On 2014.04.07 at 08:39 +0100, Nathan Sidwell wrote: > On 04/06/14 10:50, Markus Trippelsdorf wrote: > > On 2014.04.06 at 09:13 +0100, Nathan Sidwell wrote: > >> On 04/04/14 18:33, Nathan Sidwell wrote: > >> > >>> I'm testing a patch that makes the test in the loop: > >>> > >>> if (TREE_PUBLIC (base_binfo) > >> > >> Hm, binfo's aren't noted that way, it's encoded in BINFO_ACCESS and > >> SET_BINFO_ACCESS in search.c. We'd need to move those back to cp.h or > >> expose an > >> interface in search.c. This is looking like a rat hole ... > > That was my rustiness, it is not as complicated. > > Here is a patch to implement the behaviour we discussed: > *) only consider public bases > *) only consider polymorphic bases, unless Weffc++ is also specified > > I have tested this in the usual manner, and the new testcases pass with the > patch and fail without it. > > Markus, if you could give this a try and see whether it fixes the problem you > reported, that'd be great.
Well, all I've tested is if the warnings during an LLVM build vanish. And this is indeed the case with your new patch, too. -- Markus
