test_top_level_halt_burns_spilled_state_gas()

Documentation for tests/amsterdam/eip8037_state_creation_gas_cost_increase/test_state_gas_call.py::test_top_level_halt_burns_spilled_state_gas@622620d8.

Generate fixtures for these test cases for Amsterdam with:

fill -v tests/amsterdam/eip8037_state_creation_gas_cost_increase/test_state_gas_call.py::test_top_level_halt_burns_spilled_state_gas --fork Amsterdam

Verify a top-level halt burns the spilled state gas, so only the start reservoir survives. The parent calls a child that reverts or halts, then INVALIDs at the top level.

Under LIFO refills a frame's spilled state gas refills to gas_left, which the halt then zeros. Only the reservoir-funded portion survives, equal to the reservoir at frame start.

That start value equals the sized reservoir R, so for every child failure mode and reservoir_delta:

`state_gas_left_end = R`,
`tx_gas_used = tx.gas - R = gas_limit_cap`.

With the reservoir one short (reservoir_delta == -1) the child's SSTORE spills one unit from gas_left, which is refilled then burned by the halt. So tx_gas_used stays gas_limit_cap. The old behavior refunded the spill, giving gas_limit_cap - 1.

Source code in tests/amsterdam/eip8037_state_creation_gas_cost_increase/test_state_gas_call.py

@pytest.mark.parametrize(
    "reservoir_delta",
    [
        pytest.param(-1, id="reservoir_one_short"),
        pytest.param(0, id="reservoir_exact"),
        pytest.param(1, id="reservoir_one_over"),
    ],
)
@pytest.mark.parametrize(
    "child_termination",
    [
        pytest.param("revert", id="child_revert"),
        pytest.param("halt", id="child_halt"),
    ],
)
@pytest.mark.valid_from("EIP8037")
def test_top_level_halt_burns_spilled_state_gas(
    blockchain_test: BlockchainTestFiller,
    pre: Alloc,
    fork: Fork,
    child_termination: str,
    reservoir_delta: int,
) -> None:
    """
    Verify a top-level halt burns the spilled state gas, so only the
    start reservoir survives. The parent calls a child that reverts or
    halts, then INVALIDs at the top level.

    Under LIFO refills a frame's spilled state gas refills to
    `gas_left`, which the halt then zeros. Only the reservoir-funded
    portion survives, equal to the reservoir at frame start.

    That start value equals the sized reservoir R, so for every child
    failure mode and `reservoir_delta`:

        `state_gas_left_end = R`,
        `tx_gas_used = tx.gas - R = gas_limit_cap`.

    With the reservoir one short (`reservoir_delta == -1`) the child's
    SSTORE spills one unit from `gas_left`, which is refilled then
    burned by the halt. So `tx_gas_used` stays `gas_limit_cap`. The
    old behavior refunded the spill, giving `gas_limit_cap - 1`.
    """
    gas_limit_cap = fork.transaction_gas_limit_cap()
    assert gas_limit_cap is not None
    sstore_state_gas = Op.SSTORE(new_value=1).state_cost(fork)

    if child_termination == "revert":
        child_code: Bytecode = Op.SSTORE(0, 1) + Op.REVERT(0, 0)
    else:
        child_code = Op.SSTORE(0, 1) + Op.INVALID

    child = pre.deploy_contract(code=child_code)

    parent = pre.deploy_contract(
        code=(
            Op.POP(
                Op.CALL(
                    gas=child_code.execution_cost(fork) - reservoir_delta,
                    address=child,
                )
            )
            + Op.INVALID
        ),
    )

    reservoir = sstore_state_gas + reservoir_delta
    tx_gas = gas_limit_cap + reservoir

    tx = Transaction(
        to=parent,
        state_gas_reservoir=reservoir,
        sender=pre.fund_eoa(),
    )

    # LIFO refills: the spill refills to `gas_left` and is burned by
    # the halt. Only the sized reservoir survives, so
    # `tx_gas_used = gas_limit_cap`.
    state_gas_left_end = reservoir
    expected_gas_used = tx_gas - state_gas_left_end

    blockchain_test(
        pre=pre,
        blocks=[
            Block(
                txs=[tx],
                header_verify=Header(gas_used=expected_gas_used),
            ),
        ],
        post={child: Account(storage={0: 0})},
    )

Parametrized Test Cases

This test generates 6 parametrized test cases across 1 fork.