General instructions for building for pip distribution

After changing setup.py or any part of the Graffiti codebase, increase the version number in setup.py appropriately, around line 37.

Then run the commands shown below to upgrade what's stored in pip

python3 setup.py prep_to_build npm_run_build sdist bdist_wheel

If you yet don't have twine installed, you need to do pip3 install twine.

Now, uploading to the pypi test servers:

python3 -m twine upload --repository-url https://test.pypi.org/legacy/ dist/* --verbose

(For User name you can use willkessler.)

Installing to a host in the cloud, from the test servers:

python3 -m pip install --index-url https://test.pypi.org/simple/ jupytergraffiti

Testing new installation on cloud host:

jupyter notebook —port=3001 —ip=127.0.0.1 —allow-root

Uploading to the pypi production servers:

python3 -m twine upload dist/*

(For User name you can use willkessler.)

General instructions for building for conda distribution

NB: this relies on jupytergraffiti being in the main (production) pypi repository. First time steps:

mkdir conda
cd conda
rm -rf jupytergraffiti
# Note that after twine upload to pip, above, there will be about 30 seconds before the new build is accessible to conda.
conda skeleton pypi jupytergraffiti
conda install conda-verify

Modify jupytergraffiti/meta.yaml to include build noarch: python and set recipe-maintainers on the last line. Modify jupytergraffiti/meta.json to include github handle: willkessler (appears unnecessary now)

(Re)building on macosx:

(After you make updates, make sure to bump the version number in jupytergraffiti/meta.yml.

conda build jupytergraffiti
export GRAFFITI_VERSION=1.0.1.18
anaconda upload ~/anaconda3/conda-bld/noarch/jupytergraffiti-$GRAFFITI_VERSION-py_0.tar.bz2

You may need to log in with willkessler and the same password you use for pip uploads. (Optional)

Then you want to convert to other architectures and upload these as well:

conda convert --platform all ~/anaconda/conda-bld/osx-64/jupytergraffiti-<version>-py36_0.tar.bz2 -o linux64_out
anaconda upload linux64_out/linux-64/jupytergraffiti-<version>-py36_0.tar.bz2

... etc for all the architectures you want to support. This last step doesn't seem to be necessary any more (01/23/2020).

Finally...

To install and test you should be able to do :

conda install -c willkessler jupytergraffiti

You can test on Udacity streamed linux desktops with:

jupyter notebook —port=3001 —ip=127.0.0.1 —allow-root

Building for Windows and building for pip on Windows

To debug issues on Windows, you need to have cygwin64 and node installed on the system. Then you need to use jupytergraffiti/package.json_windows instead of jupytergraffiti/package.json. This version of package.json has a different build step using utilities provided by Cygwin:

...
  "scripts": {
    "build": "/cygwin64/bin/rm.exe -rf graffiti-dist build ../build_for_pip/code-prep && node node_modules/gulp/bin/gulp.js prebuild && cd node_modules/.bin && r_js.cmd -o ../../build.js && cd ../.. && node ./node_modules/gulp/bin/gulp.js moveStyles && cd node_modules/.bin && r_js.cmd -o ../../../build_for_pip/buildPip.js && cd ../.. && node ./node_modules/gulp/bin/gulp.js pipMoveStyles"
  },
...

You can follow the steps above to build for pip. For creating the pip artifact, from this directory do:

python3 setup.py prep_to_build npm_run_build sdist bdist_wheel

but then you can install from this build, without uploading to pypi servers, by simply :

1. unpacking the tarball found in the dist directory
2. switching to the unpack directory
3. finally: pip install .