[cfe-commits] r55299 - in /cfe/trunk/lib: AST/Expr.cpp CodeGen/CGExprConstant.cpp

[Anders Carlsson]

24 aug 2008 kl. 20.38 skrev Daniel Dunbar:

> Nice, thanks! A few notes:
>
>
>   return builtinID ==  
> Builtin::BI__builtin___CFStringMakeConstantString ||
> -         builtinID == Builtin::BI__builtin_classify_type;
> +         builtinID == Builtin::BI__builtin_classify_type ||
> +         builtinID == Builtin::BI__builtin_huge_valf;
>
> Quite a few builtins are treated as constant exprs. We should  
> probably add another parameter to BUILTIN in Builtins.def to handle  
> this instead of building a big disjunction here.

Sounds like a good idea!

>
>
> +  llvm::Constant *VisitCallExpr(const CallExpr *E) {
> +    if (const ImplicitCastExpr *IcExpr =
> +        dyn_cast<const ImplicitCastExpr>(E->getCallee()))
> +      if (const DeclRefExpr *DRExpr =
> +          dyn_cast<const DeclRefExpr>(IcExpr->getSubExpr()))
> +        if (const FunctionDecl *FDecl =
> +            dyn_cast<const FunctionDecl>(DRExpr->getDecl()))
> +          if (unsigned builtinID = FDecl->getIdentifier()- 
> >getBuiltinID())
> +            return EmitBuiltinExpr(builtinID, E);
> +
> +    CGM.ErrorUnsupported(E, "constant call expression");
> +    return llvm::Constant::getNullValue(ConvertType(E->getType()));
> +  }
>
> This seems rather fragile and possibly broken. Two immediates issues:
> (1) The cast is being ignored, which probably means we may end up  
> emitting the wrong type.
> (2) We also should handle the case when there is no ImplicitCastExpr  
> of a builtin.
>

This code was more or less copied from CodeGenFunction::EmitCallExpr -  
we should probably make both methods use the same code path.

Anders

-------------- next part --------------
A non-text attachment was scrubbed...
Name: smime.p7s
Type: application/pkcs7-signature
Size: 2415 bytes
Desc: not available
URL: <http://lists.llvm.org/pipermail/cfe-commits/attachments/20080824/bedf5079/attachment.bin>