Improve packed_transaction parsing performance

[heifner]

See #999 (comment)
Simple speedup is to create a caching resolver that caches abis as it goes instead of repeatedly retrieving them.

[greg7mdp]

@heifner in which context can we optimize this. Is it where we parse a single packed_transaction, or do we sometimes parse a bunch of them?

[heifner]

@heifner in which context can we optimize this. Is it where we parse a single packed_transaction, or do we sometimes parse a bunch of them?

The interface for multiple trxs is rarely if ever used. I was thinking of the single packed_transaction use.

[greg7mdp]

Is it likely that we would retrieve the same abi repeatedly for a single transaction?

[heifner]

Is it likely that we would retrieve the same abi repeatedly for a single transaction?

Yes, mining trxs often have many of the same input actions.
For example: https://bloks.io/transaction/c65166655b17ca691c8c46f3baf7c40ef90dde5cda16c2a3cb00a6bc93ca0ce7

[greg7mdp]

Thanks. All clear now!

[greg7mdp]

@heifner , I'm not seeing where it tries to repeatedly retrieve abis when processing packed transactions?

[heifner]

@heifner , I'm not seeing where it tries to repeatedly retrieve abis when processing packed transactions?

https://github.com/AntelopeIO/leap/blob/main/plugins/chain_plugin/chain_plugin.cpp#L2189
The resolver created and used there does a find and creates a new abi_serializer for each call to the resolver. It does not cache the abi_serializer.