Just looking at title it seems effort is similar to Tao's tx meta.
Might wanna coordinate with him or get his thoughts on this.

Cached FeeDetails should be invalidated between epoch boundary, should it? Inputs to calculate_fee_details() such as budget_limits are often feature gates dependent, so does calculation itself.

One version of solana_runtime_transaction::transaction_meta set ttl to cached data with get_last_slot_in_epoch(); While it helped to reuse cached value within epoch, the interface was rather messy: getter needs to take current_slot as additional parameter, and returns Option<>. In case of None returned (crossed epoch), callsite needs to find all necessary inputs to refresh the cached value. Is there better way?

Epoch boundaries aren't relevant to this PR refactoring because the fee that this PR is reusing is calculated during transaction loading / execution in the SVM. Epoch boundaries are only relevant when scheduling transactions in banking stage. This PR doesn't change any behavior in the banking stage and so I don't think this will conflict at all with tx meta work that @tao-stones is working on

Codecov Report

All modified and coverable lines are covered by tests ✅

Project coverage is 82.7%. Comparing base (a4a009e) to head (abe1f46).

@@           Coverage Diff           @@
##           master    #1462   +/-   ##
=======================================
  Coverage    82.7%    82.7%           
=======================================
  Files         876      876           
  Lines      371118   371045   -73     
=======================================
- Hits       307263   307212   -51     
+ Misses      63855    63833   -22     

My early comment is due to the concern that dev could accidentally cache and reuse LoadedTransaction (at banking stage) cross epoch boundary, even it's not intentional. Perhaps that's overly concerned?

Yeah, I don't think it's a valid concern because basically nothing in LoadedTransaction should be loaded during one bank and then used in a new bank, let alone across an epoch boundary.

automerge label removed due to a CI failure