From fa15a68050e3e6b86b1ce0ca81ee393933cad9e8 Mon Sep 17 00:00:00 2001 From: Alexandr Tarelkin Date: Tue, 8 Aug 2023 14:44:49 +0300 Subject: [PATCH] review updates --- scripts/vote_2023_08_08.py | 12 ++++++------ utils/test/event_validators/common.py | 2 -- 2 files changed, 6 insertions(+), 8 deletions(-) diff --git a/scripts/vote_2023_08_08.py b/scripts/vote_2023_08_08.py index 919c1dda..cdb9c3ee 100644 --- a/scripts/vote_2023_08_08.py +++ b/scripts/vote_2023_08_08.py @@ -4,11 +4,11 @@ 1. Add Rewards Share Program top up EVM script factory 0xbD08f9D6BF1D25Cc7407E4855dF1d46C2043B3Ea 2. Add Rewards Share Program add recipient EVM script factory 0x1F809D2cb72a5Ab13778811742050eDa876129b6 3. Add Rewards Share Program remove recipient EVM script factory 0xd30Dc38EdEfc21875257e8A3123503075226E14B -4. Grant MANAGE_NODE_OPERATOR_ROLE to Agent +4. Grant NodeOperatorsRegistry.MANAGE_NODE_OPERATOR_ROLE to Agent 5. Add Launchnodes node operator with reward address 0x5a8B929EDBf3CE44526465DD2087EC7EFB59A561 6. Add SenseiNode node operator with reward address 0xE556Da28015c04F35A52B3111B9F4120E908056e 7. Set 3.1531 stETH as the allowance of Burner over the Agent's tokens -8. Grant REQUEST_BURN_MY_STETH_ROLE to Agent +8. Grant Burner.REQUEST_BURN_MY_STETH_ROLE to Agent 9. Request to burn 3.1531 stETH for cover """ @@ -75,7 +75,7 @@ def start_vote(tx_params: Dict[str, str], silent: bool = False) -> Tuple[int, Op factory=rewards_share_remove_recipient_factory, permissions=create_permissions(rewards_share_registry, "removeRecipient"), ), - # 4. Grant MANAGE_NODE_OPERATOR_ROLE to Agent + # 4. Grant NodeOperatorsRegistry.MANAGE_NODE_OPERATOR_ROLE to Agent encode_permission_grant( grant_to=contracts.agent, target_app=contracts.node_operators_registry, @@ -94,7 +94,7 @@ def start_vote(tx_params: Dict[str, str], silent: bool = False) -> Tuple[int, Op ) ] ), - # 8. Grant REQUEST_BURN_MY_STETH_ROLE to Agent + # 8. Grant Burner.REQUEST_BURN_MY_STETH_ROLE to Agent agent_forward( [ ( @@ -118,11 +118,11 @@ def start_vote(tx_params: Dict[str, str], silent: bool = False) -> Tuple[int, Op "1) Add Rewards Share Program top up EVM script factory 0xbD08f9D6BF1D25Cc7407E4855dF1d46C2043B3Ea", "2) Add Rewards Share Program add recipient EVM script factory 0x1F809D2cb72a5Ab13778811742050eDa876129b6", "3) Add Rewards Share Program remove recipient EVM script factory 0xd30Dc38EdEfc21875257e8A3123503075226E14B", - "4) Grant MANAGE_NODE_OPERATOR_ROLE to Agent", + "4) Grant NodeOperatorsRegistry.MANAGE_NODE_OPERATOR_ROLE to Agent", "5) Add Launchnodes node operator with reward address 0x5a8B929EDBf3CE44526465DD2087EC7EFB59A561", "6) Add SenseiNode node operator with reward address 0xE556Da28015c04F35A52B3111B9F4120E908056e", "7) Set 3.1531 stETH as the allowance of Burner over the Agent's tokens", - "8) Grant REQUEST_BURN_MY_STETH_ROLE to Agent", + "8) Grant Burner.REQUEST_BURN_MY_STETH_ROLE to Agent", "9) Request to burn 3.1531 stETH for cover", ] diff --git a/utils/test/event_validators/common.py b/utils/test/event_validators/common.py index 7b29d409..112ac5ec 100644 --- a/utils/test/event_validators/common.py +++ b/utils/test/event_validators/common.py @@ -14,8 +14,6 @@ # tx_events: ['A', 'B', 'A', 'B'], reference_events: ['A', 'B'] => invalid // duplicated 'A', 'B' events chain # tx_events: ['A', 'C', 'B'], reference_events: ['A', 'B', 'C'] => invalid // wrong order def validate_events_chain(tx_events: List[str], reference_events: List[str]): - print(tx_events) - print(reference_events) for ev in tx_events: idx = next((reference_events.index(e) for e in reference_events if e == ev), len(reference_events)) assert idx != len(reference_events), f"{ev} not found in the remaining {reference_events} events chain"