[clang] [Clang] Fix aggregate emitter assertion for Atomic cast (PR #207811)
Amr Hesham via cfe-commits
cfe-commits at lists.llvm.org
Mon Jul 6 12:04:54 PDT 2026
AmrDeveloper wrote:
In Clang-17+ the AST will looks like
```
`-CStyleCastExpr <line:7:3, col:16> '_Atomic(S)' <ConstructorConversion>
`-CXXConstructExpr <col:3, col:16> 'S' 'void (const S &) noexcept'
`-ImplicitCastExpr <col:16> 'const S' lvalue <NoOp>
`-DeclRefExpr <col:16> 'S' lvalue Var 0x2c290190 's' 'S'
```
In CStyleCastExpr the expr type will be atomic but the subexpr type is non atomic
This was not a problem in Clang < 16 because the AST was emitting implicit cast to make both types equals
```
`-CStyleCastExpr <line:7:3, col:16> '_Atomic(S)' <NoOp>
`-ImplicitCastExpr <col:16> '_Atomic(S)' <NonAtomicToAtomic> part_of_explicit_cast
`-ImplicitCastExpr <col:16> 'S':'S' <ConstructorConversion> part_of_explicit_cast
`-CXXConstructExpr <col:16> 'S':'S' 'void (const S &) noexcept'
`-ImplicitCastExpr <col:16> 'const S':'const S' lvalue <NoOp>
`-DeclRefExpr <col:16> 'S':'S' lvalue Var 0x60192dc19930 's' 'S':'S'
```
In CStyleCastExpr, the expr type will be atomic and the subexpr type will be implicit casted to atomic,
the emitted IR is correct but we should expect the types are equals but without having the same qualifiers
I will add release notes before land, just created this PR to make sure this approach makes sense and will not cause any issues
Godbolt: https://godbolt.org/z/v87819zhq
https://github.com/llvm/llvm-project/pull/207811
More information about the cfe-commits
mailing list