Investigate gas estimation discrepancy between 0.5 and 0.6

[arhag]

Gas estimated by EOS EVM Node is lower than what will actually be consumed when executing the EVM transaction (either by EOS EVM Contract or EOS EVM Node).

Example: https://explorer.testnet.evm.eosnetwork.com/tx/0x5f6c24668a4a8f3509002754d46f7718470f47a929002a0acb57d57aeccf479c/raw-trace

EDIT: We confirmed this was because the estimation request was sent to EOS EVM Node verison 0.5. The problem seems to be fixed in 0.6, but now the question is to understand why since that fix was not planned by us.

[arhag]

We confirmed that the gas estimation from EOS EVM Node v0.6.0-rc1 is not mismatched with the gas expectations of EOS EVM Contract v0.6.0-rc1. The problem was that we were getting gas estimations from the old 0.5 version of the node, which is different than what the 0.6 version estimates, but then sending the transaction with that estimate to the new 0.6 version of EOS EVM Contract.

The upstream silkworm changes that were brought in for the 0.6 release changed the gas estimation code, so it is not too surprising for the gas estimate to change from 0.5 to 0.6.

What is a bit surprising though is that we had not noticed a discrepancy between gas estimates from EOS EVM Node and what the EOS EVM Contract expected (to the point of rejecting the transaction) until recently, and yet the fix to that discrepancy appears to have conveniently landed with the v0.6.0-rc1 of EOS EVM Node. And this happened due to the decision to bring upstream silkworm changes in for the 0.6.0 release which was independent of the trustless bridge enhancements that are being exercised by the example EVM transaction referenced above which had the gas discrepancy issue.

That may just be a lucky coincidence, but it may be worth exploring further.