Re: [PATCH] Add test that exercises all bfd architecture, osabi, endian, etc. combinations

[Pedro Alves <palves at redhat dot com>]

On 12/09/2016 02:56 PM, Yao Qi wrote:
> On Fri, Dec 9, 2016 at 2:32 PM, Pedro Alves <palves@redhat.com> wrote:
>> On 12/09/2016 01:56 PM, Yao Qi wrote:
>>> On Fri, Dec 9, 2016 at 1:11 PM, Pedro Alves <palves@redhat.com> wrote:
>>>> (long time passed...)
>>>>
>>>> This test would have caught the rl78 and rx problems that Yao
>>>> fixed today.  Clearly I should have pushed it in earlier so we would
>>>> have caught those regressions earlier.  :-/
>>>> The only reason I didn't, was that minute mentioned above.
>>>> I'm playing with splitting this test in 4 files, to bring that down
>>>> in a parallel run.  I'll repost with that.
>>>>
>>
>> Below's what I had in mind.  Splitting in 4 brings the time down to
>> ~30 seconds for me, while splitting in 8 brings it to ~25 seconds.
>> Looks like we hit diminishing returns, so I left it at 8.
>>
>> I also added kfail/skips for rl78 and rx, otherwise the test
>> crashes GDB...  Those can be removed as soon as your patches
>> are in (I hope, I haven't tested whether the archs have
>> further problems that would be exposed by this patch).
>>
>> WDYT?
>>
> 
> That looks good to me.

OK, I'll push it in then.

> After your test is pushed in, I'll push my
> rl78 and rx patches in, and remove the kfail from this test.  

Perfect.

> My
> rl78 and rx patches should go to 7.12 branch, but I am not sure
> this test case should go to 7.12 or not.

Yeah, should probably leave it on master only.

> 
>>>
>>> I'll extend all-architectures.exp to have a test "disassemble 0x0,+4"
>>> for PR 20939.  GDB now aborts due to the "foreign frame" again.
>>
>> That seems more dependent on host architecture than
>> target architecture, I think?  I.e., cycling over
>> target architectures and disassembling won't really add
>> more coverage?
> 
> Yes, PR 20939 is about host arch, but such test does find other issues,
> like PR 20955.  Other issues are shown up after the rl78 and rx segment
> fault is fixed.

OK, I see.

Thanks,
Pedro Alves