EIP1559 blockchain tests

[qbzzt]

InvalidBlocks

1. badBlocks, checks for blocks with invalid values in the header
2. feeCap, checks that the fee cap is implemented correctly
3. gasLimit20m and 4. gasLimit40m, tests cases from the geth tests https://github.com/ethereum/go-ethereum/blob/c9116a7c19301d6906c1aff87085c2fa345a84bf/consensus/misc/eip1559_test.go#L78

ValidBlocks

5. lowDemand, checks that when gasUsed << gasTarget baseFee <- baseFee*(7/8)
6. medDemand, checks that when gasUsed = gasTarget baseFee does not change
7. highDemand, checks that when gasUsed = 2gasTarget baseFee <- baseFee(9/8)
8. checkGasLimit, check that transactions that cause gasUsed to exceed gasLimit are invalid
9. tips, check that the tipped amount given to the miner is correct
10. transTypes, check that all three transaction types (original, type 1 (EIP 2930), and type 2 (EIP 1559)) work.
11. baseFee, check the behavior of basefee and selfbalance when called in multiple transactions in multiple blocks.
12. burnVerify, verify the correct amount of wei is burned

Note: The tests needed are at https://hackmd.io/NNforJS7SMmjJ1t6jA8pgQ

[winsvega]

I think we can remove the python sanitizer scripts and json schemas from here [in a separate PR]. as retesteth checks structures validity upon test generation. and also checks that tests are updated in automatition.

[winsvega]

very cool!
lets see if we have to rename gasTarget in our structures too. this will need some modification

[winsvega]

need to review

[winsvega]

please make sure that geth and retesteth are at latest
fix the exception names.
address the comments
and regenerate the tests

[winsvega]

what happened with the branch. why changes from another branch here ?

[holgerd77]

Congrats to the merge here! 🎉 😀

[holgerd77]

Hi there,
I am just integrating these tests here: ethereumjs/ethereumjs-monorepo#1269

Just noted that the base fee is still called baseFee here in the tests, the agreement here was to use baseFeePerGas, at least for the JSON RPC.

Is this due to an outdated Geth version and will be updated?

//cc @holiman

[holiman]

Yeah, we have several PRs in the pipeline, and I think rpc naming is that last one, so we're definitely not up to spec at the moment

[holgerd77]

@holiman ah ok, thanks, I've temporarily added a hot fix on our side.

[holgerd77]

We've just run the tests from here, current failures on our side (see ethereumjs/ethereumjs-monorepo#1269):

State Tests

2 failures (total 150)

• basefeeExample
• eip1559

Blockchain Tests

10 failures (total 1275)

• badBlocks
• gasLimit20m
• gasLimit40m

Haven't gone into any analysis, so totally open if (more likely) things are wrong on our side or if there might be some test bugs, just dropping here for reference and for others to compare if they also stumble upon things.

[ryanio]

Update from EthereumJS: state tests are now passing (it was within our own code)

For the Blockchain Tests, I think these failures are due to outdated versions of the spec that test >= and <= rather than > and <? (PR)

The test comments for badBlocks still reference the outdated spec:

  # From the EIP:
  #
  # assert block.gas_target >= parent_gas_target - parent_gas_target // 1024, 'invalid block: gas target decreased too much'
  ... 
  (this block:)
  - blockHeader:
      gasLimit: 1072693248    # 1024*1024*1023 (1023/1024 of the previous gasLimit)
    transactions: []

Block gasLimit: 1072693248
Max gasLimit: 1072693248

---

For gasLimit20m, it is on the same boundary:

 - blockHeader:
      gasLimit: 20019531
    transactions: []

Block gasLimit: 20019531
Max gasLimit: 20019531

---

Same with gasLimit40m:

  - blockHeader:
      gasLimit: 39960938
    transactions: []

Block gasLimit: 39960938
Min gasLimit: 39960938

[qbzzt]

Update from EthereumJS: state tests are now passing (it was within our own code)

Good. That means the tests were useful.

For the Blockchain Tests, I think these failures are due to outdated versions of the spec that test >= and <= rather than > and <? (PR)

The tests look at exactly the border line between permitted and not permitted (for example, fail at ...247, succeed at ...248):

  - blockHeader:
      gasLimit: 1072693247
    expectException:
      '>=London': 1559BlockImportImpossible_TargetGasLow
    transactions: []
  - blockHeader:
      gasLimit: 1072693248    # 1024*1024*1023 (1023/1024 of the previous gasLimit)
    transactions: []

If the borderline was different, even by one gas to either direction, the InvalidBlocks/bcEIP1559/badBlocks would have failed on geth.

To verify I downloaded the latest geth and tried badBlocks on it. It still works.

Can you run the test with verbose output, and see on which block it has an error?

Thanks,
Ori

[ryanio]

yes they look good!

I quoted the suspect blocks in my above comment, let me know if there is a better way to identify them

[qbzzt]

yes they look good!

I quoted the suspect blocks in my above comment, let me know if there is a better way to identify them

OK, sorry I didn't realize that.

   - blockHeader:
      gasLimit: 1072693248    # 1024*1024*1023 (1023/1024 of the previous gasLimit)
    transactions: []

This block is supposed to be accepted (that's why it doesn't have an expectException: field). Does it fail? If so, what is the exception? If not, does the block with a gas limit of 1072693247 fail? If so, what exception is reported to retesteth?

It's possible that the only problem is that retesteth doesn't recognize the exception.

[ryanio]

np, yes it's failing with our internal validation of invalid gas limit as it can't be exactly equal (ethereum/EIPs#3566)