Auto merge of #130177 - workingjubilee:rollup-m4r01yn, r=workingjubilee

Rollup of 14 pull requests

Successful merges:

 - #128316 (Stabilize most of `io_error_more`)
 - #129473 (use  `download-ci-llvm=true` in the default compiler config)
 - #129529 (Add test to build crates used by r-a on stable)
 - #129778 (interpret: make typed copies lossy wrt provenance and padding)
 - #129981 (Remove `serialized_bitcode` from `LtoModuleCodegen`.)
 - #130025 (Also emit `missing_docs` lint with `--test` to fulfil expectations)
 - #130040 (unify `llvm-bitcode-linker`, `wasm-component-ld` and llvm-tools logics)
 - #130094 (Inform the solver if evaluation is concurrent)
 - #130132 ([illumos] enable SIGSEGV handler to detect stack overflows)
 - #130146 (bootstrap `naked_asm!` for `compiler-builtins`)
 - #130149 (Helper function for formatting with `LifetimeSuggestionPosition`)
 - #130152 (adapt a test for llvm 20)
 - #130162 (bump download-ci-llvm-stamp)
 - #130164 (move some const fn out of the const_ptr_as_ref feature)

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

Author: bors

Diff for: compiler/rustc_builtin_macros/src/asm.rs

@@ -812,6 +812,44 @@ pub(super) fn expand_asm<'cx>(
     })
 }
 
+pub(super) fn expand_naked_asm<'cx>(
+    ecx: &'cx mut ExtCtxt<'_>,
+    sp: Span,
+    tts: TokenStream,
+) -> MacroExpanderResult<'cx> {
+    ExpandResult::Ready(match parse_args(ecx, sp, tts, false) {
+        Ok(args) => {
+            let ExpandResult::Ready(mac) = expand_preparsed_asm(ecx, args) else {
+                return ExpandResult::Retry(());
+            };
+            let expr = match mac {
+                Ok(mut inline_asm) => {
+                    // for future compatibility, we always set the NORETURN option.
+                    //
+                    // When we turn `asm!` into `naked_asm!` with this implementation, we can drop
+                    // the `options(noreturn)`, which makes the upgrade smooth when `naked_asm!`
+                    // starts disallowing the `noreturn` option in the future
+                    inline_asm.options |= ast::InlineAsmOptions::NORETURN;
+
+                    P(ast::Expr {
+                        id: ast::DUMMY_NODE_ID,
+                        kind: ast::ExprKind::InlineAsm(P(inline_asm)),
+                        span: sp,
+                        attrs: ast::AttrVec::new(),
+                        tokens: None,
+                    })
+                }
+                Err(guar) => DummyResult::raw_expr(sp, Some(guar)),
+            };
+            MacEager::expr(expr)
+        }
+        Err(err) => {
+            let guar = err.emit();
+            DummyResult::any(sp, guar)
+        }
+    })
+}
+
 pub(super) fn expand_global_asm<'cx>(
     ecx: &'cx mut ExtCtxt<'_>,
     sp: Span,

Diff for: compiler/rustc_builtin_macros/src/lib.rs

@@ -94,6 +94,7 @@ pub fn register_builtin_macros(resolver: &mut dyn ResolverExpand) {
         line: source_util::expand_line,
         log_syntax: log_syntax::expand_log_syntax,
         module_path: source_util::expand_mod,
+        naked_asm: asm::expand_naked_asm,
         option_env: env::expand_option_env,
         pattern_type: pattern_type::expand,
         std_panic: edition_panic::expand_panic,

Diff for: compiler/rustc_builtin_macros/src/test_harness.rs

@@ -326,6 +326,8 @@ fn mk_main(cx: &mut TestCtxt<'_>) -> P<ast::Item> {
     let main_attr = ecx.attr_word(sym::rustc_main, sp);
     // #[coverage(off)]
     let coverage_attr = ecx.attr_nested_word(sym::coverage, sym::off, sp);
+    // #[allow(missing_docs)]
+    let missing_docs_attr = ecx.attr_nested_word(sym::allow, sym::missing_docs, sp);
 
     // pub fn main() { ... }
     let main_ret_ty = ecx.ty(sp, ast::TyKind::Tup(ThinVec::new()));
@@ -355,7 +357,7 @@ fn mk_main(cx: &mut TestCtxt<'_>) -> P<ast::Item> {
 
     let main = P(ast::Item {
         ident: main_id,
-        attrs: thin_vec![main_attr, coverage_attr],
+        attrs: thin_vec![main_attr, coverage_attr, missing_docs_attr],
         id: ast::DUMMY_NODE_ID,
         kind: main,
         vis: ast::Visibility { span: sp, kind: ast::VisibilityKind::Public, tokens: None },

Diff for: compiler/rustc_codegen_gcc/src/back/lto.rs

@@ -272,7 +272,6 @@ fn fat_lto(
             }*/
         }
     };
-    let mut serialized_bitcode = Vec::new();
     {
         info!("using {:?} as a base module", module.name);
 
@@ -317,7 +316,6 @@ fn fat_lto(
                     unimplemented!("from uncompressed file")
                 }
             }
-            serialized_bitcode.push(bc_decoded);
         }
         save_temp_bitcode(cgcx, &module, "lto.input");
 
@@ -337,7 +335,7 @@ fn fat_lto(
     // of now.
     module.module_llvm.temp_dir = Some(tmp_path);
 
-    Ok(LtoModuleCodegen::Fat { module, _serialized_bitcode: serialized_bitcode })
+    Ok(LtoModuleCodegen::Fat(module))
 }
 
 pub struct ModuleBuffer(PathBuf);

Diff for: compiler/rustc_codegen_llvm/src/back/lto.rs

@@ -314,7 +314,6 @@ fn fat_lto(
             }
         }
     };
-    let mut serialized_bitcode = Vec::new();
     {
         let (llcx, llmod) = {
             let llvm = &module.module_llvm;
@@ -342,9 +341,7 @@ fn fat_lto(
         serialized_modules.sort_by(|module1, module2| module1.1.cmp(&module2.1));
 
         // For all serialized bitcode files we parse them and link them in as we did
-        // above, this is all mostly handled in C++. Like above, though, we don't
-        // know much about the memory management here so we err on the side of being
-        // save and persist everything with the original module.
+        // above, this is all mostly handled in C++.
         let mut linker = Linker::new(llmod);
         for (bc_decoded, name) in serialized_modules {
             let _timer = cgcx
@@ -355,7 +352,6 @@ fn fat_lto(
             info!("linking {:?}", name);
             let data = bc_decoded.data();
             linker.add(data).map_err(|()| write::llvm_err(dcx, LlvmError::LoadBitcode { name }))?;
-            serialized_bitcode.push(bc_decoded);
         }
         drop(linker);
         save_temp_bitcode(cgcx, &module, "lto.input");
@@ -372,7 +368,7 @@ fn fat_lto(
         }
     }
 
-    Ok(LtoModuleCodegen::Fat { module, _serialized_bitcode: serialized_bitcode })
+    Ok(LtoModuleCodegen::Fat(module))
 }
 
 pub(crate) struct Linker<'a>(&'a mut llvm::Linker<'a>);

Diff for: compiler/rustc_codegen_ssa/src/back/lto.rs

@@ -41,18 +41,14 @@ pub struct ThinShared<B: WriteBackendMethods> {
 }
 
 pub enum LtoModuleCodegen<B: WriteBackendMethods> {
-    Fat {
-        module: ModuleCodegen<B::Module>,
-        _serialized_bitcode: Vec<SerializedModule<B::ModuleBuffer>>,
-    },
-
+    Fat(ModuleCodegen<B::Module>),
     Thin(ThinModule<B>),
 }
 
 impl<B: WriteBackendMethods> LtoModuleCodegen<B> {
     pub fn name(&self) -> &str {
         match *self {
-            LtoModuleCodegen::Fat { .. } => "everything",
+            LtoModuleCodegen::Fat(_) => "everything",
             LtoModuleCodegen::Thin(ref m) => m.name(),
         }
     }
@@ -68,7 +64,7 @@ impl<B: WriteBackendMethods> LtoModuleCodegen<B> {
         cgcx: &CodegenContext<B>,
     ) -> Result<ModuleCodegen<B::Module>, FatalError> {
         match self {
-            LtoModuleCodegen::Fat { mut module, .. } => {
+            LtoModuleCodegen::Fat(mut module) => {
                 B::optimize_fat(cgcx, &mut module)?;
                 Ok(module)
             }
@@ -81,7 +77,7 @@ impl<B: WriteBackendMethods> LtoModuleCodegen<B> {
     pub fn cost(&self) -> u64 {
         match *self {
             // Only one module with fat LTO, so the cost doesn't matter.
-            LtoModuleCodegen::Fat { .. } => 0,
+            LtoModuleCodegen::Fat(_) => 0,
             LtoModuleCodegen::Thin(ref m) => m.cost(),
         }
     }

Diff for: compiler/rustc_const_eval/src/const_eval/eval_queries.rs

@@ -94,7 +94,7 @@ fn eval_body_using_ecx<'tcx, R: InterpretationResult<'tcx>>(
     let intern_result = intern_const_alloc_recursive(ecx, intern_kind, &ret);
 
     // Since evaluation had no errors, validate the resulting constant.
-    const_validate_mplace(&ecx, &ret, cid)?;
+    const_validate_mplace(ecx, &ret, cid)?;
 
     // Only report this after validation, as validaiton produces much better diagnostics.
     // FIXME: ensure validation always reports this and stop making interning care about it.
@@ -391,7 +391,7 @@ fn eval_in_interpreter<'tcx, R: InterpretationResult<'tcx>>(
 
 #[inline(always)]
 fn const_validate_mplace<'tcx>(
-    ecx: &InterpCx<'tcx, CompileTimeMachine<'tcx>>,
+    ecx: &mut InterpCx<'tcx, CompileTimeMachine<'tcx>>,
     mplace: &MPlaceTy<'tcx>,
     cid: GlobalId<'tcx>,
 ) -> Result<(), ErrorHandled> {

Diff for: compiler/rustc_const_eval/src/const_eval/machine.rs

@@ -1,16 +1,16 @@
-use std::borrow::Borrow;
+use std::borrow::{Borrow, Cow};
 use std::fmt;
 use std::hash::Hash;
 use std::ops::ControlFlow;
 
 use rustc_ast::Mutability;
-use rustc_data_structures::fx::{FxIndexMap, IndexEntry};
+use rustc_data_structures::fx::{FxHashMap, FxIndexMap, IndexEntry};
 use rustc_hir::def_id::{DefId, LocalDefId};
 use rustc_hir::{self as hir, LangItem, CRATE_HIR_ID};
 use rustc_middle::mir::AssertMessage;
 use rustc_middle::query::TyCtxtAt;
 use rustc_middle::ty::layout::{FnAbiOf, TyAndLayout};
-use rustc_middle::ty::{self, TyCtxt};
+use rustc_middle::ty::{self, Ty, TyCtxt};
 use rustc_middle::{bug, mir};
 use rustc_span::symbol::{sym, Symbol};
 use rustc_span::Span;
@@ -24,8 +24,8 @@ use crate::fluent_generated as fluent;
 use crate::interpret::{
     self, compile_time_machine, err_ub, throw_exhaust, throw_inval, throw_ub_custom, throw_unsup,
     throw_unsup_format, AllocId, AllocRange, ConstAllocation, CtfeProvenance, FnArg, Frame,
-    GlobalAlloc, ImmTy, InterpCx, InterpResult, MPlaceTy, OpTy, Pointer, PointerArithmetic, Scalar,
-    StackPopCleanup,
+    GlobalAlloc, ImmTy, InterpCx, InterpResult, MPlaceTy, OpTy, Pointer, PointerArithmetic,
+    RangeSet, Scalar, StackPopCleanup,
 };
 
 /// When hitting this many interpreted terminators we emit a deny by default lint
@@ -65,6 +65,9 @@ pub struct CompileTimeMachine<'tcx> {
     /// storing the result in the given `AllocId`.
     /// Used to prevent reads from a static's base allocation, as that may allow for self-initialization loops.
     pub(crate) static_root_ids: Option<(AllocId, LocalDefId)>,
+
+    /// A cache of "data range" computations for unions (i.e., the offsets of non-padding bytes).
+    union_data_ranges: FxHashMap<Ty<'tcx>, RangeSet>,
 }
 
 #[derive(Copy, Clone)]
@@ -99,6 +102,7 @@ impl<'tcx> CompileTimeMachine<'tcx> {
             can_access_mut_global,
             check_alignment,
             static_root_ids: None,
+            union_data_ranges: FxHashMap::default(),
         }
     }
 }
@@ -766,6 +770,19 @@ impl<'tcx> interpret::Machine<'tcx> for CompileTimeMachine<'tcx> {
         }
         Ok(())
     }
+
+    fn cached_union_data_range<'e>(
+        ecx: &'e mut InterpCx<'tcx, Self>,
+        ty: Ty<'tcx>,
+        compute_range: impl FnOnce() -> RangeSet,
+    ) -> Cow<'e, RangeSet> {
+        if ecx.tcx.sess.opts.unstable_opts.extra_const_ub_checks {
+            Cow::Borrowed(ecx.machine.union_data_ranges.entry(ty).or_insert_with(compute_range))
+        } else {
+            // Don't bother caching, we're only doing one validation at the end anyway.
+            Cow::Owned(compute_range())
+        }
+    }
 }
 
 // Please do not add any code below the above `Machine` trait impl. I (oli-obk) plan more cleanups

Diff for: compiler/rustc_const_eval/src/interpret/discriminant.rs

@@ -7,7 +7,7 @@ use rustc_target::abi::{self, TagEncoding, VariantIdx, Variants};
 use tracing::{instrument, trace};
 
 use super::{
-    err_ub, throw_ub, ImmTy, InterpCx, InterpResult, Machine, Readable, Scalar, Writeable,
+    err_ub, throw_ub, ImmTy, InterpCx, InterpResult, Machine, Projectable, Scalar, Writeable,
 };
 
 impl<'tcx, M: Machine<'tcx>> InterpCx<'tcx, M> {
@@ -60,7 +60,7 @@ impl<'tcx, M: Machine<'tcx>> InterpCx<'tcx, M> {
     #[instrument(skip(self), level = "trace")]
     pub fn read_discriminant(
         &self,
-        op: &impl Readable<'tcx, M::Provenance>,
+        op: &impl Projectable<'tcx, M::Provenance>,
     ) -> InterpResult<'tcx, VariantIdx> {
         let ty = op.layout().ty;
         trace!("read_discriminant_value {:#?}", op.layout());

Diff for: compiler/rustc_const_eval/src/interpret/machine.rs

@@ -10,6 +10,7 @@ use rustc_apfloat::{Float, FloatConvert};
 use rustc_ast::{InlineAsmOptions, InlineAsmTemplatePiece};
 use rustc_middle::query::TyCtxtAt;
 use rustc_middle::ty::layout::TyAndLayout;
+use rustc_middle::ty::Ty;
 use rustc_middle::{mir, ty};
 use rustc_span::def_id::DefId;
 use rustc_span::Span;
@@ -19,7 +20,7 @@ use rustc_target::spec::abi::Abi as CallAbi;
 use super::{
     throw_unsup, throw_unsup_format, AllocBytes, AllocId, AllocKind, AllocRange, Allocation,
     ConstAllocation, CtfeProvenance, FnArg, Frame, ImmTy, InterpCx, InterpResult, MPlaceTy,
-    MemoryKind, Misalignment, OpTy, PlaceTy, Pointer, Provenance, CTFE_ALLOC_SALT,
+    MemoryKind, Misalignment, OpTy, PlaceTy, Pointer, Provenance, RangeSet, CTFE_ALLOC_SALT,
 };
 
 /// Data returned by [`Machine::after_stack_pop`], and consumed by
@@ -578,6 +579,15 @@ pub trait Machine<'tcx>: Sized {
         ecx: &InterpCx<'tcx, Self>,
         instance: Option<ty::Instance<'tcx>>,
     ) -> usize;
+
+    fn cached_union_data_range<'e>(
+        _ecx: &'e mut InterpCx<'tcx, Self>,
+        _ty: Ty<'tcx>,
+        compute_range: impl FnOnce() -> RangeSet,
+    ) -> Cow<'e, RangeSet> {
+        // Default to no caching.
+        Cow::Owned(compute_range())
+    }
 }
 
 /// A lot of the flexibility above is just needed for `Miri`, but all "compile-time" machines