MIKE LEVIN AI SEO

Future-proof your skills with Linux, Python, vim & git as I share with you the most timeless and love-worthy tools in tech through my two great projects that work great together.

Pushing Blogslicer to PyPI (From Jupyter Notebook)

I successfully created a pattern for releasing packages built in Jupyter Notebooks into PyPI, using nbdev from Fast.ai. I pushed my package, Blogslicer, to PyPI, but still have more work to do before I can add it to my blog release script. Read more to find out how I did it!

Successfully Releasing Packages Built in Jupyter Notebooks to PyPI with nbdev

By Michael Levin

Sunday, May 1, 2022

I hate smart quotes. I don’t think they’re good for SEO (search engine optimization). I adhere to a principle called “The Weakest Link In The Chain”. Unfortunately such a good metaphor has been tarnished by that awful game show of the same name years back, but the concept will last longer than memories of that atrocity, so I will keep using it.

Every Little Thing gets done, or ELTgd, as I’m currently acronym’ing it. I get smart quotes when I copy/paste out of other more word-processing oriented software into vim where I blog. There’s many ways I could address fixing this, but I have a release script that slices & dices my single-long-page text files that I blog in (one journal for life, one journal per website) into separate, individually optimized pages.

I actually don’t sweat the optimization too much. I let my confidence that I will gradually massage and organize sites over time allow me to write at full-speed. And I do write fast. My typing speed isn’t really that fast. Neither is my thinking speed. It’s just that both are decent speed and in sync with each other, so I write just a bit slower than I can talk, and that means I can publish a lot… a whole lot, and so I do… now… as a result of operation ELTgd.

This particular blog post on Pythonic Ally is to see how quickly I can get a package called “dumbquotes” into PyPi. For you see, the next step in my blogslicer (also soon to be in PyPI) is to strip out smart quotes, and I don’t see why that can’t and shouldn’t be a separate package of potentially broader use than blogslicer if made separate.

Okay, so before I do dumbquotes, I want to get blogslicer into PyPi.org. I use a developer tool called nbdev from the illustrious Australian and all around cool guy, Jeremy Howard of Fast.ai. If you’re into Machine Learning, I suggest all his stuff over the competing Google TensorFlow because with TensorFlow, all things lead to the vendor-trap. But with fast.ai, all roads lead to generic GPU-based less vendor dependt goodness.

But I digress. I’ve already refined blogslicer a bit in Github. Making it pip installable should be a breeze. 1, 2, 3… 1?

Quickly edit the index.ipynb which gets turned into the README.md which becomes the Github repo homepage documentation. Important! Okay, done but my trick for switching between Notebook mode and command-line mode arguments is causing an error on:

#hide
from blogslicer.core import *

So I’m just commenting out that import and tabling the issue for now. Next? Look at my release system. Even though I don’t NEED to bump the version, that is usually my next step. And I’m doing everything manually first. Then I’ll put it into my release script.

$ nbdev_bump_version
Old version: 0.0.1
New version: 0.0.2

Okay, that went off without a hitch. Next!

$ nbdev_build_docs

…oops! A glorious error, probably related to the above problem importing blogslicer.core into the index.ipynb. nbdev has its weirdness. It’ll just be a matter of practice, but I’ll skip it for now. Next!

$ nbdev_build_lib
Converted 00_core.ipynb.
Converted index.ipynb.

Okay, that ran without a hitch and that’s the important one. Next!

nbdev_clean_nbs

No output, but that means it ran well. Always a good idea before:

git commit -am "Preparing for first PyPI release"
(output removed)
(py310) ubuntu@LunderVand:/mnt/c/Users/mikle/github/blogslicer$ git push
(output removed)

Next!

$ mkdir dist
$ python3 -m build
(lots of output removed)
Successfully built blogslicer-0.0.2.tar.gz and blogslicer-0.0.2-py3-none-any.whl

And finally…

$ twine upload dist/*
Uploading distributions to https://upload.pypi.org/legacy/
Uploading blogslicer-0.0.2-py3-none-any.whl
100% ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ 12.6/12.6 kB • 00:00 • 461.3 kB/s
Uploading blogslicer-0.0.2.tar.gz
100% ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ 13.7/13.7 kB • 00:00 • 718.8 kB/s

View at:
https://pypi.org/project/blogslicer/0.0.2/

Success! But without the documentation successfully built, both the Github repo page and PyPi package page contain a generic nbdev template. Ugh! That can’t stand. Okay, diagnose that quickly. No rabbit holes!

Okay, I was able to uncomment the index.ipynb package import by changing the Jupyter environment check to control where the arguments come from like so:

if hasattr(__builtins__, "__IPYTHON__") or __name__ != '__main__':
    from IPython.display import display, Markdown

    h1 = lambda text: display(Markdown(f"# {text}"))
    h2 = lambda text: display(Markdown(f"## {text}"))
    h3 = lambda text: display(Markdown(f"### {text}"))

    folder_name = "../pythonically"
    blog_title = "Pythonic Ally Blog"
    blog_slug = "blog"
    author = "Mike Levin"
else:
    h1 = lambda text: print(f"# {text}")
    h2 = lambda text: print(f"## {text}")
    h3 = lambda text: print(f"## {text}")

    aparser = argparse.ArgumentParser()
    add_arg = aparser.add_argument
    add_arg("-p", "--path", required=True)
    add_arg("-t", "--title", required=True)
    add_arg("-s", "--slug", required=True)
    add_arg("-a", "--author", required=True)
    args = aparser.parse_args()

    folder_name = args.path
    blog_title = args.title
    blog_slug = args.slug
    author = args.author

I know it worked without breaking the command-line (using “python” in the command) input because you’re reading this. And so I’ve finally completed my pattern for creating packages in Jupyter Notebooks (really, JupyterLab) and having them also work as parameterized command-line tools releasable into PyPI. I still have a little work to go to make it work without the python interpreter prefix. I’ve got some research to do on how pip installing results in CLI-runnable commands.

So now onto dumb-quotes. I’m going to hold off on putting the actual PyPI release into my everyday frequently used blog release script. I thing that may cause more harm than good. But definitely a topic for later is bringing the PyPI package up-to-date with my fast-and-furious daily code tweaking.

Okay, so one more time, but from scratch (blogslicer was already well underway).

Categories