Re: froggy

[Chris Moller <cmoller at redhat dot com>]

Phil Muldoon wrote:
> Chris Moller wrote:
>
>>>   I suspect the real answer is to write a new
>>> gdb backend, based on gdb/config/i386/linux.mh (and, I guess,
>>> gdb/config/i386/linux64.h).  (Judging from something you said a while
>>> back, this may have been obvious to you all along; I just got my Aha! a
>>> couple hours ago.)
>>>     
>
> Are you sure that there aren't ptrace semantics elsewhere in the code?
> Or perhaps harder to track down ... code that relies on ptrace
> semantics? We were discussing breakpoints and instruction stepping in
> the meeting and this would seem like a prime area. Because ...

All of the x86 linux ptrace and waitpid calls are from within the source
files selected by the linux.mh config, so I'm reasonably certain that
the scope of the delta is contained within those files.  Signal stuff,
I'm not so sure--there appears to be a level of indirection that's
making it hard to be sure.

>
>>> I've spent the last hour or so looking over the code represented by the
>>> linux.mh config and adapting it to froggy is a long way from
>>> trivial--several months, at a minimum, possibly as much as six
>>> months or
>>> more.
>>>     
>>
>> After a bit more discussion today, it looks like this is the way we'll
>> go.  Eric would like to see it ready for Fedora 11, which freezes in
>> April
>
> For a shippable product in Fedora 11 that surpasses or equals the
> reliability of existing GDB in Fedora 10, this seems a little
> aggressive to me.  

Me too...

> Tracking regressions in the GDB and utrace test-suites with these
> changes and triaging the bugs will be a huge job.  I don't wish to be
> negative energy - I think this is a great project, and am looking
> forward to it.  But the alpha freeze for F11 is 2009-01-20 and the
> beta freeze is 2009-03-10
> (https://fedoraproject.org/wiki/Releases/11/Schedule)

Assuming I'm right about the scope of the files that will be affected.
the only way, probably, that this will work on this short a schedule is
that there's a lot of complexity in those files that stem from the
interaction of fork(), exec*(), ptrace(), wait*(), and signal() in a
parent/child-debugger/debugged environment that simply go away with
froggy/utrace.  But I could be wrong...

>
> All being said, looking forward to it!
>
> Regards
>
> Phil
>

-- 
Chris Moller

  I know that you believe you understand what you think I said, but
  I'm not sure you realize that what you heard is not what I meant.
      -- Robert McCloskey

Attachment: signature.asc
Description: OpenPGP digital signature