This is an automated email from the ASF dual-hosted git repository.
tqchen pushed a commit to branch main
in repository https://gitbox.apache.org/repos/asf/tvm-ffi.git
The following commit(s) were added to refs/heads/main by this push:
new 2562bf40 [TEST] Skip the flaky shared DAG hash scaling test (#857)
2562bf40 is described below
commit 2562bf40e54f13a2b6a233633c0bbe06e72c0089
Author: Yaxing Cai <[email protected]>
AuthorDate: Mon Oct 19 19:59:23 2026 +0800
[TEST] Skip the flaky shared DAG hash scaling test (#857)
## Why
`test_shared_dag_hash_scaling_not_exponential` compares the mean time of
hashing depth-18 and depth-19 shared DAGs against a 10x ratio. Each hash
takes a few microseconds, so a single scheduling stall pushes the ratio
past the threshold. It failed the aarch64 wheel job of the Publish wheel
runs for v0.1.13, v0.1.14-post1 (twice) and v0.1.15 (`d18=0.000006s
d19=0.000155s`), and mainline-only CI on cp315. Every failure holds the
whole PyPI upload until the job is re-run. #612 and #702 already relaxed
its threshold.
## What changed
Skip the test with `@pytest.mark.skip`, giving the reason.
## Validation
- `tests/python/test_dataclass_hash.py`: `152 passed, 1 skipped`; the
skip reason is reported.
- `ruff check` and `ruff format --check` pass. `check version
consistency` currently fails on every PR because `main` still has patch
15 after the `v0.1.15` tag; #856 fixes that.
---
tests/python/test_dataclass_hash.py | 1 +
1 file changed, 1 insertion(+)
diff --git a/tests/python/test_dataclass_hash.py
b/tests/python/test_dataclass_hash.py
index a786bec9..f673ccd2 100644
--- a/tests/python/test_dataclass_hash.py
+++ b/tests/python/test_dataclass_hash.py
@@ -843,6 +843,7 @@ def _make_shared_binary_dag(depth: int) -> object:
return node
[email protected](reason="flaky: microsecond timing ratio is sensitive to CI
scheduling noise")
def test_shared_dag_hash_scaling_not_exponential() -> None:
"""Hashing shared DAGs should scale roughly linearly in depth."""
d18 = _make_shared_binary_dag(18)
