[commit] Re: testcase fix ping: Re: gdb.base/info-os.exp FAILs/ERRORs [Re: [commit] 'info os' additions]

[Stan Shebs <stanshebs at earthlink dot net>]

On 6/24/12 11:37 AM, Jan Kratochvil wrote:
> On Sat, 12 May 2012 22:18:20 +0200, Stan Shebs wrote:
> > On 5/12/12 8:40 AM, Jan Kratochvil wrote:
> > > On Sat, 12 May 2012 03:32:12 +0200, Stan Shebs wrote:
> > > (gdb) FAIL: gdb.base/info-os.exp: get shared-memory regions
> > >
> > > as apparently the .c testfile creates constant resources and expects they are
> > > free, therefore multiple runs clash.  There can be even multiple runs by
> > > different users.  This is why for example gdbserver_start does:
> > >       # Loop till we find a free port.
> > > [...]
> > >                       incr portnum
> > >
> > > Could you fix this part?
> > Huh, didn't think about parallelization!  I'll fix it up.
> This is still flipping FAIL/PASS, I have setup additional local ignore rule
> for it but it would be nice to have it fixed for gdb-7.5.

OK, here's the fix I've just committed.  Tested by running several make 
checks in parallel from a script - without this patch it's just about 
guaranteed one or more of the runs will fail.

Just as a caveat, although the test program cleans up after itself, I 
notice that the keys used seem to gradually get larger with successive 
runs, but not always monotonically.  It could be that the recovery 
process is slow, or perhaps a kernel bug - the one I'm working with is a 
little old.  If these tests start failing due to hitting the retry 
limit, we may want to investigate further.

Stan
stan@codesourcery.com

2012-07-03  Stan Shebs  <stan@codesourcery.com>

    * gdb.base/info-os.c (main): Retry resource acquisition until an
    available one is found.
    * gdb.base/info-os.exp: Collect resource keys from the program
    and use them in matching.

Attachment: ospar-patch-1
Description: Text document