astropy: test_erfa_planet fails on i386
On 32 bit x86, TestPositionKittPeak.test_erfa_planet fails by newer returning.
I could locate the loop in this loop in solar_system.py. The absolute value of delta_light_travel_time here jumps between 7.91119372419e-11 AU s / m, 7.91011123661e-11 AU s / m, and 1.08248758387e-14 AU s / m.
Python version is 2.7.12, erfa 1.3.0. This didn’t happen before (with Python 2.7.12rc1, erfa 1.2.0).
Full build log here
Unfortunately, I have no tests in between, so it may be anything… I also just had another mystic regression on i386 (with the CPL library), so it is not excluded that it is caused by the C compiler (gcc-6.2.0) or so. But if anyone could have a look to the mentioned loop and could give some advise?
About this issue
- Original URL
- State: closed
- Created 8 years ago
- Comments: 22 (21 by maintainers)
Indeed, it has been a bit sad that we even had to go the erfa route in the first place, instead of realising that it was good that people wanted to use the code and that that would mean bugs would be found. But in fairness, I don’t think the people who work on IAU committees get all that much credit from the community for their often thankless tasks. On a more positive note, perhaps we should specifically include feedback upstream in the astropy paper 2…
Yea, they are not really carefully acknowledging where they got the solution from 😦 The patch was done in liberfa/erfa#41 and sent to upstream later, and one would usually expect that this origin is mentioned somewhere in SOFA (Announcement, Changelog,
THANKSfile or whatever). But IAU is different…