@ceastlund turns out I already opened a WIP PR!

I followed your suggestions ie I added dummy 4.08 modules pointing to the unchanged latest_version.ml and then I upgraded latest_version.ml and wrote the appropriate migrations in version_4_07.ml.

I went for the dumbest approach when writing the node migration functions as I wasn't sure where I was heading. I only just recently started cleaning it up a bit but it's still fairly simple and we can probably factor a lot of code there.

One thing I noticed is that I always passed the same version to the to_node and of_node functions for a given migration "direction" so I extracted the migration type that I pass around instead of those functions. I'm wondering if there is any good reason to pass it different versions. If not we could probably remove them from that argument.

Good point about the version arguments. If you like I'll make a PR to remove those arguments.

For the test failures, it might help if we wrote some tests directly against the History interface. We could then test individual conversions, downgrade and upgrade, against both randomized and hard-coded inputs, and choose the hard-coded ones to stress each AST change. We'd want to test round-tripping, and that output satisfies the right grammar. I think these tests would be easier to use to get the conversions right, and the Ppx_ast tests would be more sanity checks.

Want to make a PR adding some tests and I'll take a look at it?

Yeah I'll try to add such tests!