[Debian-med-packaging] Bug#881149: htslib FTBFS: test failures on i386

[Adrian Bunk]

On Wed, Nov 08, 2017 at 10:25:31AM +0200, Graham Inggs wrote:
> Package: htslib
> Version: 1.5-1
> Severity: serious
> Tags: buster sid
> 
> As per logs from reproducible builds [1], htslib 1.5-1 recently
> (around 2017-09-25) started to FTBFS on i386 with the following test
> failures:

The relevant change is gcc 6 -> 7.

> test_vcf_api:
> /build/1st/htslib-1.5/test/test-vcf-api /tmp/o6gEIzeYrS/test-vcf-api.bcf
> bcf_get_format_float didn't produce the expected output.
> 
> .. failed ...
> 
> test_vcf_sweep:
> /build/1st/htslib-1.5/test/test-vcf-sweep /tmp/o6gEIzeYrS/test-vcf-api.bcf
> 
> The outputs differ:
> /build/1st/htslib-1.5/test/test-vcf-sweep.out
> /build/1st/htslib-1.5/test/test-vcf-sweep.out.new
> .. failed ...
>...

There has been some amount of FTBFS due to gcc 7 seemingly hitting not 
standards performance behaviour of x87 more frequently than previous
gcc versions.

It is actually not clear that the code compiled with older gcc works 
always correctly, it might just not hit problems in the tests.

Some time ago I read through the upstream bug for the likely related #819617.

With multiarch allowing installation of amd64 packages even on an i386
installation, I'm wondering whether we would lose anything by just doing
an RM of htslib and rdeps on i386 - that would be the cleanest solution.

cu
Adrian

-- 

       "Is there not promise of rain?" Ling Tan asked suddenly out
        of the darkness. There had been need of rain for many days.
       "Only a promise," Lao Er said.
                                       Pearl S. Buck - Dragon Seed