I developed and published my first application to PyPi last week. This involved a number of different stages and has involved into a project that I'm using to test various areas I haven't explored much in my professional work, and tools that I haven't had a chance to play with in my professional experience.

In the beginning there was a problem...

This project started 7 months ago when I wanted to convert audiobook times into the number of book pages read. To follow along, you can see the project at this stage here.

At this point it used a lambda to parse the input before turning it into a datetime. It worked, but as can be seen, I didn't really know the power of the Python datetime libraries at this point. It included some basic test coverage, and everything worked reasonably well. I knew I wanted to convert it to a pip installable package at some point, but I moved on to other things.

A refactor and a feature

There was a small refactor at one point to change how an import happened, as well as removing a test.

Shortly after this, I updated the code to allow for hours, minutes and seconds as well as minutes and seconds. Extra tests were added, but the lambda had an offspring.

Test coverage and mutation testing

The next set of changes came near the end of June when I was looking at testing as part of the 100 days of Python course and I added test coverage, trying to get up to 100% test coverage outside of the main file. This involved refactoring and some pytest options.

The other thing that I added at this point was mutation testing. At this point I didn't really explore the output of the mutation testing, which would come back to bite me later in the project. This is the repository at this point.

Logging

The next step was to add some logging for this application. I did this using logbook as part of the 100 days of code course when doing the logging section.

The logging was flawed at this stage, and I only found this out later when turning it into a pip installable application. Overall, very little logging needed to be added. I may change around the logging in the future.

For the third day of the logging module, I had some spare time outside of the logging, so I added the setup.py and additional components needed to install this through pip. It was soon after this I realised that my logs followed me around.

Bug fixes, refactoring and CI/CD

The final set of changes were made a couple of days ago when doing the refactoring part of 100 days of Python.

The first thing I did in this step was to fix the logging to stop it working on the current directory. I did this using the path library. I also turned it into a function with an argument to allow this. I originally had lru_cache from functools in front of it to stop each invocation recreating the log object, but I removed this while debugging and never readded it. I've raised a ticket on this with a good first issue label if anyone wants to contribute.

I also added circleci to automate testing of the project on new pull requests. Learning to use circleci was interesting, though I'm glad I put it off for so long. In the previous state without workflows I think I would have been disappointed. Workflows make things much nicer.

As part of the automated testing with circleci, I also had failures because of mutmut's tests failing. I previously hadn't seen anything wrong, or perhaps didn't investigate the failures enough because it didn't stop deployments. I now have a better understanding of how mutmut works. It is a useful tool to point out holes in your testing, and what could go wrong due to those holes. One of the big areas I had to patch up was the logging.

After adding the automated tests, I also wanted to get automated deployments working to PyPi. This was a relatively easy task that didn't involve much extra work, but did involve more exploration of workflows. This is the first point I would have ran into the limitations of the circleci workflow before version 2.1 of the API. As part of this, I did have to convert the readme to restructured text, because for some reason the markdown formatting wasn't working, despite updating setuptools and pip on my installation.

Conclusion

I'm pretty happy about where the project currently is. I would like to look at using additional interfaces. The big one that would benefit me is making it be triggered by Alexa. I'll play with this at some point, possibly after the 100 days of code though, as I don't think it's related to any of the topics covered. I have also raised issues around documentation for the code of conduct and the contributors guide.

I may make a web interface and API for it when I get to those parts of the course, but I can't see myself using this myself or hosting it in the longterm. A better bet for the web interface would be the speedtesting application that I've been working on. I'll talk about that more next time though.