[cfe-commits] r113015 - in /cfe/trunk: lib/CodeGen/CGExpr.cpp test/SemaCXX/blocks.cpp

[Douglas Gregor]

On Sep 3, 2010, at 2:36 PM, Fariborz Jahanian wrote:

> Author: fjahanian
> Date: Fri Sep  3 16:36:02 2010
> New Revision: 113015
> 
> URL: http://llvm.org/viewvc/llvm-project?rev=113015&view=rev
> Log:
> Cope with llvm's reference to bool type of 'i1' vs. clang's
> type of 'i8' for the same for __block variables of
> type bool. refixes radar 8382559.
> 
> Modified:
>    cfe/trunk/lib/CodeGen/CGExpr.cpp
>    cfe/trunk/test/SemaCXX/blocks.cpp
> 
> Modified: cfe/trunk/lib/CodeGen/CGExpr.cpp
> URL: http://llvm.org/viewvc/llvm-project/cfe/trunk/lib/CodeGen/CGExpr.cpp?rev=113015&r1=113014&r2=113015&view=diff
> ==============================================================================
> --- cfe/trunk/lib/CodeGen/CGExpr.cpp (original)
> +++ cfe/trunk/lib/CodeGen/CGExpr.cpp Fri Sep  3 16:36:02 2010
> @@ -335,6 +335,23 @@
>   llvm::Value *Value = EmitExprForReferenceBinding(*this, E, ReferenceTemporary,
>                                                    ReferenceTemporaryDtor,
>                                                    InitializedDecl);
> +  if (E->getType()->isBooleanType()) {
> +    // special handling for __block variable of bool type bound to
> +    // a reference type.
> +    bool block_byref_var = false;
> +    if (const BlockDeclRefExpr *BE = dyn_cast<BlockDeclRefExpr>(E))
> +      block_byref_var = BE->isByRef();
> +    else if (const DeclRefExpr *BD = dyn_cast<DeclRefExpr>(E)) {
> +      const NamedDecl *ND = BD->getDecl();
> +      if (const VarDecl *VD = dyn_cast<VarDecl>(ND))
> +        block_byref_var = VD->hasAttr<BlocksAttr>();
> +    }
> +    if (block_byref_var) {
> +      const llvm::Type *T = ConvertTypeForMem(E->getType());
> +      T = llvm::PointerType::getUnqual(T);
> +      Value = Builder.CreateBitCast(Value, T);
> +    }
> +  }

I'm really uncomfortable with this change. EmitExprForReferenceBinding needs to be doing the right thing for block references of bool type, and adding this code here feels like a workaround.

	- Doug