Enable appveyor builds #1224
Enable appveyor builds #1224
Conversation
QuLogic
force-pushed
the
appveyor
branch
5 times, most recently
from
10bfb6b
to
9e4fbc4
Compare
|
Well, it took quite a few tries, but I think it's finally at a point where I'll have to leave it. There doesn't appear to be a |
QuLogic
force-pushed
the
appveyor
branch
11 times, most recently
from
161e385
to
6f78f25
Compare
|
@QuLogic - it would be great to get to the point where we at least check installation (even if the tests don't actually run). I don't have access to a Windows machine particularly conveniently, but have gone through the process of installing cartopy on windows on a fair few occassions - most of that knowledge got encoded into conda-forge. Is there anything you need from me that can help move us to a situation where we are able to test installation on Windows? |
|
Installation is actually working now; I did cop a few bits from the conda-forge build. Python 2.7 is passing as well, but 3.6 seems to be broken due to something in pyproj. |
| - set PACKAGES=%PACKAGES% pillow pytest pytest-xdist filelock pep8 pyshp shapely six requests pyepsg owslib | ||
| - conda install --quiet %PACKAGES% | ||
| - set INCLUDE=%CONDA_PREFIX%\Library\include;%INCLUDE% | ||
| - set LIB=%CONDA_PREFIX%\Library\lib;%LIB% |
There was a problem hiding this comment.
Also, using the variables from the conda-forge build does not work here. I don't even know why I need to set these two explicitly; I thought running python from the environment would grab a compiler with these correctly set.
QuLogic
force-pushed
the
appveyor
branch
2 times, most recently
from
6d11c30
to
8546c72
Compare
|
Ah, success. Had to move all packages into the environment creation so that Also, turned off xdist, which seemed to hang before. It's only 90 vs 60 seconds (on Travis). Of course, the bigger slowdown is the fact that AppVeyor doesn't run the builds in parallel. |
|
This is looking really promising. Are you OK with disabling doctests on Windows? I'm not convinced that the reduced readability is worth the benefit of asserting documentation correctness for Windows (given we do assert correctness is unix, so they clearly aren't garbage). |
|
I think I'd prefer to just not run doctests on windows. |
|
Comments should be handled now. |
|
Amazing. Thanks for this @QuLogic! |
Rationale
#1221
Implications
This tests Python 3.6 latest everything and Python 2.7 oldest everything, on 64-bit
, then 32-bit. Because builds do not run in parallel, I ordered it from most relevant to least (as I see it.) It seems like conda-forge may no longer do 32-bit builds, so we may not need to check those. I also left out a few things that are already tested on Travis.