debugging a race

[geoffleach]

I'm debugging the tests of a large package of modules that interface to an even-larger C++ package. It was going well until some previously-passing tests began to fail. I profiled a test in an attempt to locate the failure, and --surprise-- the test passed. From this I conclude that there's a race condition somewhere.

No threads.

Given that races tend to be highly context sensitive, I'm looking for hints.

Here's the test driver

use Audio::TagLib;

$proto = 'Audio::TagLib::ID3v2::Frame';

foreach $method (qw{toString}) {
    $return = $proto->can($method);
    print "$method: ", defined $return ? 'ok' : 'not ok', "\n";
}

Result is "not ok" . With -d:DProf its "ok"

My perl5 (revision 5 version 12 subversion 4) configuration:

osname=linux, osvers=2.6.32-131.2.1.el6.x86_64, archname=i386-linux-thread-multi

[Anonymous Monk]

Why do you believe it is a race condition (particularly since you're not using threads), rather than something else, such as an execution environment difference?

[sundialsvc4]

I agree.   This could be a red herring assumption.   Start walking back through the commits in the version control system until you return to a point where the tests stop failing.   Check-out that version to confirm that they have stopped.   Now, compare the test scripts to make sure that they haven’t changed, and look over the tests that they perform to be sure that they still make sense in terms of the changes that have happened.   Now examine the source-code changes.   (This is one reason for the rune, “commit often.”)

The git version control system even provides a bug-chasing “binary search” to help you zero in on “the commit” that caused a particular malfunction to surface.

Unfortunately, debugging environments of various sorts are very often very different environments from the regular execution environment.   Software often runs very, very differently ... which in my experience makes them not-so-useful except for microscopic troubleshooting.

[geoffleach]

Thanks for the two replies.

My assumption that there was a race somewhere was based on the fact that the problem went away when I ran the test under profiling.

Otherwise, I have a baseline on which the test I quoted runs correctly. I'm n the process of walking back module changes, but its painful, because failures are not consistent. By that I mean that the test will pass, and then somewhat later, under what appear to be the same conditions, will fail.