Profiling parallel tests

[Eric Wilhelm]

# from Andy Armstrong
# on Friday 22 August 2008 06:39:

>>> Correct. But I think it's worth clearing out the structures that
>>> relate to tests that have completed too - otherwise get_any() still
>>> has a lot of corpses to step over.
>>
>> There seems to already be some pruning in _find_next_job():
>>
>>        # Prune any exhausted items.
>>        shift @$seq while @$seq && _is_empty( $seq->[0] );
>>
>> That's doing something different from what you're suggesting?
>
>Dunno :)

Yes you do ;-)  This is what we were talking about on irc.

The @$rules contains a lot of $seq = [...] items.  When a $seq is done, 
it becomes [] and gets left in @$rules.

So, Nicholas, what you have there is the only pruning (and also the 
cause of the answer from get_all() changing.)  But this means that 
get_all() is continually checking the pile of []'s.

I think the $self->_is_done() thing could replace the get_all() in 
the "spinner if($self->get_all())" bit, so long as you set the done 
flag when you find the last empty sequence.  At that point, having 
these $seq = [] corpses might not matter so much.

--Eric
-- 
"I've often gotten the feeling that the only people who have learned
from computer assisted instruction are the authors."
--Ben Schneiderman
---------------------------------------------------
    http://scratchcomputing.com
---------------------------------------------------