2.10: test error on big endian systems

[olebole]

The version 2.10 fails on one test when compiled on big endian systems like PowerPC 64 bit or IBM zSystems with :

FAIL: test_integration01.py
===========================


> WARNING: This executable has been compiled using a version of the ATLAS library without support for multithreading. Performance will be degraded.

sh: 1: convert: not found

> WARNING: FLAGS parameter not found in catalog GAIA-DR1_1334+3754_r46.cat


*** PLPLOT WARNING ***
Driver does not support hardware solid fills, switching to software fill.

Running Test #1 ...
SCAMP execution          ... passed
Minimum AS contrast: 44.1086 expected: > 22.00 ... passed
Minimum XY contrast: 48.8126 expected: > 22.90 ... passed
# of matched detections: 189983 expected: > 180100 ... passed
# of matched references: 2226 expected: > 1640 ... passed
Internal dispersion: 0.0354290517341998 expected: < 0.035 ... failed
Reference dispersion: 0.09366372231264102 expected: < 0.097 ... passed
1 errors.
FAIL test_integration01.py (exit status: 1)

Full build logs here for S390x (official Debian architecture), and here for PowerPC 64 bit (additional port).
I could ofcourse just adjust the limit for the internal dispersion, but I am afraid that this may hide a real problem.

[ebertin]

Well actually all the numbers are different from the ones I get on x86 systems but I don't have presently direct access to a big-endian system. What would you suggest for me to have access to a big-endian system (outside of the Debian buildchain)? Thanks.

[olebole]

I have no good idea to to get you access. I can however access s390x directly myselt, if that would help.
Also, the issue is a regression and did not appear on 2.9.

[ebertin]

OK, most likely a consequence of the change in the way endianity is handled then. Thanks, I'll take it from there.