Use Azure Pipelines and upload to anaconda

[mattip]

Copy the pandas-wheels approach of using Azure and uploading to anaconda

[charris]

Progress!

[charris]

The windows wheels seem to be 1.18.2?

[mattip]

Yes, since the build tag is 1.18.2
Still todo:

• fix the openblas windows linking
• use tokens to upload to anaconda

[mattip]

@tylerjereddy thanks. I added a NUMPY_STAGING_UPLOAD_TOKEN secret to this repo, which I got from the multibuild-wheels-staging group, but anaconda upload is failing. Any ideas? Is there a way to get more info from the anaconda client?

log of the upload

numpy/dist/numpy-1.19.0.dev0+9eb7560-cp36-cp36m-win_amd64.whl
Using Anaconda API: https://api.anaconda.org
Using "multibuild-wheels-staging" as upload username
Processing 'numpy/dist/numpy-1.19.0.dev0+9eb7560-cp36-cp36m-win_amd64.whl'
Detecting file type...
File type is "pypi"
Extracting python package attributes for upload
Creating package "numpy"

##[error]Bash exited with code '1'.

[mattip]

I can run anaconda -t $SECRET upload -u multibuild-wheels-staging numpy/dist/numpy*.whl from the command line, so something must be off with the script I copied from pandas-wheels. Adding some debug in c0a3da2

[mattip]

Hah, the $TOKEN starts with "$(N", not the contents of the secret.

[mattip]

Maybe the problem is I added the secrets to the repo, not the azure pipeline.

[mattip]

Success!!!. Let's see if I can get 1.18.3 releases onto that site, and I will remove the 1.19 pre-releases, they belong in the https://anaconda.org/scipy-wheels-nightly/ directory

[mattip]

Making one more commit, that I will back out, to test the nightly cron job and that linux32 passes on master HEAD.

[charris]

@mattip are you sure this should be merged? There are still the 32bit linux failures which look like failure it install OpenBLAS

[mattip]

There are still the 32bit linux failures

Those are due to the v1.18.3 tag and some incompatibility with the uname Azure reports for the 32-bit build here vs. the uname it reports on numpy/numpy. The openblas_support.py is more robust on the master branch so it all passes, as you can see at both the https://anaconda.org/multibuild-wheels-staging (which are the pre-release wheels) and the https://anaconda.org/scipy-wheels-nightly/numpy/files (which are the weekly wheels built off HEAD).

[charris]

Thanks Matti.

[charris]

.appveyor.yml still exists, should it be removed?

[mattip]

Unlike travis, appveyor errors out if there is no .appveyor.yml so I had to add an empty one.

[charris]

The appveyor hook can be disabled in the settings, I'll do that.

[charris]

I've disabled all the webhooks for appveyor and travis.

[charris]

There were three for appveyor and I deleted the failing one, which is probably the one I should have kept. Oh well...