Skip to content

Commit a4cb569

Browse files
Fixes for aioca 1.8.* and general CI cleanup issue
aioca 1.8 changed its atexit handling to work better with other EPICS libraries. Note that cothread isn't quite so nice so there still needs to be some accomodation for that. Sometimes the CI would be unable to delete its temporary dir used for test results, thus failing the build.
1 parent 35bb833 commit a4cb569

File tree

2 files changed

+23
-3
lines changed

2 files changed

+23
-3
lines changed

.github/workflows/code.yml

Lines changed: 5 additions & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -95,7 +95,11 @@ jobs:
9595
env:
9696
CIBW_BUILD: ${{ matrix.python }}*64
9797
CIBW_TEST_EXTRAS: dev
98-
CIBW_TEST_COMMAND: pytest {project}/tests --cov-report xml:${{ matrix.cov_file }} --junit-xml=${{ matrix.results_file }}
98+
# Added a sleep command afterwards to let all cleanup finish; sometimes
99+
# cibuildwheel reports it cannot clean up the temp dir used for testing,
100+
# and fails the build. Sleeping seems to give pytest enough time to
101+
# fully clean up.
102+
CIBW_TEST_COMMAND: pytest {project}/tests --cov-report xml:${{ matrix.cov_file }} --junit-xml=${{ matrix.results_file }} && sleep 2
99103
# Run with faulthandler and -s in the hope we get a stack trace on seg fault on windows...
100104
CIBW_TEST_COMMAND_WINDOWS: python -X faulthandler -m pytest -s {project}/tests --cov-report xml:${{ matrix.cov_file }} --junit-xml=${{ matrix.results_file }}
101105
# Disable auditwheel as it isn't compatible with setuptools_dso approach

tests/conftest.py

Lines changed: 18 additions & 2 deletions
Original file line numberDiff line numberDiff line change
@@ -6,9 +6,20 @@
66
import string
77
import subprocess
88
import sys
9+
from packaging.version import Version
910
from typing import Any
1011
import pytest
1112

13+
# It is necessary to ensure we always import cothread.catools before aioca as
14+
# doing this import will cause an EPICS context to be created, and it will also
15+
# register an atexit function to delete it.
16+
# If we were to import aioca first it would create an EPICS context, note it has
17+
# created it, then try to delete it in its own atexit function which will
18+
# collide with cothread's attempts to do the same (despite the fact in this
19+
# configuration cothread did not create the context)
20+
if os.name != "nt":
21+
import cothread.catools
22+
1223
from softioc import builder
1324
from softioc.builder import ClearRecords, SetDeviceName, GetRecordNames
1425

@@ -73,12 +84,17 @@ def cothread_ioc():
7384

7485

7586
def aioca_cleanup():
76-
from aioca import purge_channel_caches, _catools
87+
from aioca import purge_channel_caches, _catools, __version__
7788
# Unregister the aioca atexit handler as it conflicts with the one installed
7889
# by cothread. If we don't do this we get a seg fault. This is not a problem
7990
# in production as we won't mix aioca and cothread, but we do mix them in
8091
# the tests so need to do this.
81-
atexit.unregister(_catools._Context._destroy_context)
92+
# In aioca 1.8 the atexit function was changed to no longer cause this
93+
# issue, when coupled with ensuring cothread is always imported first.
94+
if Version(__version__) < Version("1.8"):
95+
unregister_func = _catools._Context._destroy_context
96+
atexit.unregister(unregister_func)
97+
8298
# purge the channels before the event loop goes
8399
purge_channel_caches()
84100

0 commit comments

Comments
 (0)