Thanks for the quick PR! A few style suggestions. Other than that, my main concern is that we get this tested.

Let's ensure we have tests that cover both cases. You can change https://github.com/facebook/relay/blob/d5d1c07d7a44856f82c9a652697e9662a091d4c6/compiler/crates/relay-transforms/tests/refetchable_fragment/mod.rs#LL18C60-L18C67 to build a custom schema config and then use a pattern like this to make it conditional based on a comment in the fixture file.

Then please add tests which ensure we cover both cases.

Finally, does this introduce a possibility that a type would have neither a an id field or a token field? Do we have validation anywhere to catch that?

I'll look into adding tests for this using a flag like that, thanks for the pointers!

Finally, does this introduce a possibility that a type would have neither a an id field or a token field? Do we have validation anywhere to catch that?

We still fetch the id field if it exists. And we always fetch the fetchable id. As long as the type you add fetchable to has an id field we should be OK.But I suppose we could add validation for the scenario where tokenFieldEnabled: false and the type does not have the id field available.

Is there a reason this PR hasn't been accepted yet? I'm running into this problem and without editing the generated files, I can't see another way around this until this PR is merged?

Apologies for loosing track of this. Let me import/rebase this and play around with it a bit.

@captbaritone has imported this pull request. If you are a Meta employee, you can view this diff on Phabricator.

Import failed due to this PR being old and our internal tool getting stuck when trying to rebase. I'm going to take these commits, rebase them myself and open a new PR to import. The author info on the commit should remain

Rebased here: #4889 Going to close this PR. We can continue the conversation on that PR.