Compute tick angles by advancing a small amount in axes coordinates

[lpsinger]

This handles cases when a plot extends over a small number of pixels in the WCS projection, but a large distance or number of pixels in display coordinates. Addresses #45.

Here is a modified test script. The ticks are red and extra-long to make them stand out better.

#!/usr/bin/env python
from matplotlib import pyplot as plt
from wcsaxes import WCSAxes
import astropy.wcs

wcs = astropy.wcs.WCS()
wcs.wcs.ctype = ['RA---TAN', 'DEC--TAN']
wcs.wcs.crval = [90, 70]
wcs.wcs.cdelt = [16, 16]
wcs.wcs.crpix = [1, 1]
wcs.wcs.radesys = 'ICRS'
wcs.wcs.equinox = 2000.0
fig = plt.figure(figsize=(3, 3), dpi=72)
ax = WCSAxes(fig, [0.1, 0.1, 0.8, 0.8], wcs=wcs)
fig.add_axes(ax)
ax.set_xlim(1, -1)
ax.set_ylim(-1, 1)
ax.grid(color='gray', alpha=0.5)
ax.coords['ra'].set_ticks(color='red', size=20)
ax.coords['dec'].set_ticks(color='red', size=20)
fig.savefig('test.png', dpi=72)

[lpsinger]

Isn't this line a no-op because world0 is assigned to again in the following line?

[astrofrog]

Indeed - we should try and figure out when this was introduced (I'll take a look at git blame)

[astrofrog]

This is a great idea, thanks! Will leave a few comments below. I think we should then wait for #41 to be merged since it adds image testing, then we can add a regression test for this.

[astrofrog]

If these are always going to be affine transformations with no rotation, I wonder whether it would be more efficient to simply find out what the scale from axes to data coordinates is and multiply/divide by these later on?

[lpsinger]

If these are always going to be affine transformations with no rotation, I wonder whether it would be more efficient to simply find out what the scale from axes to data coordinates is and multiply/divide by these later on?

Yes, but it would be clearer and nearly as efficient to pass all of the ticks to transform() as one big vector.

You might be able to eliminate the multiple transformation objects by using a blended transform.

[astrofrog]

@lpsinger - could you rebase this now that #41 has been merged? Also, it would be great if you could add the example you have above as an image test. To do this, you need to add a test to test_image.py then generate the reference image by running the tests with:

py.test --generate-reference True wcsaxes/tests/test_image.py

If you don't have time to do this, we could merge it as-is and I or @anizami could add the image test.

[anizami]

@lpsinger @astrofrog - I actually made a minor change so you can just generate the reference image by py.test --generate-reference wcsaxes/tests/test_image.py

[astrofrog]

@anizami - great!

[lpsinger]

@lpsinger - could you rebase this now that #41 has been merged?

Updated via force push.

[lpsinger]

Also, it would be great if you could add the example you have above as an image test.

Matplotlib tries to limit the number of test images to keep the repository size small. Is it OK if I make one image test that handles #43, #44, and #46?

[astrofrog]

@lpsinger - I think that #43/#44 and #46 should be separate since they are testing very different things. However, I think the reference images won't be too big as long as you don't actually plot an image.

[lpsinger]

@lpsinger - I think that #43/#44 and #46 should be separate since they are testing very different things. However, I think the reference images won't be too big as long as you don't actually plot an image.

Alright, shall I make one test for #43/#44 and then rebase #43/#44 into one PR?

[astrofrog]

@lpsinger - that'd be great, thanks!

[lpsinger]

@lpsinger @astrofrog - I actually made a minor change so you can just generate the reference image by py.test --generate-reference wcsaxes/tests/test_image.py

I have a unit test, but I cannot generate a reference image. I get:

$ py.test-2.7 --generate-reference True wcsaxes/tests/test_image.py
usage: py.test-2.7 [options] [file_or_dir] [file_or_dir] [...]
py.test-2.7: error: unrecognized arguments: --generate-reference

EDIT: never mind, figured it out.

[astrofrog]

@lpsinger - just to check, are you still planning on adding a reference image to this PR? It would be great to have a regression test to avoid this issue creeping back in future.

[lpsinger]

@lpsinger - just to check, are you still planning on adding a reference image to this PR?

Yes, I thought I would wait until after #49 was closed.

[astrofrog]

@lpsinger - ok, #49 has now been merged, so feel free to add a test here - thanks!

[lpsinger]

@lpsinger - ok, #49 has now been merged, so feel free to add a test here - thanks!

Done. By the way, I had to dig a little in the source code to remind myself of the --generate-reference option. Could you document it, please?

[anizami]

@astrofrog - Since this options seems more relevant for developers, where do you recommend we document this?

[lpsinger]

@astrofrog - Since this options seems more relevant for developers, where do you recommend we document this?

If your goal is to eventually put this code into Astropy, you could imitate their Developer Documentation section, which includes Testing Guidelines.

[astrofrog]

Thanks @lpsinger!

[astrofrog]

@anizami - can you open an issue for now to remind ourselves to document the --generate-reference and we can chat about it later?

[anizami]

@astrofrog - Sure!

[astrofrog]

@lpsinger - it looks like this doesn't work properly if the axes are not square. Let me know if you have time to look into how to fix this today. If not, then I can look into it.

[lpsinger]

@lpsinger - it looks like this doesn't work properly if the axes are not square.

Yes, I'll take a look. Do you have an example script?

[astrofrog]

@lpsinger - yes, simply take your example and make the figure e.g. 6x3:

w = WCS()
w.wcs.ctype = ['RA---TAN', 'DEC--TAN']
w.wcs.crval = [90, 70]
w.wcs.cdelt = [16, 16]
w.wcs.crpix = [1, 1]
w.wcs.radesys = 'ICRS'
w.wcs.equinox = 2000.0
fig = plt.figure(figsize=(6, 3))
ax = WCSAxes(fig, [0.1, 0.1, 0.8, 0.8], wcs=w)
fig.add_axes(ax)
ax.set_xlim(1, -1)
ax.set_ylim(-1, 1)
ax.grid(color='gray', alpha=0.5, linestyle='solid')
ax.coords['ra'].set_ticks(color='red', size=20)
ax.coords['dec'].set_ticks(color='red', size=20)
fig.savefig('angles.png')