28 GCC HEAD regressions, 7 new, with your patch on 2004-06-03T00:30:06Z.

[Andrew Pinski]

On Jun 3, 2004, at 17:32, Matt Austern wrote:
>>
>> The new failures are:
>> native g++.sum g++.old-deja/g++.mike/eh31.C
>> native g++.sum g++.old-deja/g++.mike/init1.C
>> native g++.sum g++.old-deja/g++.mike/p2736.C
>> native g++.sum g++.old-deja/g++.mike/p9706.C
>> native g++.sum g++.old-deja/g++.other/init5.C
>
> These tests have a common feature: they  count constructions and 
> destructions.  If they've all started failing at once, there's 
> probably one common cause.  Almost certainly not my patch, though: 
> none of these have any templates in them, let alone explicitly 
> instantiated class templates.
>
>> native g++.sum g++.old-deja/g++.other/builtins9.C
>
> Something different from the previous crop of failures.  Still no 
> templates.  Hard for me to imagine what would have caused this one to 
> fail, actually.
>
>> native g++.sum g++.old-deja/g++.pt/overload11.C
>
> At last! A template!  But no class templates, no explicit 
> instantiations.  (This test defines and uses a function template whose 
> argument is a pointer-to-function type.)


Yes these were all caused by my patch to cp-tree.h, I committed a patch
which fixes every one except for the builtins9.C testcase which is
caused by my patch but is different from the others.

Here is my view of the problem:
This is a bug in the C++ front-end where there are two decls
for some reason.  One for the builtin function and one for the
function that is there in the code.  One way to fix this is to
do what the C front does and only combine the decls.

Thanks,
Andrew Pinski