[lldb-dev] Process attach and pgid of forked process under Linux

[Andrew MacPherson]

Ok great, I'm attaching a patch here that just uses getpgid() to determine
the pgid to use with waitpid(). I get the same unit test results before and
after the patch (there are a few tests that fail consistently for me). I
tried using a simple -1 with waitpid() but this results in a hang in the
unit tests. This is probably something worth investigating but for now this
patch does resolve the issues around waitpid() when attaching.

I mentioned a few issues I ran into with the unit tests in the email I just
pinged with a Linux detach patch, basically there appears to be a bug or
limitation somewhere in that doing a "continue" in asynchronous mode in a
unit test breaks things and this is needed for most attach-related debug
testing. I can probably look into this (along with the couple of failing
tests I have) but likely won't be able to this week. I haven't filed a bug
for the consistently-failing tests on my end since it doesn't appear to be
happening for others and I haven't had a chance to investigate here, let me
know if I should file a bug anyway.


On Tue, Mar 25, 2014 at 6:30 PM, Todd Fiala <tfiala at google.com> wrote:

> I'd suggest trying the change, running the tests, and see if anything new
> fails.
>
> And if this fixes currently broken behavior, add a test to make sure we
> don't regress it.  I'm pretty sure if it breaks something we'll know pretty
> quickly (either via bugs or our own usage).  At which point - we should add
> a test so we don't regress in the future and perhaps add a few comments as
> to the reason.
>
>
> On Tue, Mar 25, 2014 at 9:54 AM, <jingham at apple.com> wrote:
>
>> Why are you waiting for process groups?  That's not something we have to
>> do on Mac OS X.
>>
>> Jim
>>
>>
>> On Mar 25, 2014, at 1:17 AM, Andrew MacPherson <andrew.macp at gmail.com>
>> wrote:
>>
>> > Currently under Linux if you attach to a process whose process group id
>> is not equal to its process id (such as the child process of a fork() call)
>> the calls to waitpid() that pass -1*pid will return ECHILD since the pid
>> argument refers to a process group that doesn't exist. These calls occur in
>> Host::MonitorChildProcessThreadFunction() and the Linux ProcessMonitor.
>> >
>> > Changing -1*pid to simply -1 or to -1*getpgid(pid) resolves the issue
>> but it's not clear if this is the right fix as I'm unsure how other OSes
>> deal with this scenario.
>> >
>> > Any thoughts?
>> >
>> > Thanks,
>> > Andrew
>> > _______________________________________________
>> > lldb-dev mailing list
>> > lldb-dev at cs.uiuc.edu
>> > http://lists.cs.uiuc.edu/mailman/listinfo/lldb-dev
>>
>> _______________________________________________
>> lldb-dev mailing list
>> lldb-dev at cs.uiuc.edu
>> http://lists.cs.uiuc.edu/mailman/listinfo/lldb-dev
>>
>
>
>
> --
>  Todd Fiala | Software Engineer |  tfiala at google.com |  650-943-3180
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/lldb-dev/attachments/20140325/69fdf014/attachment.html>
-------------- next part --------------
A non-text attachment was scrubbed...
Name: waitpid-getpgid.patch
Type: text/x-diff
Size: 1239 bytes
Desc: not available
URL: <http://lists.llvm.org/pipermail/lldb-dev/attachments/20140325/69fdf014/attachment.patch>