[pull] forks/amsterdam from ethereum:forks/amsterdam

.github/workflows/test.yaml

@@ -56,8 +56,28 @@ jobs:
           uvx --from actionlint-py actionlint
 
   py3:
+    name: py3 (${{ matrix.label }})
     runs-on: [self-hosted-ghr, size-xl-x64]
     needs: static
+    strategy:
+      fail-fast: false
+      matrix:
+        include:
+          - label: pre-cancun
+            from_fork: Frontier
+            until_fork: Shanghai
+          - label: cancun
+            from_fork: Cancun
+            until_fork: Cancun
+          - label: prague
+            from_fork: Prague
+            until_fork: Prague
+          - label: osaka
+            from_fork: Osaka
+            until_fork: Osaka
+          - label: amsterdam
+            from_fork: Amsterdam
+            until_fork: Amsterdam
     steps:
       - uses: actions/checkout@de0fac2e4500dabe0009e67214ff5f5447ce83dd # v6.0.2
         with:
@@ -66,8 +86,8 @@ jobs:
         with:
           python-version: "3.14"
       - uses: ./.github/actions/setup-env
-      - name: Run py3 tests
-        run: tox -e py3
+      - name: Run py3 tests (${{ matrix.label }})
+        run: tox -e py3 -- --from ${{ matrix.from_fork }} --until ${{ matrix.until_fork }}
         env:
           PYTEST_XDIST_AUTO_NUM_WORKERS: auto
       - name: Upload coverage reports to Codecov

packages/testing/pyproject.toml

@@ -20,7 +20,7 @@ classifiers = [
 ]
 dependencies = [
     "click>=8.1.0,<9",
-    "ethereum-hive>=0.1.0a1,<1.0.0",
+    "ethereum-hive>=0.1.0a5,<1.0.0",
     "ethereum-execution",
     "gitpython>=3.1.31,<4",
     "PyJWT>=2.3.0,<3",

packages/testing/src/execution_testing/cli/pytest_commands/plugins/consume/simulators/enginex/conftest.py

@@ -122,6 +122,26 @@ def test_suite_description() -> str:
     )
 
 
+@pytest.fixture(scope="function", autouse=True)
+def _per_test_reporting(
+    client: Client,
+    hive_test: HiveTest,
+) -> None:
+    """
+    Register a test for execution against a multi-test client.
+
+    Activate log segment capturing in the Hive backend for correct
+    client log reporting in the multi-test client case.
+
+    Parameter order matters: `client` listed before `hive_test`
+    ensures pytest sets up `client` first and tears it down last.
+    This guarantees `hive_test` teardown (`test.end()`) runs while
+    the hive node still exists, before `client` teardown calls
+    `mark_test_completed` / `client.stop()`.
+    """
+    hive_test.register_multi_test_client(client)
+
+
 @pytest.fixture(scope="function")
 def client(
     multi_test_hive_test: HiveTest,
@@ -179,6 +199,7 @@ def client(
         multi_test_client_manager.register_client(
             group_identifier, resolved_client
         )
+        resolved_client.multi_test = True
 
     try:
         yield resolved_client

packages/testing/src/execution_testing/specs/base.py

@@ -36,6 +36,9 @@
 from execution_testing.forks import Fork
 from execution_testing.forks.base_fork import BaseFork
 from execution_testing.test_types import Environment, Withdrawal
+from execution_testing.test_types.receipt_types import (
+    TransactionReceipt,
+)
 
 
 class HashMismatchExceptionError(Exception):
@@ -109,6 +112,7 @@ class BaseTest(BaseModel):
     gas_optimization_max_gas_limit: int | None = None
     expected_benchmark_gas_used: int | None = None
     skip_gas_used_validation: bool = False
+    expected_receipt_status: int | None = None
     is_tx_gas_heavy_test: bool = False
     is_exception_test: bool = False
 
@@ -170,7 +174,7 @@ def from_test(
     ) -> Self:
         """Create a test in a different format from a base test."""
         for k in BaseTest.model_fields.keys():
-            if k not in kwargs:
+            if k not in kwargs and k in base_test.model_fields_set:
                 kwargs[k] = getattr(base_test, k)
         return cls(**kwargs)
 
@@ -288,5 +292,32 @@ def validate_benchmark_gas(
             f"{gas_benchmark_value}"
         )
 
+    def validate_receipt_status(
+        self,
+        *,
+        receipts: List[TransactionReceipt],
+        block_number: int,
+    ) -> None:
+        """
+        Validate receipt status for every transaction in a block.
+
+        When expected_receipt_status is set, verify that all
+        receipts match. Catches silent OOG failures that roll
+        back state and invalidate benchmarks.
+        """
+        if "expected_receipt_status" not in self.model_fields_set:
+            return
+        for i, receipt in enumerate(receipts):
+            if receipt.status is not None and (
+                int(receipt.status) != self.expected_receipt_status
+            ):
+                raise Exception(
+                    f"Transaction {i} in block "
+                    f"{block_number} has receipt "
+                    f"status {int(receipt.status)}, "
+                    f"expected "
+                    f"{self.expected_receipt_status}."
+                )
+
 
 TestSpec = Callable[[Fork], Generator[BaseTest, None, None]]

packages/testing/src/execution_testing/specs/blockchain.py

@@ -858,8 +858,7 @@ def make_fixture(
         invalid_blocks = 0
         benchmark_gas_used: int | None = None
         benchmark_opcode_count: OpcodeCount | None = None
-        for i, block in enumerate(self.blocks):
-            is_last_block = i == len(self.blocks) - 1
+        for block in self.blocks:
             # This is the most common case, the RLP needs to be constructed
             # based on the transactions to be included in the block.
             # Set the environment according to the block to execute.
@@ -869,9 +868,16 @@ def make_fixture(
                 previous_env=env,
                 previous_alloc=alloc,
             )
+            block_number = int(built_block.header.number)
+            is_last_block = block is self.blocks[-1]
             if is_last_block and self.operation_mode == OpMode.BENCHMARKING:
                 benchmark_gas_used = int(built_block.result.gas_used)
                 benchmark_opcode_count = built_block.result.opcode_count
+            if built_block.result.receipts:
+                self.validate_receipt_status(
+                    receipts=built_block.result.receipts,
+                    block_number=block_number,
+                )
             include_receipts = (
                 block.include_receipts_in_output
                 if block.include_receipts_in_output is not None
@@ -953,17 +959,23 @@ def make_hive_fixture(
         invalid_blocks = 0
         benchmark_gas_used: int | None = None
         benchmark_opcode_count: OpcodeCount | None = None
-        for i, block in enumerate(self.blocks):
-            is_last_block = i == len(self.blocks) - 1
+        for block in self.blocks:
             built_block = self.generate_block_data(
                 t8n=t8n,
                 block=block,
                 previous_env=env,
                 previous_alloc=alloc,
             )
+            block_number = int(built_block.header.number)
+            is_last_block = block is self.blocks[-1]
             if is_last_block and self.operation_mode == OpMode.BENCHMARKING:
                 benchmark_gas_used = int(built_block.result.gas_used)
                 benchmark_opcode_count = built_block.result.opcode_count
+            if built_block.result.receipts:
+                self.validate_receipt_status(
+                    receipts=built_block.result.receipts,
+                    block_number=block_number,
+                )
             fixture_payloads.append(
                 built_block.get_fixture_engine_new_payload()
             )

tox.ini

@@ -125,7 +125,6 @@ commands =
         --clean \
         --until Amsterdam \
         --durations=50 \
-        --ignore=tests/ported_static \
         {posargs} \
         tests