Install error with recent commits: _wcs doesn't have WcsError

[keflavich]

I've been seeing this error:

$ python setup.py develop
Freezing version number to astropy/version.py
Traceback (most recent call last):
  File "setup.py", line 82, in <module>
    package_dirs)
  File "/Volumes/disk5/Users/adam/repos/astropy/astropy/setup_helpers.py", line 1028, in update_package_files
    extensions.extend(setuppkg.get_extensions())
  File "astropy/wcs/setup_package.py", line 206, in get_extensions
    generate_c_docstrings()
  File "astropy/wcs/setup_package.py", line 122, in generate_c_docstrings
    from astropy.wcs import docstrings
  File "/Volumes/disk5/Users/adam/repos/astropy/astropy/wcs/__init__.py", line 27, in <module>
    from .wcs import *
  File "/Volumes/disk5/Users/adam/repos/astropy/astropy/wcs/wcs.py", line 82, in <module>
    WcsError = _wcs.WcsError
AttributeError: 'module' object has no attribute 'WcsError'

when trying to install from the latest astropy source. On one of my machines, I was able to get around the issue by doing rm -r build, but on my other, the problem persists.

[keflavich]

Sorry, I misreported the error. The issue actually comes from 9d982a3, and is absent in the prior commit, e3e3414. This is true on both machines.

[mdboom]

Have you at any time done a setup.py build --inplace or setup.py develop? That will leave an old version of the _wcs.so extension around, which doesn't have this new WcsError member. I that's what it is, I'll have to again unleash my recurring rant about why any tool that builds things "in place" is an abomination 😉. If that's not what it is, I'm at a bit of a loss.

Try doing git clean -fxd to clean out your git working directory and see if it fixes things.

EDIT: I see setup.py develop at the top of the report. I just have to recommend not doing that, or become accustomed to git clean -fxd. I know that's a minority opinion around here...

[keflavich]

Brilliant, thanks. I personally prefer adding git clean -fxd to my workflow rather than avoiding python setup.py develop - I tend not to edit anything but .py files and prefer to have a shorter workflow for reloading python file changes.

That said, feel free to unleash the rant.

[keflavich]

FYI, note that this is causing problems on RTFD also:
https://readthedocs.org/builds/astroquery/709230/

[keflavich]

On readthedocs, even using a wipe doesn't solve the issue. Any thoughts on that? I don't think you can pass any git commands directly to rtfd, and I guess they don't really delete the whole directory? That seems quite strange to me.

[embray]

Indeed, most of the time setup.py develop is fine--it only bugs out if changes can occur in extension modules. This happens to me most often when changing branches, so sometimes I either do git fresh (which is an alias I have for git clean -d -x -f -e .tox so that it doesn't wipe away my tox envs).

Sometimes if I suspect this is going to happen I just check out a branch into a separate working copy. Usually it's not an issue though.

I don't know why you would be having problems with this in RTD but I don't think that's really an Astropy issue.

[keflavich]

Thanks for those tips. However, the RTD issue is the exact same error message I reported above. @kbarbary reported a similar issue on another RTD build. There are other problems with that build, but I think the failure to build astropy is something we should be able to figure out?

[kbarbary]

@keflavich To me, it really looks like the problem is that the RTD "wipe" is not really deleting everything. Maybe we should ask the RTD guys what's going on. Off-topic for this PR though.

[astrofrog]

Try pressing the wipe button several times in a row - it sometimes goes through different paths.

[keflavich]

Tried that. I think contacting RTD is probably best.

[keflavich]

@cdeil posted the issue at RTD: readthedocs/readthedocs.org#422

While this is perhaps no longer an astropy issue, it is affecting a lot of astropy dependencies, so I think it's good to keep track of it here for a little while longer.

[kbarbary]

Thanks for filing the issue @cdeil!

[cdeil]

@erikholscher confirmed that this is an issue with wipe on readthedocs at readthedocs/readthedocs.org#422 (comment) and they are working on it:

This ticket will be kept open to fix the problem w/ wipe not working.

So IMO this astropy issue can be closed ... astropy affiliated package devs will still find it with the search button.