Author: Armin Rigo <[email protected]>
Branch:
Changeset: r46531:20bbf61dca0f
Date: 2011-08-16 12:05 +0200
http://bitbucket.org/pypy/pypy/changeset/20bbf61dca0f/
Log: Found (I think) the source of the occasional issue in this threading
test.
diff --git a/pypy/module/thread/test/test_ll_thread.py
b/pypy/module/thread/test/test_ll_thread.py
--- a/pypy/module/thread/test/test_ll_thread.py
+++ b/pypy/module/thread/test/test_ll_thread.py
@@ -52,6 +52,13 @@
assert get_ident() == state.z.ident
state.seen_value = state.z.value
state.z = None
+ # I think that we would need here a memory barrier in order
+ # to make the test pass reliably. The issue is that the
+ # main thread may see 'state.done = 1' before seeing the
+ # effect of the other assignments done above. For now let's
+ # emulate the write barrier by doing a system call and
+ # waiting a bit...
+ time.sleep(0.012)
state.done = 1
def g(i):
_______________________________________________
pypy-commit mailing list
[email protected]
http://mail.python.org/mailman/listinfo/pypy-commit
