ?Sized>(&'a T);` is valid, `struct S<T: ?Sized>(T);` // is not. Look for invalid "bare" parameter uses, and suggest using indirection. - let mut visitor = - FindTypeParam { param: param.name.ident().name, invalid_spans: vec![], nested: false }; + let mut visitor = FindTypeParam { param: param.name.ident().name, .. }; visitor.visit_item(item); if visitor.invalid_spans.is_empty() { return false; @@ -5228,7 +5250,7 @@ fn hint_missing_borrow<'tcx>( /// Used to suggest replacing associated types with an explicit type in `where` clauses. #[derive(Debug)] pub struct SelfVisitor<'v> { - pub paths: Vec<&'v hir::Ty<'v>>, + pub paths: Vec<&'v hir::Ty<'v>> = Vec::new(), pub name: Option<Symbol>, } @@ -5599,7 +5621,7 @@ fn point_at_assoc_type_restriction<G: EmissionGuarantee>( ); // Search for the associated type `Self::{name}`, get // its type and suggest replacing the bound with it. - let mut visitor = SelfVisitor { paths: vec![], name: Some(name) }; + let mut visitor = SelfVisitor { name: Some(name), .. }; visitor.visit_trait_ref(trait_ref); for path in visitor.paths { err.span_suggestion_verbose( @@ -5610,7 +5632,7 @@ fn point_at_assoc_type_restriction<G: EmissionGuarantee>( ); } } else { - let mut visitor = SelfVisitor { paths: vec![], name: None }; + let mut visitor = SelfVisitor { name: None, .. }; visitor.visit_trait_ref(trait_ref); let span: MultiSpan = visitor.paths.iter().map(|p| p.span).collect::<Vec<Span>>().into(); @@ -5640,8 +5662,8 @@ fn get_deref_type_and_refs(mut ty: Ty<'_>) -> (Ty<'_>, Vec<hir::Mutability>) { /// `param: ?Sized` would be a valid constraint. struct FindTypeParam { param: rustc_span::Symbol, - invalid_spans: Vec<Span>, - nested: bool, + invalid_spans: Vec<Span> = Vec::new(), + nested: bool = false, } impl<'v> Visitor<'v> for FindTypeParam { @@ -5679,3 +5701,38 @@ impl<'v> Visitor<'v> for FindTypeParam { } } } + +/// Look for type parameters in predicates. We use this to identify whether a bound is suitable in +/// on a given item. +struct ParamFinder { + params: Vec<Symbol> = Vec::new(), +} + +impl<'tcx> TypeVisitor<TyCtxt<'tcx>> for ParamFinder { + fn visit_ty(&mut self, t: Ty<'tcx>) -> Self::Result { + match t.kind() { + ty::Param(p) => self.params.push(p.name), + _ => {} + } + t.super_visit_with(self) + } +} + +impl ParamFinder { + /// Whether the `hir::Generics` of the current item can suggest the evaluated bound because its + /// references to type parameters are present in the generics. + fn can_suggest_bound(&self, generics: &hir::Generics<'_>) -> bool { + if self.params.is_empty() { + // There are no references to type parameters at all, so suggesting the bound + // would be reasonable. + return true; + } + generics.params.iter().any(|p| match p.name { + hir::ParamName::Plain(p_name) => { + // All of the parameters in the bound can be referenced in the current item. + self.params.iter().any(|p| *p == p_name.name || *p == kw::SelfUpper) + } + _ => true, + }) + } +} diff --git a/compiler/rustc_trait_selection/src/lib.rs b/compiler/rustc_trait_selection/src/lib.rs index e2b22f7bab7..fc0cf8f140a 100644 --- a/compiler/rustc_trait_selection/src/lib.rs +++ b/compiler/rustc_trait_selection/src/lib.rs @@ -19,6 +19,7 @@ #![feature(assert_matches)] #![feature(associated_type_defaults)] #![feature(box_patterns)] +#![feature(default_field_values)] #![feature(if_let_guard)] #![feature(iter_intersperse)] #![feature(iterator_try_reduce)] diff --git a/compiler/rustc_type_ir/src/interner.rs b/compiler/rustc_type_ir/src/interner.rs index f5448baf8c0..3355c0c2104 100644 --- a/compiler/rustc_type_ir/src/interner.rs +++ b/compiler/rustc_type_ir/src/interner.rs @@ -1,3 +1,4 @@ +use std::borrow::Borrow; use std::fmt::Debug; use std::hash::Hash; use std::ops::Deref; @@ -383,12 +384,12 @@ pub trait Interner: defining_anchor: Self::LocalDefId, ) -> Self::LocalDefIds; - type ProbeRef: Copy + Debug + Hash + Eq + Deref<Target = inspect::Probe<Self>>; - fn mk_probe_ref(self, probe: inspect::Probe<Self>) -> Self::ProbeRef; + type Probe: Debug + Hash + Eq + Borrow<inspect::Probe<Self>>; + fn mk_probe(self, probe: inspect::Probe<Self>) -> Self::Probe; fn evaluate_root_goal_for_proof_tree_raw( self, canonical_goal: CanonicalInput<Self>, - ) -> (QueryResult<Self>, Self::ProbeRef); + ) -> (QueryResult<Self>, Self::Probe); } /// Imagine you have a function `F: FnOnce(&[T]) -> R`, plus an iterator `iter` diff --git a/compiler/rustc_type_ir/src/solve/inspect.rs b/compiler/rustc_type_ir/src/solve/inspect.rs index 5452ac44158..3af2f8dbaf2 100644 --- a/compiler/rustc_type_ir/src/solve/inspect.rs +++ b/compiler/rustc_type_ir/src/solve/inspect.rs @@ -49,7 +49,7 @@ pub type CanonicalState<I, T> = Canonical<I, State<I, T>>; pub struct GoalEvaluation<I: Interner> { pub uncanonicalized_goal: Goal<I, I::Predicate>, pub orig_values: Vec<I::GenericArg>, - pub final_revision: I::ProbeRef, + pub final_revision: I::Probe, pub result: QueryResult<I>, } |