Added failing example for test_pix2ang

[astrofrog]

@lpsinger - #80 is not related to PPC, it is just a failure that happened to appear in that build. This PR adds that specific example to the hypothesis test and this fails for me locally on Mac. I don't have time to investigate why this is happening, so if anyone has any time please feel free to!

[astrofrog]

This fails on Travis, as expected (I think healpy doesn't get installed on AppVeyor)

[cdeil]

Here's a small example to reproduce the issue:

import numpy as np
import astropy_healpix

nside = 2 ** 27
frac = 2 / 3
ipix = int(frac * 12 * nside ** 2)
healpix = astropy_healpix.HEALPix(nside, order='nested')
lon, lat = healpix.healpix_to_lonlat(ipix)
phi = lon.rad
theta = np.pi / 2 - lat.rad

print('nside:', nside)
print('ipix=', ipix)
print('lat.dtype=', lat.dtype)
print('phi=', phi)
print('lat.deg=', lat.deg)
print('distance_to_south_pole=', np.pi - theta)

The problem is that this pixel is placed directly at the south pole:

nside: 134217728
ipix= 144115188075855872
lat.dtype= float64
phi= 0.0
lat.deg= -90.0
distance_to_south_pole= 0.0

healpy places is (presumably correctly) above the south pole

import healpy as hp
theta, phi = hp.pix2ang(nside, ipix, nest=True)
print(theta.dtype)
print('phi=', phi)
print('distance_to_south_pole=', np.pi - theta)

float64
phi= 0.7853981633974483
distance_to_south_pole= 6.083373360610267e-09

@dstndstn - could you please try to run this example through only your C code? Is it a bug there, or is it a problem that's introduced in the Python wrapper? (e.g. because of float or int precision / size or rounding issues?)

[cdeil]

@astrofrog - I now see #34 #38 . Looks like this case is just an example of a known issue: precision near the poles.

[cdeil]

I restarted CI after merging #94.

Looks like the phi is still incorrect for this test case?
https://travis-ci.org/astropy/astropy-healpix/jobs/381277304#L1064

So @dstndstn @mreineck - looks like there are other precision issues in the C code.

[dstndstn]

has anyone else had issues installing healpy on OSX? After doing a "pip install healpy --user", which seemed to succeed, I get

python -c "import healpy"

/Users/dstn/Library/Python/3.6/lib/python/site-packages/healpy/pixelfunc.py:107: UserWarning: Warning: cannot import _healpy_pixel_lib module warnings.warn("Warning: cannot import _healpy_pixel_lib module") Traceback (most recent call last): File "<string>", line 1, in <module> File "/Users/dstn/Library/Python/3.6/lib/python/site-packages/healpy/__init__.py", line 81, in <module> from ._query_disc import query_disc, query_strip, query_polygon, boundaries ImportError: dlopen(/Users/dstn/Library/Python/3.6/lib/python/site-packages/healpy/_ query_disc.cpython-36m-darwin.so, 2): Symbol not found: __ZN14T_Healpix_BaseIlE8SetNsideEl23Healpix_Ordering_Scheme Referenced from: /Users/dstn/Library/Python/3.6/lib/python/site-packages/healpy/_ query_disc.cpython-36m-darwin.so Expected in: flat namespace in /Users/dstn/Library/Python/3.6/lib/python/site-packages/healpy/_ query_disc.cpython-36m-darwin.so

…

On Tue, Jun 26, 2018 at 5:11 PM, Christoph Deil ***@***.***> wrote: I restarted CI after merging #94 <#94>. Looks like the phi is still incorrect for this test case? https://travis-ci.org/astropy/astropy-healpix/jobs/381277304#L1064 So @dstndstn <https://github.com/dstndstn> @mreineck <https://github.com/mreineck> - looks like there are other precision issues in the C code. — You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub <#84 (comment)>, or mute the thread <https://github.com/notifications/unsubscribe-auth/ABBD_aQzPkaOPJz9wiiBjjAycpME3Berks5uAqOLgaJpZM4UFIu_> .

[astrofrog]

I often have issues with pip for healpy and have to resort to using the conda package from conda-forge.

[dstndstn]

I looked into this failing test case. It's very similar, in that there is cancellation error near the pole.

Specifically, the term being squared here is ~1e-9, so z -> 1 (cancellation error; it should be 1 - 1e-18).
https://github.com/astropy/astropy-healpix/blob/master/cextern/astrometry.net/healpix.c#L1013

[dstndstn]

I've got a fix for this

[cdeil]

@dstndstn - If you're willing to use conda, there's now a quick way to get an environment with healpy and everything you need for hacking on astropy-healpix: a0b72e7

My experience is that pip install healpy sometimes works and sometimes give linker errors like yours, but using conda and the healpy from conda-forge works reliably.

[cdeil]

I've got a fix for this

Great! Please link to this issue when you make a pull request.

[dstndstn]

Could you please check whether #101 fixes this?
I'm going to be away this weekend.

[coveralls]

Coverage remained the same at 96.214% when pulling ed9305b on astrofrog:test-pix2ang into a0b72e7 on astropy:master.

[astrofrog]

I've rebased this, so it can be merged if CI passes

[cdeil]

Looks like this case is working now. @dstndstn @astrofrog - Thanks!