[PATCH] [test-suite] Fix failure to capture stderr in *.out-simple's

Daniel Sanders Daniel.Sanders at imgtec.com
Fri Dec 19 05:03:47 PST 2014


> -----Original Message-----
> From: Daniel Sanders
> Sent: 18 December 2014 22:09
> To: 'Tom Stellard'
> Cc: llvm-commits at cs.uiuc.edu
> Subject: RE: [PATCH] [test-suite] Fix failure to capture stderr in *.out-simple's
> 
> > -----Original Message-----
> > From: Tom Stellard [mailto:tom at stellard.net]
> > Sent: 18 December 2014 17:54
> > To: Daniel Sanders
> > Cc: llvm-commits at cs.uiuc.edu
> > Subject: Re: [PATCH] [test-suite] Fix failure to capture stderr in *.out-simple's
> >
> > On Thu, Dec 18, 2014 at 11:45:09AM +0000, Daniel Sanders wrote:
> > > I'm experiencing the problem on my x86_64 Debian Jessie system while
> > cross-compiling to Mips and running the resulting binaries under qemu. perf
> is
> > run on the host system so I believe it affects everyone with a sufficiently new
> > perf.
> > >
> > > I haven't been able to check whether the same problem affects mips/mipsel
> > Debian Jessie systems because the new perf isn't compatible with the kernels
> I
> > currently have to use for these systems. The test-suite falls back on ctimeit
> for
> > these systems.
> > >
> >
> > It looks like r218054 may solve this issue by make perf non-default
> > on Linux.  Does this solve your issue?  If so maybe we can merge this
> > commit instead of the patch.
> >
> > -Tom
> 
> I'll give it a try overnight to be sure but I think it will fix the problem too. It also
> corrects the usage of perf in the same way that my patch does so I believe it
> will fix the USE_PERF case too.

Yes, it works for me. Given that you suggested this merge, I've assumed your approval and merged it in r224573.

> > > > -----Original Message-----
> > > > From: Tom Stellard [mailto:tom at stellard.net]
> > > > Sent: 17 December 2014 20:13
> > > > To: Daniel Sanders
> > > > Cc: llvm-commits at cs.uiuc.edu
> > > > Subject: Re: [PATCH] [test-suite] Fix failure to capture stderr in *.out-
> > simple's
> > > >
> > > > On Fri, Dec 12, 2014 at 10:39:48PM +0000, Daniel Sanders wrote:
> > > > > A very late ping. I'd completely forgotten about this patch but I just ran
> > into
> > > > the problem again while testing LLVM 3.5.1 (cross-compilation from
> x86_64
> > to
> > > > mips) with clean test-suite sources.
> > > >
> > > > How critical is this fix for 3.5.1?  Will the test-suite fail on new Debian
> > > > for everyone, or is it just a Mips issue?
> > > >
> > > > -Tom
> > > >
> > > > > ________________________________
> > > > > From: Daniel Sanders
> > > > > Sent: 15 August 2014 15:24
> > > > > To: llvm-commits at cs.uiuc.edu
> > > > > Subject: [PATCH] [test-suite] Fix failure to capture stderr in *.out-
> simple's
> > > > >
> > > > > Hi,
> > > > >
> > > > > I've just dist-upgraded one of my Debian machines and I'm now finding
> > that a
> > > > large number of tests are failing the reference check on that machine. All
> > the
> > > > missing lines appear in test.log so it seems that stderr is no longer
> > redirected to
> > > > the *.out-simple along with stdout. This patch fixes this by redirecting
> > stderr to
> > > > stdout on the perf command.
> > > > >
> > > > > Daniel Sanders
> > > > > Leading Software Design Engineer, MIPS Processor IP
> > > > > Imagination Technologies Limited
> > > > > www.imgtec.com<http://www.imgtec.com/>
> > > > >
> > > >
> > > > > Index: tools/timeit.sh
> > > > >
> > > >
> >
> ================================================================
> > > > ===
> > > > > --- tools/timeit.sh	(revision 213378)
> > > > > +++ tools/timeit.sh	(working copy)
> > > > > @@ -26,7 +26,7 @@
> > > > >  	shift 2
> > > > >  done
> > > > >
> > > > > -perf stat -o stats $@ < $INPUT > $OUTPUT
> > > > > +perf stat -o stats $@ < $INPUT > $OUTPUT 2>&1
> > > > >
> > > > >  EXITCODE=$?
> > > > >
> > >




More information about the llvm-commits mailing list