Sandbox BGS does not accept events with missing `prev` field

[DavidBuchanan314]

IIUC, as of repo v3, the prev field is optional. I got rid of it while updating picopds to repo v3. However, I couldn't get the sandbox BGS to accept my commit events until I added dummy prev fields (set to null) to my commit objects and firehose events (I'm not sure if both were strictly needed, but I changed both at once and it started working after that).

[bnewbold]

Thanks for the report, will take a look!

[whyrusleeping]

hrm... the prev field on the generated struct probably wants to be omitempty, but i dont think thats actually going to cause an issue.

[whyrusleeping]

I see some signature validation failures on retroid repos

[DavidBuchanan314]

Do you have timestamps for those validation failures? I might be able to figure out what I was doing at the time.

At one point I'd forgotten to update my _atproto DNS record, although I don't think that would manifest itself as signature validation errors

[whyrusleeping]

yeah i saw the bad dns record one.

I'm 90% sure whats going on here is that you're signing over the 'commit' with the 'prev' field entirely missing, where when i check the signature i'm computing the bytes to have 'prev=null'.
Since i'm properly validating the signatures on the typescript code, i can assume that they are doing the same thing here. Actually removing the prev field is going to be tricky, missing field != null will be fun...

[bnewbold]

👀

yeah, prev is both optional and nullable per the v3 spec. it was nullable previously and we want to allow just not including it at all, but also wanted to be as compatible as possible during the transition.

some languages should be able to validate whatever is in the commit. golang will have a bit of trouble... I guess trying both variations shouldn't be too bad performance-wise as long as it is really clear which is expected and 99.99% of signature validations are happy path?

this could be a major problem for, eg, label signing and validation though (several fields, with no-value-provided common for several), will want to ensure there is no ambiguity in that case.

cc: @dholms