Re: PING *objc, NeXT runtime, testsuite changes*

[IainS <developer at sandoe-acoustics dot co dot uk>]

On 16 Feb 2010, at 16:52, Mike Stump wrote:

> On Feb 16, 2010, at 7:58 AM, IainS wrote:
> > 3/ wrapper headers and target logic to remove non-informative error  
> > cruft from Darwin ObjC tests.
> > http://gcc.gnu.org/ml/gcc-patches/2009-12/msg01081.html
>
> I think this is a step forward, so Ok.

philosophically, it would be better to be supporting ObjC1 & 2 .. but  
it's very hard to make progress in a sea of reported errors.

>  I'm assuming that a run with -fnext-runtime and -fgnu-runtime  
> exposes no regressions (contrib/compare_tests) on darwin and linux.

It shows up one ICE on linux that I could avoid by skipping the test -  
but ACAICS the test should not ICE on linux so I'll leave for someone  
else to say that it should be eliminated.

> I do wonder about some of the specifics, like using xfail instead of  
> skip, but those issues I think will work themselves out in time.

Hm. so long as the intention is met that once __OBJC2__ is defined the  
tests are executed - I guess I don't mind which way it's done.

> Also, I don't think I like:
>
>  -#include <objc/Object.h>
>  +#include "../objc-obj-c++-shared/Object1.h"
>
> in general.  My wish would be for the gnu runtime to set feature  
> bits that can be directly tested, when it differs from the next- 
> runtime, and then we could use those feature bits in the testcases.   
> Likewise for the next runtime, though, they might just have a small  
> number of coarse (OBJC2) features.

I'm not sure that this is possible - the header contents (including  
logical purpose) are now different between NeXT and FSF objc - so I  
think we're stuck with a divergence.  What I've done is in line with  
what's been done on the 4.2/LLVM tree.

cheers,
Iain