Log software + packages versions

[jc-cisneros]

In this issue we will test the write_version_log feature from gslab_make (see https://github.com/gslab-econ/gslab_make/issues/56).

Task list

• [x] Add the new feature to the make scripts in each submodule.
• [x] Check that a versions.log file is created in each submodule's log folder.
• [x] Check that versions.log correctly prints the versions from the active environment.
• [x] Robustness tests (i.e., the script should produce a result for all possible user behavior and throw appropriate error messages when relevant)

[jc-cisneros]

@gentzkow, following the latest conversations on package versions (see https://github.com/gslab-econ/gslab_make/issues/56), I added a write_versions_log feature to produce a log file that lists all the software + packages used in a conda environment, their versions and a timestamp. This feature can be added to any project using gslab_make so that we can always check the versions used in any commit. A caveat of the current code is that it requires a conda environment to be active (i.e, if a user has all the dependencies installed locally, the versions.log file would not be created).

The make.log files now include a new line when the version logs are correctly created:

An example of how a versions.log file looks within a module can be found on the latest commit.

If this sounds good, I can proceed with the double PR (on both template and gslab_make).

[gentzkow]

Great! This looks very good.

I'm thinking we may want to incorporate this into template and then make the default to not specify explicit version numbers in the setup file. We can note in a comment in that file that users can add explicit version numbers if they want to freeze a package at a particular revision or if they're getting conflicts caused by package updates.

[jc-cisneros]

On a98903e, I updated the conda_env.yaml to reflect the new default behavior per https://github.com/gentzkow/template/issues/71#issuecomment-1309251979. I will proceed with the PR for the issue.

[jc-cisneros]

Summary:

On this issue we tested the a new gslab_make feature to log the package versions. The feature worked as expected and was added to the gslab_make master branch.

Flagging some of the lessons and opportunities for future updates:

• The write_versions_log.py feature currently returns the output of running conda list with the active environment and a timestamp. We could add some extra formatting to differentiate Python, R, and Julia packages from each other (see https://github.com/gentzkow/template/pull/72#discussion_r1020986095).
• The current behavior of this function is to create an identical versions.log file for each module. In the future we can try to produce a versions.log that prints only the packages that were used in each module or script (see https://github.com/gentzkow/template/pull/72#discussion_r1020986973).