Pandas 3.0 and edgartools
How a dependency issue with pandas was fixed in edgartools
On Wednesday I noticed one of my applications failed to start after deployment. In the logs the issue was a missing dependency - pytz - a Python library used for timezone calculations. That dependency was used by edgartools , but not specified by edgartools in the pyproject.toml. Instead the library pulled pytz as a transitive dependency via pandas. This was a serious issue since it meant the new installations of edgartools would face the same issue and users would not be able to install it. So clearly it had to be fixed, but also I needed to figure out why this issue happened.
Well actually for a really good reason. Pandas 3.0 was released on Tuesday night. This had been a long awaited and publicized release but I had not been paying attention. More accurately my attention was diverted from foundation data libraries towards foundation models and so I missed a really important event. That new version removed pytz as a dependency and so exposed an error on my part, which was to add a line of code to edgartools that worked only because older versions of pandas included pytz.
So Release 5.11.1 was an emergency version of edgartools that removed pytz as a dependency. If you are affected by the issue then install 5.11.1 or later and that should solve the issue. I also pinned the version of pandas to below 3.0 to prevent new installations from picking up the latest release of Pandas. The obvious downside is of course that this prevents users from using both edgartools and Pandas 3.0 in the same environment but I am actively working on fixing this. I also work on ways to prevent this from happening in the future.
Something else I noticed was the massive spike in installations of edgartools on Python 3.10 since the Pandas 3.0 release.
Pandas 3.0 dropped 3.10 support it could be that as jobs started to run pip would need to backtrack through several installations to determine the correct dependency tree. Hopefully this settles down in the next couple of days.
Conclusion
Pandas 3.0 support should be available soon so you won't be forced to choose between the latest version of a foundational library and edgartools. In any event this was a learning experience for me. Sorry if you were affected and thanks for your continued support. To get started with edgartools head over to the Github repo and leave a star. Cheers!