CPANPLUS::Dist::Build 0.44 considered harmful
Andreas J. Koenig
producing bad reports.
Several of my statistics indicate that CPANPLUS::Dist::Build=0.44 never
produces PASS reports and instead produces many false FAIL reports.
Here a few samples:
http://pause.perl.org:3000/reports_by_field?distv=ack-1.92;field=mod%3ACPANPLUS%3A%3ADist%3A%3ABuild
http://pause.perl.org:3000/solved?distv=Business-OnlinePayment-SagePay-0.10
It seems that I have identified about 40 distros that each got more than
10 false FAIL reports.
Please turn off your smokers if you're using CP:D:B 0.44
Let me know if you need more data.
Thanks,
--
andreas
pe...@0ne.us
Thanks for warning us about this! I quickly looked at some of the
reports and I can tell you with confidence that a lot of my FAIL reports
is bad. I don't think it's CP:D:B's fault but the VM's fault. It seems
like their local time is skewed/slowed/whatever and it causes confusion
when running the makefile. Here's one sample report:
http://www.nntp.perl.org/group/perl.cpan.testers/2009/12/msg6478612.html
I've now turned off the smoker on those VMs and is investigating the
time problem. I'm using NetBSD 5.0.1 and FreeBSD 7.2-RELEASE on a
VirtualBox 3.0.12 host running Ubuntu 9.10. If anyone have any tips,
please let me know!
P.S. Great use of your new regression analyzer! Big kudos to you :)
Chris 'BinGOs' Williams
> Hello,
>
> Thanks for warning us about this! I quickly looked at some of the
> reports and I can tell you with confidence that a lot of my FAIL reports
> is bad. I don't think it's CP:D:B's fault but the VM's fault. It seems
> like their local time is skewed/slowed/whatever and it causes confusion
> when running the makefile. Here's one sample report:
>
> http://www.nntp.perl.org/group/perl.cpan.testers/2009/12/msg6478612.html
>
> I've now turned off the smoker on those VMs and is investigating the
> time problem. I'm using NetBSD 5.0.1 and FreeBSD 7.2-RELEASE on a
> VirtualBox 3.0.12 host running Ubuntu 9.10. If anyone have any tips,
> please let me know!
>
> P.S. Great use of your new regression analyzer! Big kudos to you :)
Using Vmware Server 2.0 on Ubuntu 9.04 ( and before that on Windows Server 2003 )
I noticed the time skew problems.
I run ntpd on the vms which tends to resolve the issue.
FreeBSD though is a bit more flunky. Even with ntpd running it generally skews out
and ntpd will give up on trying to sync.
On FreeBSD vms I disable ntpd and set a cron job as root to run every 5 minutes to
execute "/etc/rc.d/ntpdate start" ( Actually, I have this wrapped in a shell script so
I can redirect the stdout/stderr from that command to /dev/null, so that root doesn't
get an email from cron every 5 minutes >:) ).
Regarding the email subject, as far as I can tell from the reports I looked through they
were all relating to time issues with Makefile, CP::D::Build wouldn't be involved
CP::D::MM would.
Cheers,
--
Chris Williams
aka BinGOs
PGP ID 0x4658671F
http://www.gumbynet.org.uk
==========================
Andreas J. Koenig
> Regarding the email subject, as far as I can tell from the reports I looked through they
> were all relating to time issues with Makefile, CP::D::Build wouldn't be involved
> CP::D::MM would.
Sorry for the bad subject line and thank you for the fine grained
details.
Do you have suggestions how we can prevent this from happening again?
Are the timestamps of the untarred distributions in the future on such
systems? Presumably this would play an ugly game also on CPAN.pm based
smokers, right?
--
andreas
Graham TerMarsch
> Evidence is rather strong to me that CPANPLUS::Dist::Build 0.44 is
> producing bad reports.
>
> Several of my statistics indicate that CPANPLUS::Dist::Build=0.44 never
> produces PASS reports and instead produces many false FAIL reports.
>
> matter-JUnit-0.03&field=mod%3ACPANPLUS&field=mod%3ACPANPLUS%3A%3ADist%3A%3A
> Build&field=meta%3Afrom&field=mod%3ACPANPLUS%3A%3ADist%3A%3ABuild
I can't speak for the others, but for TAP::Formatter::JUnit, its apparent that
the smoker isn't installing any of the dependencies listed in META.yml. It
lists them all as "ERROR: <dep> is not installed" but then marches ahead
nonetheless and (obviously) the build fails.
--
Graham TerMarsch
pe...@0ne.us
hour and fire off the smokers for new dists. My FreeBSD box after a few
days would lag terribly, losing several minutes per hour! The skew got
so bad, what you explained happened - untarred distros were in the
future. I assume it would affect any toolchain modules trying to
install/smoke them?
I wonder how it would affect a "normal" CPAN user trying to install
the modules on a system? Is it possible for EU::MM/CP:D:MM/friends to
automatically detect this "funkiness" and re-run the makefile, so
nothing will blow up if this situation happens on a server? I'm willing
to test this theory and investigate possible ways to patch the toolchain
to do this, but only if somebody tells me it's feasible :)
I've been busy having christmas potlucks with friends/family, and
once I have some free time I'll get NTPd running on the boxes and watch
them closely for any clock skew issues. Maybe I can add a filter to my
outbound mailer to discard reports if it sees the "makefile out of date"
message and send alerts to me... :)
Thanks again!