tests: move provided.al2 to the deprecated lint list and stop rerunning a non-rerun-safe sync test

[roger-zhangg]

Which issue(s) does this change fix?

N/A

Why is this change necessary?

Two failures from last night's integration runs (32094973517, 32122795247).

For context, the pytest 9.1 unblock (#9189) landed cleanly — sync-code passes again, and other-and-e2e went from 135 errors → 0 (all 81 nested-stack logs tests now run). These two are the unrelated leftovers.

1. test_lint_supported_runtimes — fails in both runs, systematic

cfn-lint marks provided.al2 EOL as of 2026-07-31, so sam validate --lint now exits 1 while the test asserts 0:

W2531: Runtime 'provided.al2' was deprecated on '2026-07-31'.
Error: Linting failed. At least one linting rule was matched to the provided template.

A time bomb rather than a flake — it began failing after that date and fails every run until fixed.

2. TestSyncAdlCasesWithCodeParameter.test_sync_code_function_without_dependencies — sync-watch

Reruns cannot help this test and actively hide why it failed. It edits test data in place (update_file writes over code/before/python_function_no_deps/app.py) and the class-scoped fixture that deploys the stack does not re-run, so once attempt 1 has mutated the tree the retries run against inconsistent state.

That is exactly what the reported error was. The function came back:

Unable to import module 'app': No module named 'numpy'

so _get_lambda_response returned None (error payloads have no body) and the failure surfaced as TypeError: the JSON object must be str, bytes or bytearray, not NoneType — three wasted deploy cycles and none of the original cause. Pre-bump this job had 0 reruns and the test passed.

How does it address the issue?

provided.al2 moves out of supported_runtimes and into test_lint_deprecated_runtimes, which derives the expected code (W2531/E2531/E2533) from cfn-lint's LmbdRuntimeLifecycle.json by date — so it needs no further edits as the runtime moves through create-block and update-block. Same treatment as ruby3.2 in #8883 and nodejs20.x in #8947, except those were deleted rather than moved.

I checked the rest of the list: only provided.al2 is flagged today. java8.al2 is still clean and is the likely next one.

The sync test gets @pytest.mark.flaky(reruns=0), so it reports the real first-attempt failure immediately instead of a misleading artifact three deploys later. Transient invoke flakes are still covered — _get_lambda_response retries internally.

What side effects does this change have?

reruns=0 means a genuine pre-mutation flake in that one test now fails the build rather than being retried. That is the intended trade: the test cannot pass on a rerun anyway, so the retries only cost time and obscure the cause.

Worth flagging for a follow-up: 26 update_file calls write into code/before — test_sync_adl.py (6), test_sync_code.py (2), test_sync_watch.py (20) — so the same rerun-unsafety applies wherever one of those fails after mutating. Only the test that actually failed is marked here; making them rerun-safe (restoring the tree, or re-running the infra sync) is a larger change and belongs on its own.

Testing

• With SAM_CLI_DEV=1 (as CI sets), test_lint_supported_runtimes and all 7 test_lint_deprecated_runtimes cases pass — 8 passed. Note a local run without SAM_CLI_DEV tests the installed sam, whose older bundled cfn-lint does not flag provided.al2
• Verified reruns=0 is honoured: 1 attempt versus 4 for an unmarked test
• tests/integration/{sync,validate} collect 153 tests; black and ruff clean

Mandatory Checklist

PRs will only be reviewed after checklist is complete

• Review the generative AI contribution guidelines
• Add input/output type hints to new functions/methods
• Write design document if needed (Do I need to write a design document?)
• Write/update unit tests
• Write/update integration tests