Steffen ,

I tested (briefly) the slave-driver branch. It failed some tests but
still appeared to work w/ the swarm service provided the plugin was
not enabled at startup - probably just an ordering issue between
pluginmanager and taskmanager. The effect was a SEGV when the slave
thread tried to ->post_event to it's service event.

Of the plugins and other code in padre/trunk/ only swarm is using
Padre::Service and the swarm audience is virtually non-existent at
present.

Perhaps you and I can workshop the idea of Service, certainly the
original implementation is brutal and the mechanism for communicating
with a service not well defined.

On Thu, Jan 28, 2010 at 9:08 PM, Steffen Mueller <[email protected]> wrote:
> Hi Pete,
>
> Peter Lavender wrote:
>>> There's this branch in branches/Padre-slave-driver that has the memory
>>> consumption/threading optimizations that Adam blogged about. It works
>>> well on my computer (linux), but it's a pretty devious and significant
>>> change. To the best of my knowledge, it's also only been tested by me
>>> despite my multiple suggestions on IRC that others should give it a
>>> shot. Now, I guess our development model or our developers mindset is
>>> such that it won't get a lot of testing time *unless* I just dump it
>>> into the trunk and let Murphy figure it out. I'm certain there'll be
>>> breakage.
>
>> Right.. it's a bit like that.. I had a patch waiting for ages for
>> someone to check out in the end I added it to trunk.  It was very simple
>> really, nothing like what you're doing, but it's the same end result..
>> it's not noticed until it hits trunk and then a release.
>
> It's not really surprising either. We're a volunteer group. I think all
> of us are in it for our own enjoyment. Testing branches is no fun.
>
>>> This suggests two possible routes:
>>> a) Make a release now, then merge the branch into trunk. Buys us time to
>>> fix things.
>>> b) Merge the branch into trunk and delay the release until we're certain
>>> we've fixed the most serious issues. Could make Padre 0.56 an even more
>>> significant release (cf. Adam's blog entry).
>>>
>>> I'd love to have this in the upcoming (hyped) 0.56 release. But it's
>>> quite risky and it might be better to stick with a). What do you guys think?
>>
>> Ok, I'll do a switch here and try it out myself.
>
> Thanks! The branch may be behind trunk by a few revisions (check svn
> info if in doubt).
>
>> I'm quite open on this matter myself.  If we want to wait that's fine.
>>
>> I'm quite a fan of least changes at once to make sure we can at least
>> see what's going on at the time  a major change comes in.
>
> Given that we're usually using the frequent Padre releases to whittle
> out bugs ("the problem manifested between 0.38 and 0.39." vs. the more
> usual "bisection tells me the problem manifested with change 10000") and
> that's probably a difficult to break habit (cf. comment about volunteer
> groups) it's probably better to push out a 0.56 release without my
> dubious changes.
>
>> In the recent past many of the 'big' changes we've seen to Padre has
>> come from Adam where incremental changes aren't overly an issue and I
>> know I've done releases with Adam's changes only partially applied, with
>> major gains seen up to 2 releases later.
>>
>> Given what I understand of the changes from Adams Blog, I will say that
>> making it for 0.56 make sense, but it means we need the time to do it.
>
> Yes. If we were to merge my branch to trunk, we'd have to delay 0.56 for
> sure. Anything else would be unfair to the beta testers^W^W users.
>
>> The only issue I see is that there are already a large number of changes
>> listed already, and Sebastian's concern with problems he's seeing means
>> merging now just to squeeze in before 0.56 might make sorting out any
>> issues harder than it needs to be.
>
> Agreed.
>
>> While it might hurt a little bit in terms of the "marketing" done for
>> 0.56, I think we could make 0.56 with what's already done, with the
>> merge to trunk after the release being your changes -  and two weeks to
>> get it ironed out, keeping major changes out until your changes are
>> stabilised.
>
> Adam's changes have impacted the apparent performance *a lot*. When I
> tried them for the first time, I really was amazed. Just as he's written
> in his blog, 0.56 is going to be *the* release to try out and to upgrade
> to. My changes are quite different if still important. They wouldn't
> really be visible to the users (modulo Scalars Leaked) unless they're
> very memory bottlenecked.
>
>> This gives us a fairly clear and visible progression with major changes
>> to Padre.
>
> Agreed!
>
> Best regards,
> Steffen
>
> PS: I'd particularly appreciate testing with Padre::Service related
> functionality. Not sure where that's used. For sure in ::Plugin::Swarm,
> though.
> _______________________________________________
> Padre-dev mailing list
> [email protected]
> http://mail.perlide.org/mailman/listinfo/padre-dev
>
_______________________________________________
Padre-dev mailing list
[email protected]
http://mail.perlide.org/mailman/listinfo/padre-dev

Reply via email to