Bug#795589: libmath-planepath-perl: FTBFS: Failed 1/103 test programs. 1/34754 subtests failed.

[Niko Tyni]

Control: reopen 795589

On Sat, Aug 15, 2015 at 05:26:44PM +0200, Salvatore Bonaccorso wrote:
> On Sat, Aug 15, 2015 at 04:53:19PM +0200, gregor herrmann wrote:
> > Control: tag -1 + unreproducible
> > On Sat, 15 Aug 2015 14:34:29 +0100, Chris Lamb wrote:
> > 
> > > Source: libmath-planepath-perl
> > > Version: 119-1
> > > Severity: serious
> > > Justification: fails to build from source
> > > 
> > > Dear Maintainer,
> > > 
> > > libmath-planepath-perl fails to build from source on unstable/amd64:

> > >   t/Base-Digits.t               (Wstat: 0 Tests: 80 Failed: 1)
> > >     Failed test:  48

> > Can someone reprocude it or does anybody have an idea?
> > It still builds fine for me (amd64 sid cowbuilder).
> 
> JFTR, as well here for me both with pbuilder (amd64, sid, cowbuilder)
> and with an sbuild chroot (but only done two testbuilds).

Works for me too, but the failure is reproducible (heh) on the
reproducible.debian.net CI system. Interesting. I'm reopening this for
now, hope that's OK.

I note that the failing logs have this extra line:

  # round_up_pow(5559060566555524,3) i=33 post 5559060566555523,16677181699666569 want 33,34

which comes from t/Base-Digits.t:198 and shows the failure point.

So for some reason the failing builds thinks that the next power of three
from 3^33 + 1 (which is 5559060566555524) is 3^33 (5559060566555523)
rather than 3^34 (16677181699666569). This is clearly wrong.

I see that the Changes file lists round_up_pow() as a new feature,
so it's plausible that there's a problem with it. Maybe a CPU-specific
rounding difference in float handling? reproducible.debian.net lists
the CPU type as 'AMD Opteron 62xx class CPU' FWIW.

I'm Cc'ing the upstream author, Kevin Ryde, who's probably interested.
A current log of a failing build can be found at
 https://reproducible.debian.net/rb-pkg/unstable/amd64/libmath-planepath-perl.html

I'll see if I can arrange access to a host where the failure is reproducible.
-- 
Niko Tyni   ntyni at debian.org