Re-execution of a transaction currently requires using TransactionExecutor::execute_transaction. This is an unnecessary hurdle because it requires a DataStore and a TransactionAuthenticator impl. The only thing that should be necessary is a MastForestStore impl as required by the VM.

If I understand correctly, re-execution of transactions during proving will go away once the necessary VM APIs are exposed, but re-execution itself is basically a solved problem in LocalTransactionProver::prove, which doesn't require a DataStore or TransactionAuthenticator. We should be able to repurpose the way this is set up and the prover host implementation for the purpose of a re-execution API in miden-tx that miden node would be able to use.

I think a separate implementation like this is preferable to re-using TransactionExecutor for this, as it would be overloaded with the concerns of execution and re-execution, which would introduce complication at various points:

• Foreign account loading happens only when the kernel emits the miden::account::before_foreign_load event. Skipping this just based on whether the foreign account id's advice map key is present isn't enough. Additionally:
  • the foreign account code from TransactionInputs::foreign_account_code would have to be loaded during tx preparation.
  • the account procedure index map would have to be built with TransactionInputs::foreign_account_code as well. Currently it isn't because this field is assumed to be empty when executing.
• The DataStore interface is not designed for re-execution and many of its methods shouldn't be called during re-execution since all data should already be present. However, building an API on top of TransactionExecutor for re-execution requires implementing these, and we'd have to do this in some crude way, e.g. always returning errors à la "should not be called".
• We'll probably have to do more things like this one in the tx executor to make it fit for re-execution.

Separately, it seems like TransactionInputs has too many concerns at once that make it harder to understand. That seems like a consequence of it being used for execution and re-execution (mentioned also by @igamigo here):

• TransactionInputs::foreign_account_code being empty before execution, but containing code after execution.
• Advice inputs being overwritten with the contents from the advice provider after execution.

One initial idea to address these two is to bring back TransactionWitness for the purpose of re-execution to make this concern clearer. ExecutedTransaction would again be convertible into TransactionWitness and miden-node submit tx API would take this as an input next to a ProvenTransaction for the purpose of re-execution. We can then add TransactionWitness::execute(&self) -> Result<ExecutedTransaction, _> to re-execute. As mentioned in the beginning, this would take the LocalTransactionProver infrastructure as a basis for re-execution. This is just an initial idea, though.

Context: 0xMiden/node#1493 (comment)