develooper Front page | perl.perl5.porters | Postings from September 2011

Re: Some preliminary cpan smoking reports

Thread Previous | Thread Next
Steffen Mueller
September 4, 2011 23:25
Re: Some preliminary cpan smoking reports
Message ID:
On 09/04/2011 11:32 PM, Chip Salzenberg wrote:
> On Sun, Sep 4, 2011 at 1:26 PM, Reverend Chip <
> <>> wrote:
>  > There may be something else unexplained going on.  I just did a
>  > perlbrew-like localized installation of magicflags1 branch, and used
>  > cpanm to build Moose 2.0204, and it compiled and passed its self tests,
>  > which I'm happy about of course.  But the smoke report suggests that
>  > should not have worked.
> Actually the smoke test said that 2.0004 is the one that failed, not
> 2.0204.  I have reproduced the failure.  But the failure seems unrelated
> to magicflags1; it's in t/basics/require_superclasses.t and has to do
> with version satisfaction.  And 2.0204 does work with magicflags1.  So I
> think it's a non-issue for both the patch and the smoke process (except
> insofar as the smoke needs a bit of a CPAN index update :-)).

Yeah, can't update your minicpan mirror while running a smoke of two 
different perls. They'd just be testing different versions of 
distributions randomly.

I've now restarted the smoke and am using the newly implemented 
parallelism hacks to put a lot higher load on poor dromedary. From last 
night, there are now already 4000 reports (total, across two perls). 
That means we might be done by the end of the working week if all goes well.

I have not updated the minicpan mirror. This was on purpose: If I had 
done that, I couldn't possibly go back and finish the other smoke if 
deemed useful. And having another minicpan is currently too wasteful of 
scratch disk. It's more important to get this done so that Dennis can do 
his admin thing with dromedary.

Furthermore, restarting the smokers turns out to not be free: The 
smokers will have to rebuild the distributions that are commonly 
required as dependencies. This is compounded by the fact that I am now 
running six* independent smokers in parallel, all of which have a 
separate build directory by necessity. Alas, I'll keep an eye on it.

Back on a train, so don't trust anything I say or do.


* Yes. On a quad core machine. That's also on purpose. There are quite a 
lot of distributions that just waste wallclock time waiting for input 
that never comes. While I blacklisted a lot of them thanks to George's 
advice and Andreas' work, I assume there's still plenty to make the 
over-commitment a net win.

Thread Previous | Thread Next Perl Programming lists via nntp and http.
Comments to Ask Bjørn Hansen at | Group listing | About