fix: retryable excess gas refund

[gzeoneth]

When we have _gasPriceBid higher than actual L2 gas price, or otherwise have deposit > l2callvalue + maxSubmissionCost + gascost, our refund is insufficient because we only refund gasPrice * gasLeft, even if gasPrice < _gasPriceBid

nitro/arbos/tx_processor.go

Line 335 in 69fd4e3

refund := arbmath.BigMulByUint(gasPrice, gasLeft)

We should refund excess gas provided from L1, i.e.
max(deposit - l2callvalue - maxSubmissionCost - gascost, 0)

The retryable can use existing fund in L2 from account to provide gas, but the refund will be bounded by excessDeposit - gascost.

[codecov]

Codecov Report

Merging #635 (349b1ba) into master (69fd4e3) will decrease coverage by 4.64%.
The diff coverage is 75.00%.

@@            Coverage Diff             @@
##           master     #635      +/-   ##
==========================================
- Coverage   52.32%   47.67%   -4.65%     
==========================================
  Files         210      199      -11     
  Lines       25080    21065    -4015     
  Branches      486      486              
==========================================
- Hits        13122    10042    -3080     
+ Misses      10635     9699     -936     
- Partials     1323     1324       +1     

[PlasmaPower]

I'm not sure we want to move excess deposit to the fee refund address. I thought the idea was that the deposit would go to the L2 address.

[gzeoneth]

I'm not sure we want to move excess deposit to the fee refund address. I thought the idea was that the deposit would go to the L2 address.

well that's the behaviour in classic and also documented
https://developer.offchainlabs.com/docs/l1_l2_messages#other-parameters

Credit-Back Address: Address to which all excess gas is credited on L2; i.e., excess ETH for base submission cost (MaxSubmissionCost - ActualSubmissionCostPaid) and excess ETH provided for L2 execution ((GasPrice x MaxGas) - ActualETHSpentInExecution).

or do you mean we should refund max(min(deposit - l2callvalue - maxSubmissionCost, gasprice * maxgas) - gascost, 0) instead

[gzeoneth]

discussed offline, I think the conclusion is to refund excessSubmissionCost + gasprice * maxgas - gascost
constrained by gasrefund + l2callvalue + maxSubmissionCost <= deposit