Slightly different gene structures produced on i386

[satta]

Hi,

I am investigating a build failure on Debian sid i386 (see https://buildd.debian.org/status/fetch.php?pkg=genomethreader&arch=i386&ver=1.7.3%2Bdfsg-2&stamp=1579736664&raw=0), the only failure across all supported architectures (see https://buildd.debian.org/status/package.php?p=genomethreader).
The reason for this failure is that apparently this 32-bit version outputs two exons in the U89959 test case differently. I confirmed that this issue also appears in the binary release version downloaded from the official GenomeThreader site by copying the testdata and testsuite directories from the GenomeThreader source into the extracted binary distribution directory and running testsuite.rb:

(sid-i386)root@debian:/tmp/gth-1.7.3-Linux_i386-32bit/testsuite# ./testsuite.rb
  1: gth                                                         : ok
  2: gth -help                                                   : ok
  3: gth -help+                                                  : ok
  4: gth -version                                                : ok
  5: gth regression test (assertion in gthsafilter)              : ok
  6: gth regression test (-gff3out -intermediate)                : ok
  7: gth regression test (-gff3out -intermediate -fastdp)        : ok
  8: gth regression test (-gff3out -skipalignmentout)            : ok
  9: gth regression test (-gff3out -skipalignmentout -fastdp)    : ok
[...]
 50: fastdp (U89959)                                             : failed
     [ problem: unexpected return code: 1 != 0
       in: /tmp/gth-1.7.3-Linux_i386-32bit/testsuite/stest_testsuite/test50 ]
 51: fastdp (U89959, introncutout)                               : failed
     [ problem: unexpected return code: 1 != 0
       in: /tmp/gth-1.7.3-Linux_i386-32bit/testsuite/stest_testsuite/test51 ]
[...]

(sid-i386)root@debian:/tmp/gth-1.7.3-Linux_i386-32bit/testsuite# cat /tmp/gth-1.7.3-Linux_i386-32bit/testsuite/stest_testsuite/test50/stdout_2
1333,1336c1333,1336
< 1877523	gth	exon	105010	105218	0.861	+	.	Parent=gene174
< 1877523	gth	five_prime_cis_splice_site	105219	105220	0	+	.	Parent=gene174
< 1877523	gth	three_prime_cis_splice_site	105303	105304	0	+	.	Parent=gene174
< 1877523	gth	exon	105305	105428	0.887	+	.	Parent=gene174
---
> 1877523	gth	exon	105010	105223	0.86	+	.	Parent=gene174
> 1877523	gth	five_prime_cis_splice_site	105224	105225	0	+	.	Parent=gene174
> 1877523	gth	three_prime_cis_splice_site	105308	105309	0	+	.	Parent=gene174
> 1877523	gth	exon	105310	105428	0.891	+	.	Parent=gene174
1345,1348c1345,1348
< 1877523	gth	exon	105009	105218	0.862	+	.	Parent=gene175
< 1877523	gth	five_prime_cis_splice_site	105219	105220	0	+	.	Parent=gene175
< 1877523	gth	three_prime_cis_splice_site	105303	105304	0	+	.	Parent=gene175
< 1877523	gth	exon	105305	105428	0.887	+	.	Parent=gene175
---
> 1877523	gth	exon	105009	105223	0.86	+	.	Parent=gene175
> 1877523	gth	five_prime_cis_splice_site	105224	105225	0	+	.	Parent=gene175
> 1877523	gth	three_prime_cis_splice_site	105308	105309	0	+	.	Parent=gene175
> 1877523	gth	exon	105310	105428	0.891	+	.	Parent=gene175

These tests were run in an i386 Debian sid chroot.

Is this a bug or is this deviation known or acceptable?

[gordon]

Unfortunately this deviation is known and we have to live with it. It is an artifact of floating point operations working slightly differently on some architectures and in some rare cases this can lead to different results.

I spent a lot of time trying to fix that back in the day, but it doesn't seem to be possible, these are rounding errors.

I would just disable these specific tests on this platform.

[satta]

Okay, thanks for the information! I will skip these tests on i386 then.