Bug#599992: perl: FTBFS on mips: lib/Memoize/t/speed FAILED at test 2

[Niko Tyni]

found 599992 5.10.1-14
severity 599992 important
thanks

On Tue, Oct 12, 2010 at 11:46:08PM +0300, Niko Tyni wrote:
> On Tue, Oct 12, 2010 at 05:30:34PM +0200, Cyril Brulebois wrote:
> > Source: perl
> > Version: 5.10.1-15
> > Severity: serious
> > Justification: FTBFS
> 
> > your package no longer builds on mips:
> > | lib/Memoize/t/speed...........................................FAILED at test 2
> 
> I doubt Memoize is really broken. There certainly weren't any related
> changes between 5.10.1-14 and -15. Could you please give it back so we
> see if that succeeds?

It did succeed.

> This test is timing sensitive by nature: it calculates Fibonacci numbers
> recursively until the calculation of the next one takes more than ten
> seconds, then checks that a memoized run takes less than one tenth
> of that.
> 
> It seems conceivable that a load spike on a slow machine could make the
> memoized run (which itself is trivial CPU-wise) to take two seconds,
> which would cause the failure.

Non-deterministic uncommon build failures haven't been considered release
critical in the past, so downgrading the severity. Even if the release
team thinks otherwise, this shouldn't block the -15 testing transition
as -14 has the same issue.

I think the probability of this failure mode could be made smaller
by raising the $LONG_RUN value in lib/Memoize/t/speed.t (currently 10
seconds) at the cost of longer build times.
-- 
Niko Tyni   ntyni at debian.org