Improve packed_transaction parsing performance

[greg7mdp]

Resolves #1025.

Introduce a caching_resolver class, which provides the same API as a resolver, but which caches abi_serializers as they are retrieved, so we pay the cost of retrieval only once.

[greg7mdp]

@heifner if I create a abi_serializer cache when parsing the packed_transaction sent by the user, can I reuse the same cache for serializing the trace result? In my current version we do recreate the cache from the trace.

[heifner]

@heifner if I create a abi_serializer cache when parsing the packed_transaction sent by the user, can I reuse the same cache for serializing the trace result? In my current version we do recreate the cache from the trace.

Short answer: no.
The transaction itself or another transaction executed between the time of trx input and trx execution can setabi which would change the abi. Your cache could check for a diff abi_sequence on the account, but you would have to look up the account. You would save the creation of the abi_serializer, but not sure it is worth the extra complexity.

[greg7mdp]

Looks like I have some test failures. I looked at one an it looks like we are trying to retrieve the account before we create it? Am I understanding this right?

[heifner]

Looks like I have some test failures. I looked at one an it looks like we are trying to retrieve the account before we create it? Am I understanding this right?

You can ignore that. It is expected.

[greg7mdp]

Hey, thanks for the tip, I debugged with the core file (super easy) and it pointed me to the problem. Apparently the cache was returning an incorrect abi_serializer. I'm not 100% sure why, but I think it was because we construct a std::optional containing a std::reference_wrapper, and when constructing it from an empty std::optional<abi_serializer> it didn't work quite as expected. Anyways I tried to be very explicit and it seems to work.