Can't mine more than one block when transitioning from another VM to London

[kclowes]

What was wrong?

This test: https://github.com/ethereum/eth-tester/blob/master/tests/backends/test_pyevm.py#L33-L56

Fails if there is a transition to the LondonVM and more than one block is mined on London. The stacktrace looks like:

    def test_custom_virtual_machines():
        if not is_pyevm_available():
            pytest.skip("PyEVM is not available")

        backend = PyEVMBackend(vm_configuration=(
            (0, FrontierVM),
            (3, LondonVM),
        ))

        # This should be a FrontierVM block
        VM_at_2 = backend.chain.get_vm_class_for_block_number(2)
        # This should be a LondonVM block
        VM_at_3 = backend.chain.get_vm_class_for_block_number(3)

        assert issubclass(VM_at_2, FrontierVM)
        assert not issubclass(VM_at_2, LondonVM)
        assert issubclass(VM_at_3, LondonVM)

>       backend.mine_blocks(3)

VM_at_2    = <class 'abc.FrontierVM'>
VM_at_3    = <class 'abc.LondonVM'>
backend    = <eth_tester.backends.pyevm.main.PyEVMBackend object at 0x7fa0e93c2a90>

tests/backends/test_pyevm.py:52:
_ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ 
.venv-et-iii/lib/python3.9/site-packages/eth_utils/functional.py:45: in inner
    return callback(fn(*args, **kwargs))
eth_tester/backends/pyevm/main.py:429: in mine_blocks
    block = self.chain.mine_block(**mine_kwargs)
.venv-et-iii/lib/python3.9/site-packages/eth/chains/base.py:703: in mine_block
    return self.mine_block_extended(*args, **kwargs).block
.venv-et-iii/lib/python3.9/site-packages/eth/chains/base.py:709: in mine_block_extended
    mine_result = vm.mine_block(current_block, *args, **kwargs)
.venv-et-iii/lib/python3.9/site-packages/eth/vm/base.py:336: in mine_block
    self.validate_block(block_result.block)
.venv-et-iii/lib/python3.9/site-packages/eth/vm/base.py:563: in validate_block
    self.validate_header(block.header, parent_header)
.venv-et-iii/lib/python3.9/site-packages/eth/vm/base.py:604: in validate_header
    cls.validate_gas(header, parent_header)
.venv-et-iii/lib/python3.9/site-packages/eth/vm/forks/london/__init__.py:55: in validate_gas
    validate_gas_limit(header.gas_limit, parent_gas_limit)
_ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ 

gas_limit = 30029122, parent_gas_limit = 60058244

    def validate_gas_limit(gas_limit: int, parent_gas_limit: int) -> None:
        low_bound, high_bound = compute_gas_limit_bounds(parent_gas_limit)
        if gas_limit < low_bound:
>           raise ValidationError(
                f"The gas limit {gas_limit} is too low. It must be at least {low_bound}"
E               eth_utils.exceptions.ValidationError: The gas limit 30029122 is too low. It must be at least 59999595

How can it be fixed?

Not sure yet.

[fselmo]

Moving the conversation here but it seems like this all stems from this issue and ultimately this PR.

Removing these lines will get this test to pass again and will break 2 tests related to keeping the gas limit constant. I'm not sure why we wouldn't want to test the actual behavior of the gas limit over keeping it constant but I suppose we can discuss that here.