Restore loop inference fallback#269
Conversation
|
What tests are failing without this? And: couldn't you just use the meshmode array context in those examples? |
The pattern of failures looks like: (full log) |
|
Have you tried something along these lines? diff --git a/test/test_bc.py b/test/test_bc.py
index f34f1776..5fd76890 100644
--- a/test/test_bc.py
+++ b/test/test_bc.py
@@ -52,9 +52,10 @@ from mirgecom.gas_model import (
import grudge.op as op
from mirgecom.simutil import get_box_mesh
-from meshmode.array_context import ( # noqa
- pytest_generate_tests_for_pyopencl_array_context
- as pytest_generate_tests)
+from meshmode.array_context import PytestPyOpenCLArrayContextFactory
+from arraycontext import pytest_generate_tests_for_array_contexts
+pytest_generate_tests = pytest_generate_tests_for_array_contexts(
+ [PytestPyOpenCLArrayContextFactory])
logger = logging.getLogger(__name__)As recommended by the |
Yes, this causes the same error. |
|
🤷 I just tried the patch above in a fresh emirge install, and it fixes the issue for me. Note that this needs to be done in every test file. |
Weird, this patch does seem to work in CI without this PR. Maybe something wrong in my environment. |
|
Closing, as it seems to be unnecessary. |
This reverts a part of #264 (in particular, parts of 9af49ba).
With this PR (and
something like illinois-ceesd/mirgecom#898illinois-ceesd/mirgecom#1048), mirgecom is able to run from the main branches of the software stack.Side note: We carry a similar patch as this PR also in our production branch: illinois-ceesd@bfd22a5
Related: inducer/meshmode#333 (?)
Please squash