Background

To check the effects of Gas cost increases for state access opcodes, a modified geth-version with support for the to-be-eip was used.

Setup

The experiment was run on Goerli blocks from number=3321372 to number=3321425, using a fast-synced geth with gcmode=archive, thus not pruning state.

A modified version of standardTraceBlockToFile was added to the geth instance, making it possible to trace blocks using alternate rules, like so:

    debug.standardTraceBlockToFile(blockhash,{"overrides":{
        "yoloV2Block": 0
        }})

The analysis.js was loaded, and the traces were generated (those are 500+Mb, I decided not to include them in this repo), containing the trace for 53 blocks, totalling 106 transactions.

The experiment

When running with new rules, two things can typically happen:

1. The execution flow is broken. The easiest way to detect this is to compare the number of
   operations executed. If the exact same operations were executed in both traces, the flow was not broken.
2. The execution flow was not broken.

For 1), it is interesting to analyse:

A. Would this particular execution have been 'salvaged' if more gas had been provided by the external caller? B. If not, would the POKE precompile have 'salvaged' it? C. If not, would the tx-access list have 'salvaged' it?

For 2), it is interesting to analyse:

A) What was the difference in gas used by the transaction? Did the alternative transaction consume more or less?

Goerli results.

There were 100 transactions that were OK, as in, the execution finished the same way in both the 'canon' version and the 'alt' (gas-cost-increase version):

txhash	steps(canon)	steps(alt)	OK?	gas(canon)	gas(alt)	diff
0x26888356	15570	15570	ok	469226	518626	10.00 %
0x9f795798	1	1	ok	0	0	0.00 %
0xf89eb38e	340	340	ok	31967	34367	7.00 %
0xa0084529	15630	15630	ok	412982	425182	2.00 %
0xcf9553b2	15630	15630	ok	412982	425182	2.00 %
0x5768df5e	1	1	ok	0	0	0.00 %
0xb33b6d6d	1	1	ok	0	0	0.00 %
0xb6dce779	15630	15630	ok	412982	425182	2.00 %
0x20313f8f	1411	1411	ok	76144	89944	18.00 %
0xbf98737d	1	1	ok	0	0	0.00 %
0x04b95f1b	258	258	ok	29362	29862	1.00 %
0x5867d811	15630	15630	ok	412982	425182	2.00 %
0x6e4496c7	15630	15630	ok	412982	425182	2.00 %
0x7e005bb0	1	1	ok	0	0	0.00 %
0x84a35acb	15630	15630	ok	412982	425182	2.00 %
0xe10171a6	1	1	ok	0	0	0.00 %
0x3fd24de9	377	377	ok	15488	15188	-2.00 %
0x3cca6923	377	377	ok	15488	15188	-2.00 %
0xd787848a	377	377	ok	15488	15188	-2.00 %
0xb70224df	15630	15630	ok	412982	425182	2.00 %
0x8f60fe0d	377	377	ok	15488	15188	-2.00 %
0xcd7e7ce6	15630	15630	ok	412982	425182	2.00 %
0x5891f568	6325	6325	ok	298960	312160	4.00 %
0x578a51ef	23990	23990	ok	906184	953884	5.00 %
0x848ebe81	258	258	ok	29362	29862	1.00 %
0x9fdf8667	377	377	ok	15488	15188	-2.00 %
0xef2b2bf2	258	258	ok	29362	29862	1.00 %
0x6770fd04	1	1	ok	0	0	0.00 %
0xf276ac64	6325	6325	ok	298960	312160	4.00 %
0x94826b3d	1	1	ok	0	0	0.00 %
0xf7c56b8f	15630	15630	ok	432182	445082	2.00 %
0x8973962a	6332	6332	ok	298980	307680	2.00 %
0x171ab5c7	6257	6257	ok	169712	155612	-9.00 %
0x9577d126	258	258	ok	29362	29862	1.00 %
0x9301dd6d	6325	6325	ok	298960	312160	4.00 %
0x2c10f55a	258	258	ok	29362	29862	1.00 %
0x44392316	1	1	ok	0	0	0.00 %
0x75daeceb	258	258	ok	29362	27862	-6.00 %
0x0222d6cf	15630	15630	ok	432182	445082	2.00 %
0x44df7cfa	377	377	ok	15488	9188	-41.00 %
0xbdc1c6cf	6325	6325	ok	298960	312160	4.00 %
0x23029eec	6325	6325	ok	298960	268160	-11.00 %
0x2c207531	15630	15630	ok	412982	425182	2.00 %
0x1a8d48e5	6325	6325	ok	298960	312160	4.00 %
0x37f83d52	258	258	ok	29362	27862	-6.00 %
0x5395285e	1	1	ok	0	0	0.00 %
0xadfacf22	502	502	ok	19854	24754	24.00 %
0x9331a030	1	1	ok	0	0	0.00 %
0x5b7ea6ac	377	377	ok	15488	15188	-2.00 %
0xce51e39e	15630	15630	ok	412982	425182	2.00 %
0x27147082	377	377	ok	15488	15188	-2.00 %
0x9e4b02ee	258	258	ok	29362	29862	1.00 %
0x561ca5d8	6332	6332	ok	298980	312180	4.00 %
0x9ea5e140	6332	6332	ok	298980	268180	-11.00 %
0xdbead36b	377	377	ok	15488	15188	-2.00 %
0x8c33128d	258	258	ok	29362	29862	1.00 %
0x0150ed20	340	340	ok	12767	12467	-3.00 %
0x0f9a5a0f	377	377	ok	15488	15188	-2.00 %
0x26eecdb0	258	258	ok	29362	29862	1.00 %
0x68622456	6325	6325	ok	298960	268160	-11.00 %
0x7a180660	1	1	ok	0	0	0.00 %
0xd356e16c	15630	15630	ok	432182	445082	2.00 %
0xda048c47	377	377	ok	15488	15188	-2.00 %
0xdf322b2a	1	1	ok	0	0	0.00 %
0xa66cdfe8	1	1	ok	0	0	0.00 %
0x550770b3	15731	15731	ok	1149087	1209187	5.00 %
0xcf57527b	6257	6257	ok	169712	155612	-9.00 %
0x113897c1	377	377	ok	15488	15188	-2.00 %
0xb07f73f6	258	258	ok	29362	27862	-6.00 %
0xe1883f43	1	1	ok	0	0	0.00 %
0xf07e2c52	210	210	ok	1518087	1518087	0.00 %
0x55c2d6df	377	377	ok	15488	15188	-2.00 %
0x7499547c	15570	15570	ok	469226	518626	10.00 %
0xb917dc67	1	1	ok	0	0	0.00 %
0x420cd0e2	6332	6332	ok	298980	307680	2.00 %
0x90b1ec09	377	377	ok	15488	15188	-2.00 %
0xd5d7cf5c	6257	6257	ok	169712	150612	-12.00 %
0xfadc30ea	15630	15630	ok	412982	425182	2.00 %
0x0dffecb4	6325	6325	ok	298960	312160	4.00 %
0xf7151f6f	258	258	ok	29362	29862	1.00 %
0x5c72a21a	15630	15630	ok	412982	425182	2.00 %
0x1d29f0b2	377	377	ok	15488	15188	-2.00 %
0xcdd7a3d5	377	377	ok	15488	15188	-2.00 %
0x672f1027	1	1	ok	0	0	0.00 %
0x2f4c2c25	258	258	ok	29362	27862	-6.00 %
0x60f5268a	1	1	ok	0	0	0.00 %
0xac1c6cfe	377	377	ok	15488	15188	-2.00 %
0xd41659a7	1	1	ok	0	0	0.00 %
0x71435959	1	1	ok	0	0	0.00 %
0x10e72aa7	6257	6257	ok	169712	150612	-12.00 %
0xfd89999c	6243	6243	ok	169658	155558	-9.00 %
0xd690c177	15630	15630	ok	412982	425182	2.00 %
0xadef95d2	1	1	ok	0	0	0.00 %
0x9e1662e4	377	377	ok	15488	15188	-2.00 %
0xa97ab7a7	1	1	ok	0	0	0.00 %
0x643d441f	1	1	ok	0	0	0.00 %
0x50f7d372	6332	6332	ok	298980	312180	4.00 %
0xff5bbcff	15731	15731	ok	1149087	1209187	5.00 %

OK: 98, broken: 8
Canon gas used: 17862517, alt gas used: 18284517

And 8 transactions were 'broken':

txhash	steps(canon)	steps(alt)	OK?	gas(canon)	gas(alt)	diff
0xd67b5067	557	547	BROKEN	46354	46354	0.00 %
0x8574e95d	3297	3187	BROKEN	519530	519530	0.00 %
0xfc2b04ed	4775	2638	BROKEN	199854	194824	-3.00 %
0xd96c736a	6044	274	BROKEN	124673	158954	27.00 %
0xcbd02f65	6044	274	BROKEN	124673	158954	27.00 %
0xe85c0e23	369	362	BROKEN	61215	61215	0.00 %
0xaeb93809	15119	12190	BROKEN	400331	384420	-4.00 %
0xef39d3b1	64117	61159	BROKEN	4398883	4401415	0.00 %

Broken flows

This section was written with traces generated by an earlier implementation of the spec, which undercharged some calls by 700, and where only 6 txs failed. It has not been updated with the data from the regenerated table above

Let's look deeper into the broken ones:

0xd67b5067

block_0x19ef4a84-2-0xd67b5067-742912372:

OperationsOperations
STEP      PC       OPNAME  OPCODE GAS  GASCOST DEPTH REFUND STEP      PC      OPNAME OPCODE GAS  GASCOST DEPTH REFUND    
538  2724 (0xaa4) POP      0x50   8187 2       1     0      528  2714 (0xa9a) ADD    0x1    5714 3       1     0         
539  2725 (0xaa5) POP      0x50   8185 2       1     0      529  2715 (0xa9b) DUP3   0x82   5711 3       1     0         
540  2726 (0xaa6) POP      0x50   8183 2       1     0      530  2716 (0xa9c) SWAP1  0x90   5708 3       1     0         
541  2727 (0xaa7) POP      0x50   8181 2       1     0      531  2717 (0xa9d) SUB    0x3    5705 3       1     0         
542  2728 (0xaa8) POP      0x50   8179 2       1     0      532  2718 (0xa9e) SWAP11 0x9a   5702 3       1     0         
543  2729 (0xaa9) POP      0x50   8177 2       1     0      533  2719 (0xa9f) POP    0x50   5699 2       1     0         
544  2730 (0xaaa) POP      0x50   8175 2       1     0      534  2720 (0xaa0) SWAP1  0x90   5697 3       1     0         
545  2731 (0xaab) LOG1     0xa1   8173 8150    1     0      535  2721 (0xaa1) SWAP9  0x98   5694 3       1     0         
546  2732 (0xaac) POP      0x50   23   2       1     0      536  2722 (0xaa2) POP    0x50   5691 2       1     0         
547  2733 (0xaad) POP      0x50   21   2       1     0      537  2723 (0xaa3) POP    0x50   5689 2       1     0         
548  2734 (0xaae) POP      0x50   19   2       1     0      538  2724 (0xaa4) POP    0x50   5687 2       1     0         
549  2735 (0xaaf) POP      0x50   17   2       1     0      539  2725 (0xaa5) POP    0x50   5685 2       1     0         
550  2736 (0xab0) POP      0x50   15   2       1     0      540  2726 (0xaa6) POP    0x50   5683 2       1     0         
551  2737 (0xab1) POP      0x50   13   2       1     0      541  2727 (0xaa7) POP    0x50   5681 2       1     0         
552  2738 (0xab2) POP      0x50   11   2       1     0      542  2728 (0xaa8) POP    0x50   5679 2       1     0         
553  2739 (0xab3) JUMP     0x56   9    8       1     0      543  2729 (0xaa9) POP    0x50   5677 2       1     0         
554  426 (0x1aa)  JUMPDEST 0x5b   1    1       1     0      544  2730 (0xaaa) POP    0x50   5675 2       1     0         
555  427 (0x1ab)  STOP     0x0    0    0       1     0      545  2731 (0xaab) LOG1   0xa1   5673 8150    1     0         

This tx breaks on a LOG1 instruction on the outermost depth. Therefore, it would have succeeded with some more gas from the EOA. It can also be noted that the canon tx also went OOG, so technically there was no change.

0x8574e95d

This tx also goes OOG in the uppermost depth, at step 3185, where it hits an SSTORE. The canon tx has 63K gas left at this point, but the alt has only 11K. This is also a situation where more gas from outside would have salvaged the call.

It can also be noted that for this transaction too, the canon tx also went OOG, so technically there was no change.

0xfc2b04ed

This tx is actually different - the canon tx does not go OOG. The canon transaction does a series of deep calls, and goes through 4773 steps of execution. The 'alt' transaction fails at step 2638. However, the execution differs even earlier :

OperationsOperations
STEP       PC        OPNAME  OPCODE  GAS  GASCOST DEPTH EFUSTEP       PC           OPNAME     OPCODE GAS  GASCOST DEPTH 
2591 18041 (0x4679) SWAP2    0x91   43718 3       4     0   2591 18041 (0x4679) SWAP2          0x91   4571 3       4     
2592 18042 (0x467a) SSTORE   0x55   43715 800     4     0   2592 18042 (0x467a) SSTORE         0x55   4568 2100    4     
2593 18043 (0x467b) PUSH1    0x60   42915 3       4     0   2593 18043 (0x467b) PUSH1          0x60   2468 3       4     
2594 18045 (0x467d) SSTORE   0x55   42912 5000    4     0   2594 18045 (0x467d) SSTORE         0x55   2465 3000    4     
2595 18046 (0x467e) JUMP     0x56   37912 8       4     0   2595 607 (0x25f)    RETURNDATASIZE 0x3d   8097 2       3     
2596 4899 (0x1323)  JUMPDEST 0x5b   37904 1       4     0   2596 608 (0x260)    PUSH1          0x60   8095 3       3     
2597 4900 (0x1324)  PUSH2    0x61   37903 3       4     0   2597 610 (0x262)    MLOAD          0x51   8092 3       3     
2598 4903 (0x1327)  DUP2     0x81   37900 3       4     0   2598 611 (0x263)    DUP2           0x81   8089 3       3     

At step 2594, a call at depth 4 exits with OOG. The call was initiated at step 1229, and is a DELEGATECALL with gas not being hardcoded, but GAS.

Call ladder:

1. delegatecall(GAS)
2. call(GAS)
3. delegatecall(GAS)

Since the gas is never restricted, it appears that this tx would have been salvageable with more outer gas.

0x8574e95d

The 0x8574e95d transaction has 3295 canon steps, but exits early after 3185 steps on 'alt', where an SSTORE occurs, causing OOG. The event happens on depth=1, meaning that more outer gas would have salvaged the flow.

0xfc2b04ed

The 0xfc2b04ed tx has 4773 steps, but the 'alt' is aborted after '2638' steps. However, the execution flow is disrupted at step 2594, where an SSTORE at depth=4 causes a subcall to go OOG:

OperationsOperations
STEP       PC        OPNAME  OPCODE  GAS  GASCOST DEPTH EFUSTEP       PC           OPNAME     OPCODE GAS  GASCOST DEPTH 
2591 18041 (0x4679) SWAP2    0x91   43718 3       4     0   2591 18041 (0x4679) SWAP2          0x91   4571 3       4     
2592 18042 (0x467a) SSTORE   0x55   43715 800     4     0   2592 18042 (0x467a) SSTORE         0x55   4568 2100    4     
2593 18043 (0x467b) PUSH1    0x60   42915 3       4     0   2593 18043 (0x467b) PUSH1          0x60   2468 3       4     
2594 18045 (0x467d) SSTORE   0x55   42912 5000    4     0   2594 18045 (0x467d) SSTORE         0x55   2465 3000    4     
2595 18046 (0x467e) JUMP     0x56   37912 8       4     0   2595 607 (0x25f)    RETURNDATASIZE 0x3d   8097 2       3     
2596 4899 (0x1323)  JUMPDEST 0x5b   37904 1       4     0   2596 608 (0x260)    PUSH1          0x60   8095 3       3     
2597 4900 (0x1324)  PUSH2    0x61   37903 3       4     0   2597 610 (0x262)    MLOAD          0x51   8092 3       3     
2598 4903 (0x1327)  DUP2     0x81   37900 3       4     0   2598 611 (0x263)    DUP2           0x81   8089 3       3     
2599 4904 (0x1328)  DUP6     0x85   37897 3       4     0   2599 612 (0x264)    PUSH1          0x60   8086 3       3     
2600 4905 (0x1329)  DUP12    0x8b   37894 3       4     0   2600 614 (0x266)    DUP3           0x82   8083 3       3     
2601 4906 (0x132a)  DUP10    0x89   37891 3       4     0   2601 615 (0x267)    RETURNDATACOPY 0x3e   8080 3       3     

Call ladder:

1. DCALL(GAS-10000) to 0x6b8a780124a2259b... (ctx: NA)
2. CALL(GAS) to 0x5b67626cbf5d1677...
3. DCALL(GAS) to 0xfe488f251a7c6e63... (ctx: 0x5b67626cbf5d1677...)

It seems that this one would have been salvaged by more external gas.

0xe85c0e23

The 0xe85c0e23 goes OOG at step 360, where it does a LOG2 operation, where the 'canon' trace continues another 7 steps. This happens at depth=1, meaning that more external gas would have solved it.

0xaeb93809

This is a very long trace, reaching depths of 6. The execution flow is diverted at depth 6, on step 12405, where an SSTORE causes OOG:

OperationsOperations
STEP        PC        OPNAME  OPCODE  GAS  GASCOST DEPTH EFSTEP        PC           OPNAME     OPCODE GAS  GASCOST DEPTH
12398 9075 (0x2373)  POP      0x50   50012 2       6     0  12398 9075 (0x2373)  POP            0x50   1275 2       6    
12399 9076 (0x2374)  POP      0x50   50010 2       6     0  12399 9076 (0x2374)  POP            0x50   1273 2       6    
12400 9077 (0x2375)  JUMP     0x56   50008 8       6     0  12400 9077 (0x2375)  JUMP           0x56   1271 8       6    
12401 10784 (0x2a20) JUMPDEST 0x5b   50000 1       6     0  12401 10784 (0x2a20) JUMPDEST       0x5b   1263 1       6    
12402 10785 (0x2a21) PUSH1    0x60   49999 3       6     0  12402 10785 (0x2a21) PUSH1          0x60   1262 3       6    
12403 10787 (0x2a23) DUP2     0x81   49996 3       6     0  12403 10787 (0x2a23) DUP2           0x81   1259 3       6    
12404 10788 (0x2a24) SWAP1    0x90   49993 3       6     0  12404 10788 (0x2a24) SWAP1          0x90   1256 3       6    
12405 10789 (0x2a25) SSTORE   0x55   49990 5000    6     0  12405 10789 (0x2a25) SSTORE         0x55   1253 0       6    
12406 10790 (0x2a26) POP      0x50   44990 2       6     0  12406 684 (0x2ac)    RETURNDATASIZE 0x3d   9897 2       5    
12407 10791 (0x2a27) PUSH1    0x60   44988 3       6     0  12407 685 (0x2ad)    PUSH1          0x60   9895 3       5    
12408 10793 (0x2a29) PUSH2    0x61   44985 3       6     0  12408 687 (0x2af)    MLOAD          0x51   9892 3       5    
12409 10796 (0x2a2c) DUP3     0x82   44982 3       6     0  12409 688 (0x2b0)    DUP2           0x81   9889 3       5    
12410 10797 (0x2a2d) DUP7     0x86   44979 3       6     0  12410 689 (0x2b1)    PUSH1          0x60   9886 3       5    
12411 10798 (0x2a2e) PUSH2    0x61   44976 3       6     0  12411 691 (0x2b3)    DUP3           0x82   9883 3       5    
12412 10801 (0x2a31) SWAP1    0x90   44973 3       6     0  12412 692 (0x2b4)    RETURNDATACOPY 0x3e   9880 3       5    
12413 10802 (0x2a32) SWAP2    0x91   44970 3       6     0  12413 693 (0x2b5)    DUP3           0x82   9877 3       5    

Call ladder:

call 4  DCALL to 0xfd259808f4c08b12... 
	- gas = `GAS-10K`
call 3  CALL to 0xf5e574045298ee3a... 
	- gas = `GAS`
call 2  DCALL to 0x25723f81ebf997f3... 
	- gas = `GAS-10K`
call 1  CALL to 0x00200ea4ee292e25... 
	- gas = `GAS`
call 0  DCALL to 0xce29aedcdbeef0b0... 
	- gas = `GAS-10K`

Since this uses GAS (minus some constant), it seems likely that this could also be salvaged by using more external gas.

0xef39d3b1

This is another extremely long trace, with 60K+ steps. At step 61764, at depth 2, an SLOAD causes the 'alt'-trace to go OOG. The depth 2 was reached via a DELEGATECALL(GAS...), so it appears that this would also have been salvageable via more external gas.

Summary

The test on Goerli, covering 106 transactions, showed

• Out of 106 transactions, 98 were not meaningfully affected by the change,
• For those 98 transactions,
  • The 'canon' gas used was 17862517,
  • and the 'alt' gas used was 18284517, which is +2.36%.
• Out of the 8 failing transactions,
  • 2 went OOG in the 'canon' version aswell as in the alt version
  • All 6 seems to have been salvageable by using more external gas by the caller. (OBS The manual analysis is not updated after rerun with new constants)

Note: the gas here refers to the gas used by the evm execution, and does not take into concern the intrinsic gas used by the envelope transaction. Thus, a non-executing raw ether-send is listed as 0 gas.