[PATCH] Improved handling of the naked attribute for MSVC compatibility

[Richard Smith]

On Fri, May 3, 2013 at 10:02 AM, Reid Kleckner <rnk at google.com> wrote:

> Cool!  There's two instances of trailing whitespace you'll want to fix.
>
> -  if (!isa<FunctionDecl>(D)) {
> +  // Microsoft mode expects the naked attribute to only be applied to a
> +  // function definition, or else it causes an error.  In non-Microsoft
> mode,
> +  // the attribute is allowed to be attached to a function definition or
> is
> +  // otherwise warned about.
> +  //
> +  // Because the attribute is handled before the function body is parsed,
> try
> +  // to use the Declarator to determine whether this is a function
> definition
> +  // or not.
> +  bool IsFunctionDecl = isa<FunctionDecl>(D);
> +  if (S.LangOpts.MicrosoftMode &&
> +      (!IsFunctionDecl || (PD && !PD->isFunctionDefinition()))) {
>
> I'm trying to find a way to avoid passing the Declarator through to
> here, but I assume you've looked and I can't find anything either...
> All of the definition-related checks rely on Body which isn't set
> until the body parsing is finished.
>
> If it were a TagDecl we could say 'D->isBeingDefined()', but for
> functions it looks like we haven't needed that.  Maybe it's worth
> adding a IsDefinition or BeingDefined bit to FunctionDecl?  Would
> anyone else find that useful or is it a bad idea?
>

This seems reasonable to me, although I'm not sure how generally useful it
would be.


> +    S.Diag(Attr.getLoc(), diag::err_attribute_wrong_decl_type)
> +      << Attr.getName() << ExpectedFunctionDefinition;
> +    return;
> +  } else if (!IsFunctionDecl) {
>      S.Diag(Attr.getLoc(), diag::warn_attribute_wrong_decl_type)
>        << Attr.getName() << ExpectedFunction;
>      return;
> @@ -4699,7 +4715,8 @@
>
>
> On Fri, May 3, 2013 at 9:09 AM, Aaron Ballman <aaron at aaronballman.com>
> wrote:
> > MSVC handles the naked attribute differently than clang does.
> > Specifically, in MSVC the __declspec(naked) must attach to a function
> > *definition*, and failure to do so will result in an error.
> >
> > This patch brings clang's behavior more in line with MSVC's for
> > __declspec(naked) in MS compatibility mode.  This does not modify the
> > behavior of the codegen for naked functions (as is being discussed
> > elsewhere).
> >
> > ~Aaron
> >
> > _______________________________________________
> > cfe-commits mailing list
> > cfe-commits at cs.uiuc.edu
> > http://lists.cs.uiuc.edu/mailman/listinfo/cfe-commits
> >
> _______________________________________________
> cfe-commits mailing list
> cfe-commits at cs.uiuc.edu
> http://lists.cs.uiuc.edu/mailman/listinfo/cfe-commits
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/cfe-commits/attachments/20130503/0015ea17/attachment.html>