Re: [PATCH/7.10 2/2] gdbserver: Fix non-stop / fork / step-over issues

[Pedro Alves <palves at redhat dot com>]

On 08/05/2015 12:41 PM, Yao Qi wrote:
> Pedro Alves <palves@redhat.com> writes:
> 
>> I've now extended the test to run in a few different modes, along
>> a couple different axis.  One axis is with/without conditional
>> breakpoints on the target enabled.  That exposes the same fails you
>> saw on arm, on x86 gdbserver as well.  And then the other axis is
>> with/without joining _all_ threads before exiting.  If we gracefully
>> terminate the breakpoint thread (new mode), then the test should be
>> passing everywhere, because what fails is gdb's handling of the inferior
>> disappearing while a thread is stopped (and being inspected).
>> Therefore that new mode is not kfailed.
>>
>> For testing convenience, I've pushed this along with the previous patch
>> to the users/palves/gdbserver-fork-issues branch on sourceware.org.
>> Let me know if this works for you.
> 
> Thanks, Pedro.  There are no fails on arm-linux with GDBserver.  Some
> comments on your patch below,

Thanks Yao, that's great.

Thinking about this some more made me want to split both the patch
and test in two.  The process-exits-while-handling-breakpoints issues
happen independently of the fork issues, so I'll make the fork test always
join all threads before exiting.  Splitting will result in two
cleaner and more directed tests.  As per your testing, the fork test
will not need any kfail.

I'll be back to answer your questions and address your review when
I have a moment.

Thanks,
Pedro Alves