[svn.haxx.se] · SVN Dev · SVN Users · SVN Org · TSVN Dev · TSVN Users · Subclipse Dev · Subclipse Users · this month's index

Re: Ruby test failure on 1.6.x

From: Julian Foad <julianfoad_at_btopenworld.com>
Date: Tue, 12 Jan 2010 18:43:50 +0000

Philip Martin wrote:
> I build with pool debugging enabled. I have run the tests under
> valgrind as follows:
>
> (cd subversion/bindings/swig/ruby && valgrind ruby -I ../../../../../src-1.6/subversion/bindings/swig/ruby ../../../../../src-1.6/subversion/bindings/swig/ruby/test/run-test.rb test_delta.rb)
>
> I get a variety of valgrind assertions during the initialisation
> phase, but during the tests I get:
>
> ==19064== Use of uninitialised value of size 8
> ==19064== at 0x4C6CB78: (within /usr/lib/libruby1.8.so.1.8.7)
> ==19064== by 0x4C6C69E: (within /usr/lib/libruby1.8.so.1.8.7)
> ==19064== by 0x4C6CF46: (within /usr/lib/libruby1.8.so.1.8.7)
> ==19064== by 0x4C6D8FA: rb_gc (in /usr/lib/libruby1.8.so.1.8.7)
> ==19064== by 0x4C6D918: rb_gc_start (in /usr/lib/libruby1.8.so.1.8.7)
> ==19064== by 0x4C586C1: (within /usr/lib/libruby1.8.so.1.8.7)
> ==19064== by 0x4C58829: (within /usr/lib/libruby1.8.so.1.8.7)
> ==19064== by 0x4C527DF: (within /usr/lib/libruby1.8.so.1.8.7)
> ==19064== by 0x4C569BA: (within /usr/lib/libruby1.8.so.1.8.7)
> ==19064== by 0x4C52BD7: (within /usr/lib/libruby1.8.so.1.8.7)
> ==19064== by 0x4C55A9E: (within /usr/lib/libruby1.8.so.1.8.7)
> ==19064== by 0x4C58577: (within /usr/lib/libruby1.8.so.1.8.7)
> ...........
> Finished in 34.863395 seconds.
>
> 11 tests, 49 assertions, 0 failures, 0 errors
>
> The error occurs in test_apply which I assume is the sixth dot in that
> list; there are no valgrind assertions associated with that particular
> test.

Philip, please could you also run in valgring the "test_info.rb" test
file, because it is in that file that the two errors occurred that were
previously reported (on two different machine architectures), and I
don't think we did anything in particular to fix them.

I tried this just now, and don't see any output specific to that
"test_info(SvnInfoTest)" test.

I compile Subversion with "-DAPR_POOL_DEBUG", but I recall that may not
be as good as it would be if I also compiled APR the same way.

Peter Samuelson said on IRC just now:

"julianf: you're correct that the failures I reported were on
kfreebsd-i386 and sparc, but I've also seen them in older Debian build
daemon logs for other architectures. I believe I've seen one or both on
powerpc, armel (little-endian arm), alpha. it is too bad I can't get
anyone to reliably reproduce them on non-automated builds.
<peterS> one of joeswatosh's recent comments makes me wonder if these
failures would come forward more readily with, for example, apr pool
debugging"

- Julian
Received on 2010-01-12 19:44:30 CET

This is an archived mail posted to the Subversion Dev mailing list.