fix(evm): The estimated gas and the gas used for executing the message are not consistent

[cuiweixie]

Description

when execute a eth msg, the kv gas config is zeros, see comments.
but gas config is not zero in estimate gas.
these will cause estimate gas return more gas that execution really need when call precompile smart contract.
set the estimate gas to same as in execution.

---

Closes #XXX

[codecov]

Codecov Report

Merging #1943 (7abbc04) into main (b826687) will increase coverage by 0.28%.
The diff coverage is 88.88%.

❗ Current head 7abbc04 differs from pull request most recent head be75ba3. Consider uploading reports for the commit be75ba3 to get more accurate results

@@            Coverage Diff             @@
##             main    #1943      +/-   ##
==========================================
+ Coverage   69.70%   69.99%   +0.28%     
==========================================
  Files         331      327       -4     
  Lines       24514    24292     -222     
==========================================
- Hits        17088    17003      -85     
+ Misses       6543     6417     -126     
+ Partials      883      872      -11     

Files	Coverage Δ
x/erc20/keeper/evm.go	92.15% <100.00%> (ø)
x/evm/keeper/grpc_query.go	86.73% <88.00%> (-0.02%)	⬇️

... and 8 files with indirect coverage changes

[GAtom22]

Thanks for the contribution @cuiweixie!!

We appreciate you reporting this. However, this is actually not a solution to the issue. The EstimateGas is not only used on the eth_estimateGas rpc API, but also in CallEVM function. The result is that the gasUsed and estimation are both less with these changes, but still have the same difference between them. Check out these results from this test:

Test with the changes proposed here:

FAILED test_precompiles.py::test_staking - AssertionError: assert 108073 (gasUsed) == 118573 (gas estimation)

Test with current code on main:

FAILED test_precompiles.py::test_staking - AssertionError: assert 205647 (gasUsed) == 216147  (gas estimation)

The difference in both cases is 10500

[cuiweixie]

Thanks for the contribution @cuiweixie!!

We appreciate you reporting this. However, this is actually not a solution to the issue. The EstimateGas is not only used on the eth_estimateGas rpc API, but also in CallEVM function. The result is that the gasUsed and estimation are both less with these changes, but still have the same difference between them. Check out these results from this test:

Test with the changes proposed here:

FAILED test_precompiles.py::test_staking - AssertionError: assert 108073 (gasUsed) == 118573 (gas estimation)

Test with current code on main:

FAILED test_precompiles.py::test_staking - AssertionError: assert 205647 (gasUsed) == 216147  (gas estimation)

The difference in both cases is 10500

I will try to fix it in right way. If you know the right way to fix this, please tell me. I will update my pr.

[cuiweixie]

Thanks for the contribution @cuiweixie!!
We appreciate you reporting this. However, this is actually not a solution to the issue. The EstimateGas is not only used on the eth_estimateGas rpc API, but also in CallEVM function. The result is that the gasUsed and estimation are both less with these changes, but still have the same difference between them. Check out these results from this test:
Test with the changes proposed here:

FAILED test_precompiles.py::test_staking - AssertionError: assert 108073 (gasUsed) == 118573 (gas estimation)

Test with current code on main:

FAILED test_precompiles.py::test_staking - AssertionError: assert 205647 (gasUsed) == 216147  (gas estimation)

The difference in both cases is 10500

I will try to fix it in right way. If you know the right way to fix this, please tell me. I will update my pr.

@GAtom22 in the new codebase, the nix_tests(test_staking in test_precompiles.py) can passed now.

[GAtom22]

Please add a changelog entry

[CLAassistant]

All committers have signed the CLA.

[GAtom22]

Added a new test where there's a smart contract that calls the staking EVM extension, and in this case we're still having the same issue. The gas estimation is higher than the gas used:

# On the approval tx
gas used: 56951
gas estimation: 57049

# On the delegate tx
gas used: 121735
gas estimation: 122917

[cuiweixie]

Added a new test where there's a smart contract that calls the staking EVM extension, and in this case we're still having the same issue. The gas estimation is higher than the gas used:

# On the approval tx
gas used: 56951
gas estimation: 57049

# On the delegate tx
gas used: 121735
gas estimation: 122917

@GAtom22 I find this may not be a issue.
after i change callGas in go-ethereum.

the testDelegate call is same now.

# On the approval tx
gas used: 56951
gas estimation: 56921

# On the delegate tx
gas used: 120730
gas estimation: 120730

i think the difference is caused by same issue in above link.
gas used in less that gas limit in gas estimation.
but if we change gas limit to gas used that can make tx run success,
the tx will failed.

what 's more!
just found that the first approval tx to same address is no equal.

estimate gas is binary search the gas limit. when find a fake minimal gasLimit, it doesn't means that the
gasLimit == gasUsed, gasUsed may less than gasLimit, because the dynamic gas stragety.

[GAtom22]

tACK! Thanks for the contribution @cuiweixie!

[0xstepit]

Great job @cuiweixie, and thanks a lot for contributing to our codebase and improving it!