This is a good fix, thanks.

An interesting challenge with this change: since it modifies the schema, shipping a release with it could break existing databases the next time git-history file ... is run against them.

This would affect my workflow here for example: https://github.com/simonw/scrape-instances-social/blob/main/.github/workflows/scrape.yml

Options for doing this:

• Document this as a breaking change (maybe even by shipping a 1.x version) and let people know how to work around it
• Have the code tool itself learn how to modify the schema on existing tables to fix this

I'm leaning towards the second option, depending on how hard it will be to implement.

Worth noting that this really was a bug: the code was designed to create that _commit column but failed to because it mutated a copy of the item from the array, not the original object:

I think the fix is to detect if the item_table is missing that _commit column and add it.

The schema is only changed for those using the non---id branch (and worth fixing carefully for those users). But not in fact scrape-instances-social currently.

But as you've probably realised, this fix removes the need for the --id workaround you mention in tracking-mastodon. Stop setting --id, and use the simpler set of tables with a join to get the commit date: