perl-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Geoffrey Young <>
Subject Re: can't deal with ithreads, need Test::More
Date Mon, 28 Mar 2005 00:02:17 GMT

Stas Bekman wrote:
> As we start adding tests running in ithreads, we no longer can use
>, which can't deal with the test counter. when several threads
> print 'ok 1' we get:
> ok 136
> Test output counter mismatch [test 277]
> since in each thread,'s counter has a different value.
> Test::More has it right (via Test::Builder). So one possibility is for
> mp2 test suite to switch to Test::More. The problems with that approach is:
> 1) syntax of ok and skip aren't 100% compatible between Test::More and
> Test (but we can deal with that by adjusting our tests where needed).
> Prototype mismatch: sub Apache::Test::ok ($;$$) vs ($;$) at
> /home/stas/ line 75
> Prototype mismatch: sub Apache::Test::skip ($;$$$) vs none at
> /home/stas/ line 75

  use Apache::Test qw(:withtestmore);

should remedy this.

> 2) Test::More relies on Test::Builder, I see that some important fixes
> due to which we have to switch to T-M require Test-Builder 0.47 (Aug
> 2002), but looking up the changes list, it seems that many more ithreads
> related bugs were fixed in the very recent versions. So it's possible
> that we need at least 0.51 Nov 2004). 

fwiw, 0.49 is required under persistent environments (eg server-side tests),
but I guess you're doing this *.t so you're ok.

> That means that the test suite now
> will require a prerequisite which most people won't have, making the
> source build out of box w/o fetching some requirements impossible.

unless you mean dropping support in Apache-Test, it doesn't mean
this at all.  you can require Test::More from all.t for just the thread tests.

> And
> skipping those tests would suck, since we really need those threads
> things to be tested.

well, we can test them ourselves as best we can, but skipping tests on user
platforms isn't all that bad.  a large percentage of people skip them
routinely anyway, or don't care when they fail.  or at least I suspect as much.

> Another possibility is to try to hack and possible override
> things, but it's not a good idea, since it will make things too fragile
> if internals change later on.

hacking shouldn't be required.

> A related issue: A-T's -withtestmore breaks the importing mechanism.
> It's no longer possible to load A-T w/o importing all the symbols, i.e.:
> use Apache::Test (-withtestmore);

well, that's a feature, since that action is supposed to load all the
symbols that don't collide with Test::More.  but why do you _not_ want to
load them all?

> forces the import of all. It probably should be moved to
> Apache::TestMore instead.

I'll think about that a bit and see if it can work the way I want it to -
I've been using -withtestmore for about a year now and it seems to meet all
my needs without issue.


To unsubscribe, e-mail:
For additional commands, e-mail:

View raw message