[patch] Don't let invalid methods mark their RecordDecl invalid

[John McCall]

On Dec 20, 2013, at 4:06 PM, Reid Kleckner <rnk at google.com> wrote:
> +John, since it was his code.
> 
> Are there other ways we can get an invalid record decl and then do layout on a derived class to trigger PR18284?  If so, it seems like this doesn't fix the underlying issue.

A class should definitely be invalid if any of its base classes is invalid.  Making it invalid when one of its members is invalid is probably for the best; I would guess that that’s now occurring through some other channel, causing the original test to successfully not crash.

I’m not sure what about the linked patch is making a base class invalid when a method in a derived class is invalidated, though?

John.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/cfe-commits/attachments/20131220/ecc47abf/attachment.html>