LightBlock verification ignores last_block_id

[Shivani912]

In the current Model Based Testing, we ignore a few Header fields i.e. set them to null and one of them is last_block_id. According to the spec, Blocks, except for the first one, is expected to have last_block_id pointing to the previous block's block_id. It's surprising that neither Tendermint in Go nor in Rust checks this!

Also with the idea of correct-by-construction data structures, I'd expect a Block at height > 1 with a missing last_block_id to fail to translate into Domain type.

[ebuchman]

Thanks @Shivani912 . I think the reason we don't check it is because there's nothing we can validate during skipping verification, so it would only be relevant during sequential, and it doesn't seem to pose a meaningful attack vector, since you already have the blocks chained through the NextValidatorSet and through the next blocks LastCommitID (which points to the blockID of the previous block, effectively giving you the same thing as a LastBlockID).

That said, it would probably be good if the MBT was using complete data structures as much as possible, and you're right that we shouldn't even be able to construct a Block type without it for height > 1. This again calls up the #101 issue

[Shivani912]

Thanks @ebuchman and apologies if this is a totally dumb discussion. I'm still not convinced that we shouldn't have it 😄

since you already have the blocks chained through the NextValidatorSet and through the next blocks LastCommitID (which points to the blockID of the previous block, effectively giving you the same thing as a LastBlockID).

I'm not sure what you mean by LastCommitID 🤔

That said, it would probably be good if the MBT was using complete data structures as much as possible,

Yes, ideally it should but it's also interesting to see that an incomplete data structure passes verification.

[ebuchman]

I'm not sure what you mean by LastCommitID 🤔

Sorry I meant block.LastCommit.BlockID.

Not a dumb discussion at all, we should certainly strive to fill out the data structures properly. Will be especially important so we can use them for block processing tests when we get there.