[cfe-dev] Clang crash with attribute-types that have extra parens:

[Keane, Erich via cfe-dev]

Thank you very much, I’ll use this for guidance when I start working again after the new year.

Thanks!
-Erich

From: Reid Kleckner [mailto:rnk at google.com]
Sent: Thursday, December 22, 2016 9:32 AM
To: Keane, Erich <erich.keane at intel.com>; cfe-dev <cfe-dev at lists.llvm.org>
Subject: Re: [cfe-dev] Clang crash with attribute-types that have extra parens:

+cfe-dev, I must have dropped it

Richard says we shouldn't build AttributedTypes for this, even though the comments there suggest it should be used for that. The AttributedType modified type is redundant with the VectorType element type.

So, I think beefing up the implementation of VectorTypeLoc is the way to go, which seems like what you were already doing.

On Wed, Dec 21, 2016 at 3:42 PM, Reid Kleckner <rnk at google.com<mailto:rnk at google.com>> wrote:
I don't think you want to go down the path of adding VisitVectorTypeLoc. Users can't "write" vector types directly, they can only write an AttributedType which desugars to a vector type and modifies the underlying type written in the source code. Type locations are supposed to model types that the user actually wrote.

You want to build a TypeLoc that's similar in structure to what you'd get for 'int (*vs_test);', which I think is a PointerTypeLoc around a ParenTypeLoc around an BuiltinTypeLoc.

I'm still looking at this, but that's what I think I know about this area.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/cfe-dev/attachments/20161224/d5dc5fc9/attachment.html>