CASSANDRA-20883: Add support in the binary protocol to allow transactions to have multiple conditions

[dcapwell]

No description provided.

[dcapwell]

think i should likely rename this to call out its really only safe for CAS since it has 1 condition

[belliottsmith]

where is this used?

[dcapwell]

we spoke in slack

It used to be used but after moving off the all/non bit sets, nothing uses and i forgot to cleanup

[belliottsmith]

do we use this hashCode, or can we throw UnsupportedOperationException? If not, can we just use id?

[belliottsmith]

suggest: new line

[dcapwell]

code has changed a lot since this comment, i assume you changed this in your feedback branch? I assume you wanted a newline above this?

[belliottsmith]

should this use TxnCondition.None instead of null? Can simplify some other logic then.

[belliottsmith]

outter is a typo, but I prefer superset/subset for this kind of logic as it makes clearer the relationship. Also, we already have this logic in various places. The most obvious is AbstractWithKeysSerializer, though this is more annoying to use that we might like. Surely we at least have the subset bitset logic somewhere already? If not, let's at least consider adding that, but it's a shame this form of logic is getting fragmented and reproduced in different ways.

[belliottsmith]

why not standardise(ish) the two serializers? They're basically the same, we just have an implicit capacity for the small bitset. Then for this serializer we can avoid writing a kind and instead write the capacity, which implies if it can be created as a SmallBitSet or must be a LargeBitSet.

The large serializer can avoid writing out wordsInUse if <= 1, and always serialize at least one long (with value 0 if empty), and then the two are serialied identically, and we would just need the large serializer to expose methods that accept/assume the capacity so we can invoke that directly for this use case.

[belliottsmith]

Why is condition nullable? I don't believe we assign it anything non-null in real codepaths, and I don't think we should either, since we have the None condition (which takes up 1 byte rather than zero).