Hi! This is the friendly automated conda-forge-linting service.

I just wanted to let you know that I linted all conda-recipes in your PR (recipe) and found it was in an excellent condition.

@conda-forge/h5py, this is ready for review, I think (cc @hmaarrfk).

I'm not sure why cross compiling isn't working for arch and ppc64le. It should in theory work.

Travis is quite slow (mostly due to our limited access to it) so for these large build matrices, maybe it is good to do them on azure.

I'm not sure why cross compiling isn't working for arch and ppc64le. It should in theory work.

I think this is because of the use of ctypes.CDLL(...) in setup_configure.py. This should be fixed by fcbb1ff, so bb08b91 could probably be harmlessly reverted if required.

Travis is quite slow (mostly due to our limited access to it) so for these large build matrices, maybe it is good to do them on azure.

From the tiny dataset that this PR represents, the travis-ci build matrix completes faster than the azure build matrix. The linux-ppc64le builds are the slowest, but are still faster than the win-64 builds run on Azure.

I appreciate that this is highly dependent on the total job queue on travis-ci, and UK morning/early-afternoon on a Sunday is probably the best time to attempt a large parallel matrix, but even if it is slower in general I prefer simplicity of configuration (using the 'default' provider in conda-forge.yml) over speed.

Thank you for the thorough discussion, I think that this is a fine PR and should be merged. I'll let it sit here until a maintainer sees it.

Lets move on. The discussion was quite clear that the intention was to use the previous as a crutch, while a real fix is found.

Thanks for following through.