Add pypi deployment via Travis

[zurk]

Signed-off-by: Konstantin Slavnov [email protected]

[vmarkovtsev]

Do you need after_script: skip here?

[vmarkovtsev]

Better .travis.install

[vmarkovtsev]

Move it to the "build" stage. I know that not necessary but still will be more explicit

[zurk]

if I move it here deploy fails:

subprocess.CalledProcessError: Command '('cmake', '-DCMAKE_BUILD_TYPE=Release', '-DCUDA_TOOLKIT_ROOT_DIR=/usr/local/cuda-8.0', '.')' returned non-zero exit status 1
...
auditwheel: command not found
...
twine: command not found
...

and it looks like install does not run for this stage.

See full logs: https://api.travis-ci.org/v3/job/346796133/log.txt

[vmarkovtsev]

You need to debug this because it works perfectly and I did it multiple times, e.g. https://github.com/vmarkovtsev/ggmbox/blob/master/.travis.yml

[zurk]

but you do not use stages in your example...

[zurk]

@vmarkovtsev, Also I find one more problem:

[0K$ twine upload dist/*manylinux* -u $PYPI_LOGIN -p $PYPI_PASS

  0%|          | 0.00/44.2M [00:00<?, ?bytes/s]
100%|█████████▉| 44.1M/44.2M [00:00<00:00, 612kbytes/s]
                                                       
Uploading distributions to https://upload.pypi.org/legacy/
Uploading libMHCUDA-2.0.2-cp34-cp34m-manylinux1_x86_64.whl
HTTPError: 403 Client Error: The user '[secure]' is not allowed to upload to project 'libMHCUDA'. See https://pypi.org/help#project-name for more information. for url: https://upload.pypi.org/legacy/

travis_time:end:1bf1826c:start=1519741115503683766,finish=1519741119985534458,duration=4481850692
�[0Ktravis_fold:end:after_script
�[0K
Done. Your build exited with 0.

in case of error, twine returns the error message instead of 1 code (https://github.com/pypa/twine/blob/master/twine/__main__.py#L27).
I can make a PR to twine, but any ideas how to work around it for now?

[zurk]

And this is not outdated. #3 (comment)

[vmarkovtsev]

I don't see much changes...

Also script should run the tests, not install.

And be aware that this project is actually C++. There are python bindings, but there can be other bindings too. Don't nail it to Python.

[vmarkovtsev]

@zurk

dist: trusty
sudo: required

matrix:
  include:
    - language: python
      python:
        - 3.4
        - 3.5
        - 3.6
      env:
        - CUDA_TOOLKIT_ROOT_DIR=/usr/local/cuda-8.0
      install: . .python.install
      deploy:
        - if tag
        - twine upload dist/*manylinux* -u $PYPI_LOGIN -p $PYPI_PASS
      language: cpp
      install:
        - mkdir build
        - cd build
        - cmake -G "Unix Makefiles" -DCMAKE_BUILD_TYPE=Release -DCUDA_TOOLKIT_ROOT_DIR=/usr/local/cuda-8.0 ..
notifications:
  email: false

[zurk]

@vmarkovtsev I adopt your script.
for any reason

python:
  - 3.4
  - 3.5
  - 3.6

does not work at all. and any modifications of this code.
Also, deployment stage fails with gives zero-information message:

[33mInstalling deploy dependencies�[0m

!!! Script support is experimental !!!


travis_fold:end:dpl.1
travis_fold:start:dpl.2
�[33mPreparing deploy�[0m

travis_fold:end:dpl.2
travis_fold:start:dpl.3
�[33mDeploying application�[0m
Script failed with status 127

travis_fold:end:dpl.3
failed to deploy

But It should be the same as I share earlier #3 (comment). You should add PyPI login and password to Travis settings to $PYPI_LOGIN and $PYPI_PASS env vars.

[vmarkovtsev]

@zurk If setting several python versions do not work - grow the matrix and use YAML pointers to be DRY.

[vmarkovtsev]

I will add the login and password as soon as we merge this.

[zurk]

grow the matrix and use YAML pointers to be DRY.

I think I did exactly what you are saying

[vmarkovtsev]

You don't need virtualenv in this ephemeral environment, nothing to isolate from. Each job runs in a separate instance.

[vmarkovtsev]

@zurk Almost there

[zurk]

Yes, now it works without virtualenv activation. Before I have some troubles. Looks like artifact while searching how to make it work.