Auto merge of #131547 - matthiaskrgr:rollup-ui4p744, r=matthiaskrgr

Rollup of 6 pull requests

Successful merges:

 - #129079 (Create `_imp__` symbols also when doing ThinLTO)
 - #131208 (ABI: Pass aggregates by value on AIX)
 - #131394 (fix(rustdoc): add space between struct fields and their descriptions)
 - #131519 (Use Default visibility for rustc-generated C symbol declarations)
 - #131541 (compiletest: Extract auxiliary-crate properties to their own module/struct)
 - #131542 (next-solver: remove outdated FIXMEs)

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

Author: bors

compiler/rustc_codegen_llvm/src/declare.rs

@@ -84,10 +84,9 @@ impl<'ll, 'tcx> CodegenCx<'ll, 'tcx> {
         unnamed: llvm::UnnamedAddr,
         fn_type: &'ll Type,
     ) -> &'ll Value {
-        // Declare C ABI functions with the visibility used by C by default.
-        let visibility = Visibility::from_generic(self.tcx.sess.default_visibility());
-
-        declare_raw_fn(self, name, llvm::CCallConv, unnamed, visibility, fn_type)
+        // Visibility should always be default for declarations, otherwise the linker may report an
+        // error.
+        declare_raw_fn(self, name, llvm::CCallConv, unnamed, Visibility::Default, fn_type)
     }
 
     /// Declare an entry Function

compiler/rustc_codegen_ssa/src/back/write.rs

@@ -2164,8 +2164,14 @@ fn msvc_imps_needed(tcx: TyCtxt<'_>) -> bool {
             && tcx.sess.opts.cg.prefer_dynamic)
     );
 
+    // We need to generate _imp__ symbol if we are generating an rlib or we include one
+    // indirectly from ThinLTO. In theory these are not needed as ThinLTO could resolve
+    // these, but it currently does not do so.
+    let can_have_static_objects =
+        tcx.sess.lto() == Lto::Thin || tcx.crate_types().iter().any(|ct| *ct == CrateType::Rlib);
+
     tcx.sess.target.is_like_windows &&
-        tcx.crate_types().iter().any(|ct| *ct == CrateType::Rlib) &&
+    can_have_static_objects   &&
     // ThinLTO can't handle this workaround in all cases, so we don't
     // emit the `__imp_` symbols. Instead we make them unnecessary by disallowing
     // dynamic linking when linker plugin LTO is enabled.

compiler/rustc_next_trait_solver/src/solve/eval_ctxt/canonical.rs

@@ -444,7 +444,6 @@ where
     for &arg in &state.value.var_values.var_values.as_slice()
         [orig_values.len()..state.value.var_values.len()]
     {
-        // FIXME: This is so ugly.
         let unconstrained = delegate.fresh_var_for_kind_with_span(arg, span);
         orig_values.push(unconstrained);
     }

compiler/rustc_next_trait_solver/src/solve/eval_ctxt/mod.rs

@@ -92,7 +92,6 @@ where
 #[derive_where(Clone, Debug, Default; I: Interner)]
 #[derive(TypeVisitable_Generic, TypeFoldable_Generic, Lift_Generic)]
 #[cfg_attr(feature = "nightly", derive(TyDecodable, TyEncodable, HashStable_NoContext))]
-// FIXME: This can be made crate-private once `EvalCtxt` also lives in this crate.
 struct NestedGoals<I: Interner> {
     /// These normalizes-to goals are treated specially during the evaluation
     /// loop. In each iteration we take the RHS of the projection, replace it with
@@ -421,6 +420,7 @@ where
         let (normalization_nested_goals, certainty) =
             self.instantiate_and_apply_query_response(goal.param_env, orig_values, response);
         self.inspect.goal_evaluation(goal_evaluation);
+
         // FIXME: We previously had an assert here that checked that recomputing
         // a goal after applying its constraints did not change its response.
         //

compiler/rustc_next_trait_solver/src/solve/mod.rs

@@ -10,9 +10,6 @@
 //!
 //! For a high-level overview of how this solver works, check out the relevant
 //! section of the rustc-dev-guide.
-//!
-//! FIXME(@lcnr): Write that section. If you read this before then ask me
-//! about it on zulip.
 
 mod alias_relate;
 mod assembly;

compiler/rustc_next_trait_solver/src/solve/normalizes_to/mod.rs

@@ -899,7 +899,7 @@ where
                 for ty in types.iter() {
                     // We can't find the intersection if the types used are generic.
                     //
-                    // FIXME(effects) do we want to look at where clauses to get some
+                    // FIXME(effects): do we want to look at where clauses to get some
                     // clue for the case where generic types are being used?
                     let Some(kind) = ty::EffectKind::try_from_ty(cx, ty) else {
                         return Err(NoSolution);

compiler/rustc_next_trait_solver/src/solve/trait_goals.rs

@@ -108,7 +108,6 @@ where
         ecx: &mut EvalCtxt<'_, D>,
         _guar: I::ErrorGuaranteed,
     ) -> Result<Candidate<I>, NoSolution> {
-        // FIXME: don't need to enter a probe here.
         ecx.probe_builtin_trait_candidate(BuiltinImplSource::Misc)
             .enter(|ecx| ecx.evaluate_added_goals_and_make_canonical_response(Certainty::Yes))
     }
@@ -463,7 +462,6 @@ where
         // Async coroutine unconditionally implement `Future`
         // Technically, we need to check that the future output type is Sized,
         // but that's already proven by the coroutine being WF.
-        // FIXME: use `consider_implied`
         ecx.probe_builtin_trait_candidate(BuiltinImplSource::Misc)
             .enter(|ecx| ecx.evaluate_added_goals_and_make_canonical_response(Certainty::Yes))
     }
@@ -489,7 +487,6 @@ where
         // Gen coroutines unconditionally implement `Iterator`
         // Technically, we need to check that the iterator output type is Sized,
         // but that's already proven by the coroutines being WF.
-        // FIXME: use `consider_implied`
         ecx.probe_builtin_trait_candidate(BuiltinImplSource::Misc)
             .enter(|ecx| ecx.evaluate_added_goals_and_make_canonical_response(Certainty::Yes))
     }
@@ -512,8 +509,7 @@ where
             return Err(NoSolution);
         }
 
-        // Gen coroutines unconditionally implement `FusedIterator`
-        // FIXME: use `consider_implied`
+        // Gen coroutines unconditionally implement `FusedIterator`.
         ecx.probe_builtin_trait_candidate(BuiltinImplSource::Misc)
             .enter(|ecx| ecx.evaluate_added_goals_and_make_canonical_response(Certainty::Yes))
     }
@@ -539,7 +535,6 @@ where
         // Gen coroutines unconditionally implement `Iterator`
         // Technically, we need to check that the iterator output type is Sized,
         // but that's already proven by the coroutines being WF.
-        // FIXME: use `consider_implied`
         ecx.probe_builtin_trait_candidate(BuiltinImplSource::Misc)
             .enter(|ecx| ecx.evaluate_added_goals_and_make_canonical_response(Certainty::Yes))
     }
@@ -610,7 +605,7 @@ where
             return Err(NoSolution);
         }
 
-        // FIXME(-Znext-solver): Implement this when we get const working in the new solver
+        // FIXME(effects): Implement this when we get const working in the new solver
 
         // `Destruct` is automatically implemented for every type in
         // non-const environments.
@@ -631,8 +626,6 @@ where
             return Err(NoSolution);
         }
 
-        // FIXME: This actually should destructure the `Result` we get from transmutability and
-        // register candidates. We probably need to register >1 since we may have an OR of ANDs.
         ecx.probe_builtin_trait_candidate(BuiltinImplSource::Misc).enter(|ecx| {
             let certainty = ecx.is_transmutable(
                 goal.param_env,

compiler/rustc_target/src/abi/call/powerpc64.rs

@@ -10,6 +10,7 @@ use crate::spec::HasTargetSpec;
 enum ABI {
     ELFv1, // original ABI used for powerpc64 (big-endian)
     ELFv2, // newer ABI used for powerpc64le and musl (both endians)
+    AIX,   // used by AIX OS, big-endian only
 }
 use ABI::*;
 
@@ -23,9 +24,9 @@ where
     C: HasDataLayout,
 {
     arg.layout.homogeneous_aggregate(cx).ok().and_then(|ha| ha.unit()).and_then(|unit| {
-        // ELFv1 only passes one-member aggregates transparently.
+        // ELFv1 and AIX only passes one-member aggregates transparently.
         // ELFv2 passes up to eight uniquely addressable members.
-        if (abi == ELFv1 && arg.layout.size > unit.size)
+        if ((abi == ELFv1 || abi == AIX) && arg.layout.size > unit.size)
             || arg.layout.size > unit.size.checked_mul(8, cx).unwrap()
         {
             return None;
@@ -55,8 +56,15 @@ where
         return;
     }
 
+    // The AIX ABI expect byval for aggregates
+    // See https://github.com/llvm/llvm-project/blob/main/clang/lib/CodeGen/Targets/PPC.cpp.
+    if !is_ret && abi == AIX {
+        arg.pass_by_stack_offset(None);
+        return;
+    }
+
     // The ELFv1 ABI doesn't return aggregates in registers
-    if is_ret && abi == ELFv1 {
+    if is_ret && (abi == ELFv1 || abi == AIX) {
         arg.make_indirect();
         return;
     }
@@ -93,6 +101,8 @@ where
 {
     let abi = if cx.target_spec().env == "musl" {
         ELFv2
+    } else if cx.target_spec().os == "aix" {
+        AIX
     } else {
         match cx.data_layout().endian {
             Endian::Big => ELFv1,

compiler/rustc_trait_selection/src/solve/delegate.rs

@@ -223,6 +223,8 @@ impl<'tcx> rustc_next_trait_solver::delegate::SolverDelegate for SolverDelegate<
         if eligible { Ok(Some(node_item.item.def_id)) } else { Ok(None) }
     }
 
+    // FIXME: This actually should destructure the `Result` we get from transmutability and
+    // register candidates. We probably need to register >1 since we may have an OR of ANDs.
     fn is_transmutable(
         &self,
         param_env: ty::ParamEnv<'tcx>,

src/librustdoc/html/static/css/rustdoc.css

@@ -230,6 +230,9 @@ h4.code-header {
 	padding: 0;
 	white-space: pre-wrap;
 }
+.structfield {
+	margin: 0.6em 0;
+}
 
 #crate-search,
 h1, h2, h3, h4, h5, h6,