[Zope3-dev] ZEO tests error

[Tim Peters]

[Roger ineichen]
> C:\projektCompiler\writeaccess>python test.py --all

OK, you're on Windows.  Which flavor of Windows?  Which version of
Python?  Cygwin or native Windows?  If native Windows, how did you get
Zope3's C code compiled?

Here on native Windows XP Pro, compiling the C code myself with MSVC 6
+ all service packs, and using the native Windows Python 2.3.4:

C:\Code\Zope3>test.py -vv --all
Configuration file found.
Running UNIT tests at all levels
Running UNIT tests from C:\Code\Zope3
testBadUpdateTupleSize (BTrees.tests.testBTrees.IIBucketTest) ... ok
testClear (BTrees.tests.testBTrees.IIBucketTest) ... ok
...
toxml (zope.xmlpickle.xmlpickle) ... ok

======================================================================
FAIL: test_gadflyRoot
(zope.app.rdb.tests.test_gadflyrootdirective.DirectiveTest)
----------------------------------------------------------------------
Traceback (most recent call last):
  File "C:\Code\Zope3\src\zope\app\rdb\tests\test_gadflyrootdirective.py",
line 31, in test_gadflyRoot
    self.assert_('src/zope/app/rdb/tests/test/dir' in getGadflyRoot())
  File "C:\python23\lib\unittest.py", line 278, in failUnless
    if not expr: raise self.failureException, msg
AssertionError

----------------------------------------------------------------------
Ran 4945 tests in 1231.875s

FAILED (failures=1)

Running FUNCTIONAL tests at all levels
...
Ran 126 tests in 24.297s

OK

C:\Code\Zope3>


No ZEO failures, but a new gadfly failure that's almost obviously due
to Unix-specific code in the test.

You saw a bunch of ZEO failures.  I'm afraid that may not be fixable: 
the ZEO tests do things Windows hates doing, and isn't good at, in
particular creating lots of new processes.  That's slow on Windows,
unreliable on pre-NT versions of Windows, and the ZEO tests are
already choked with hacks trying to out-guess the most common ways
Windows fails to respond in a timely matter.

So an unfortunate fact of life is that several of the ZEO tests just
will fail from time to time, and especially on Windows.  When I run
ZEO tests on Windows, I try to keep the machine otherwise pretty quiet
for the duration.  The ZEO tests are very demanding, because they try
to cram inter-process interactions that usually stretch out over a
considerable time (in real life) into a fraction of a second, and
while they're at least 20x better at that than they were 2 years ago,
they're still too delicate.  Most of the delicate ones run only when
you (as you and I both did above) pass --all to test.py.