chromatic wrote:
On Friday 26 October 2007 13:05:14 Tom Heady wrote:

The method does require a single file per class, and loading perl for each of those files. If you are trying to avoid that it's not going to
help.

... and if that's the slowest part of 45-minute test runs, color me surprised.

Well, *my* test suite was 45 minutes. I think Jonathans is just a couple minutes. So depending on how many classes this could add several seconds. It'd be an interesting thing to benchmark though. Jonathan, are you sure it's worth your troubles and that you're really saving any significant amount of time?


Yes, it was about time to benchmark it.

In a single process:

real    1m4.999s
user    0m14.320s
sys     0m2.290s

In separate processes:

real    2m7.747s
user    0m58.960s
sys     0m8.570s

So I'd say that it makes a noticeable difference for tests run manually (and awaited upon), but not so much for background automated smoke tests.

This is for thirty test classes, so there's about 2 seconds of load overhead per class. I've made my Test::Class subclass fairly rich in terms of utilities and automatically loaded modules, so that the individual test classes don't need to worry about use'ing this and that.

Jon

Reply via email to