Repository navigation
1.0rc1&2: test_eloc_attributes() fail on i386 #3496
Description
This is another separation of #3380. test_eloc_attributes fails on i386:
def test_eloc_attributes():
from .. import AltAz, ITRS, GCRS, EarthLocation
el = EarthLocation(lon=12.3*u.deg, lat=45.6*u.deg, height=1*u.km)
it = ITRS(representation.SphericalRepresentation(lon=12.3*u.deg, lat=45.6*u.deg, distance=1*u.km))
gc = GCRS(ra=12.3*u.deg, dec=45.6*u.deg, distance=6375*u.km)
el1 = AltAz(location=el).location
assert isinstance(el1, EarthLocation)
assert el1.latitude == el.latitude
assert el1.longitude == el.longitude
assert el1.height == el.height
el2 = AltAz(location=it).location
assert isinstance(el2, EarthLocation)
assert el2.latitude != it.spherical.lat
> assert el2.longitude != it.spherical.lon
E assert <Longitude [ 12.3] deg> != <Longitude 12.3 deg>
E + where <Longitude [ 12.3] deg> = <EarthLocation [(0.6836029704431764, 0.14904955169200265, 0.7144726796328034)] km>.longitude
E + and <Longitude 12.3 deg> = <SphericalRepresentation lon=12.3 deg, lat=45.6 deg, distance=1.0 km>.lon
E + where <SphericalRepresentation lon=12.3 deg, lat=45.6 deg, distance=1.0 km> = <ITRS Coordinate (obstime=J2000.000): x=0.683602970443 km, y=0.149049551692 km, z=0.714472679633 km>.spherical
astropy/coordinates/tests/test_frames.py:531: AssertionError
(Update: I git confused with my logs. arm64 is not affected)
Activity
[-]1.0rc1&2: test_eloc_attributes() fail on i386, arm64[/-][+]1.0rc1&2: test_eloc_attributes() fail on i386[/+]astrofrog commentedon Feb 14, 2015
@eteq - I looked into this a bit, and it turns out that I think the test may only be passing by luck on 64-bit systems. The difference between
el2.longitudeandit.spherical.lonis 1e-11 arcseconds.A side note is that the values returned by
el2are arrays, so I think it's more strictly correct to do:So I guess the question is, are the longitudes expected to be different?
Also, this test could do with more comments to explain why values should or should not be equal.
Update test_eloc_attributes to test correctly and also be explained
eteq commentedon Feb 16, 2015
#3507 should fix this - in fact, the logitudes should not be different, but the latitudes should. So #3507 changes to using
np.allclose(ornot np.allclose) and also changeslonto the correct behavior.@olebole - can you double-check that #3507 does indeed solve this on i386? (Or if someone else can test it on i386, that's fine too)
olebole commentedon Feb 16, 2015
I tested #3507 on i386, and it works.