Re: Two other packages are now failing their debci tests on i386 (Was: Bug#1024012: [joshuaulrich/xts] Unit test failure when running on i386 (Issue #384))

[Adrian Bunk <bunk@debian.org>]

On Tue, Jan 24, 2023 at 08:29:42AM +0100, Andreas Tille wrote:
> Am Sun, Jan 22, 2023 at 07:20:34PM +0200 schrieb Adrian Bunk:
> > >  > pp <- profile(fit2,prof.lower=-0.2)
> > > +Warning message:
> > > +In mle2(minuslogl = function (a = NULL, b = NULL)  :
> > > +  convergence failure: code=52 (ERROR: ABNORMAL_TERMINATION_IN_LNSRCH)
> > >  > stopifnot(min(subset(as.data.frame(pp),param=="b")$par.vals.b)==-0.2)
> > >  > ## note that b does go below -0.2 when profiling a ...
> > >  > options(old_opt)
> > 
> > I don't have a better idea for that than sending you a MR to disable 
> > this test on i386.
> 
> I can reproduce this one in a i386 chroot and I'm working on a patch
> that disables this test.
>  
> > Ubuntu has problems with a different test in this package on ppc64el 
> > where they are building with -O3,[1] so this package seems to have
> > fragile tests.
> 
> I can include this patch as well.

The Ubuntu diff followed by

if [ "$hostarch" = "i386" ] ; then
    rm -f profbound.R
fi

works for me.

> > > and r-cran-rpf[3] failing with
> > > 
> > > ══ Failed tests ════════════════════════════════════════════════════════════════
> > > ── Failure ('test-fit.R:116'): fit w/ mcar ─────────────────────────────────────
> > > `stat` not equal to `Estat`.
> > > 1/20 mismatches
> > > [5] 4.42 - 6.69 == -2.27
> > > 
> > > [ FAIL 1 | WARN 0 | SKIP 0 | PASS 381 ]
> > > 
> > > I have no idea what to do sensibly (except patching out the said tests).
> > 
> > This might end up being similar, but there is some weirdness going on here.
> > I'll let you know if patching is needed.
> 
> I can't reproduce this one in my i386 pbuilder chroot.

It fails when testing the package in unstable/testing, but passes after 
recompiling that one with the new r-base.

> Kind regards
>     Andreas. 
>...

cu
Adrian