llvmbot wrote:
<!--LLVM PR SUMMARY COMMENT--> @llvm/pr-subscribers-clang Author: Aaron Ballman (AaronBallman) <details> <summary>Changes</summary> It is possible to get a qualified rvalue in C. Consider: struct { cont int i; } foo(); _Generic(foo().i, ...); foo() returns an rvalue for the anonymous structure, the member access expression then results in an rvalue of type const int. The lvalue to rvalue conversion we perform on the controlling expression of the generic selection expression then fails to strip any qualifiers because the expression is already an rvalue. Discussion on the WG14 reflectors suggested that the qualifiers should still be stripped from the type of the controlling expression; the standard should be corrected to make this more clear. Fixes https://github.com/llvm/llvm-project/issues/96713 --- Full diff: https://github.com/llvm/llvm-project/pull/96913.diff 3 Files Affected: - (modified) clang/docs/ReleaseNotes.rst (+4) - (modified) clang/lib/Sema/SemaExpr.cpp (+17-9) - (modified) clang/test/Sema/generic-selection.c (+10) ``````````diff diff --git a/clang/docs/ReleaseNotes.rst b/clang/docs/ReleaseNotes.rst index da967fcdda808..10a718fe67e54 100644 --- a/clang/docs/ReleaseNotes.rst +++ b/clang/docs/ReleaseNotes.rst @@ -740,6 +740,10 @@ Bug Fixes in This Version negatives where the analysis failed to detect unchecked access to guarded data. +- When passing a qualified rvalue as the controlling expression of a + ``_Generic`` selection expression, Clang now properly strips the qualifiers. + Fixes #GH96713 + Bug Fixes to Compiler Builtins ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ diff --git a/clang/lib/Sema/SemaExpr.cpp b/clang/lib/Sema/SemaExpr.cpp index db44cfe1288b6..845fd44f3b39a 100644 --- a/clang/lib/Sema/SemaExpr.cpp +++ b/clang/lib/Sema/SemaExpr.cpp @@ -1828,18 +1828,26 @@ ExprResult Sema::CreateGenericSelectionExpr( // Look at the canonical type of the controlling expression in case it was a // deduced type like __auto_type. However, when issuing diagnostics, use the // type the user wrote in source rather than the canonical one. + // + // Note: there is an edge case for this in C where you can get a + // qualified rvalue and the qualifiers are not stripped by the lvalue + // conversion applied above. e.g., + // + // struct { const int i; } foo(); + // _Generic(foo().i, ...); + // + // Therefore, in C, we will still ask for the non-atomic, unqualified + // type despite those qualifiers generally being stripped above. + QualType ControllingTypeToUse = + ControllingExpr ? ControllingExpr->getType() : ControllingType->getType(); + ControllingTypeToUse = ControllingTypeToUse.getCanonicalType(); + if (!getLangOpts().CPlusPlus && ControllingExpr) + ControllingTypeToUse = ControllingTypeToUse.getAtomicUnqualifiedType(); for (unsigned i = 0; i < NumAssocs; ++i) { if (!Types[i]) DefaultIndex = i; - else if (ControllingExpr && - Context.typesAreCompatible( - ControllingExpr->getType().getCanonicalType(), - Types[i]->getType())) - CompatIndices.push_back(i); - else if (ControllingType && - Context.typesAreCompatible( - ControllingType->getType().getCanonicalType(), - Types[i]->getType())) + else if (Context.typesAreCompatible(ControllingTypeToUse, + Types[i]->getType())) CompatIndices.push_back(i); } diff --git a/clang/test/Sema/generic-selection.c b/clang/test/Sema/generic-selection.c index 1f17896ca4cda..ff2fe6ec44eff 100644 --- a/clang/test/Sema/generic-selection.c +++ b/clang/test/Sema/generic-selection.c @@ -86,3 +86,13 @@ void GH55562(void) { struct S s = { 0 }; int i = s.a; } + +struct { const int i; } GH96713_1(void); +void GH96713_2(void) { + _Static_assert( // ext-warning {{'_Static_assert' is a C11 extension}} + _Generic(GH96713_1().i, // ext-warning {{'_Generic' is a C11 extension}} + int : 1, + const int : 0 // expected-warning {{due to lvalue conversion of the controlling expression, association of type 'const int' will never be selected because it is qualified}} + ), "" + ); +} `````````` </details> https://github.com/llvm/llvm-project/pull/96913 _______________________________________________ cfe-commits mailing list cfe-commits@lists.llvm.org https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits