Test suite fails on Python 2.7

[Remy Blank]

The test suite has been ​reported by the Gentoo Linux project to fail on Python 2.7:

Traceback (most recent call last):
  File "trac/test.py", line 441, in <module>
    unittest.main(defaultTest='suite')
  File "/usr/lib64/python2.7/unittest/main.py", line 95, in __init__
    self.runTests()
  File "/usr/lib64/python2.7/unittest/main.py", line 229, in runTests
    self.result = testRunner.run(self.test)
  File "/usr/lib64/python2.7/unittest/runner.py", line 151, in run
    test(result)
  File "/usr/lib64/python2.7/unittest/suite.py", line 65, in __call__
    return self.run(*args, **kwds)
  File "/usr/lib64/python2.7/unittest/suite.py", line 85, in run
    self._wrapped_run(result)
  File "/usr/lib64/python2.7/unittest/suite.py", line 115, in _wrapped_run
    test._wrapped_run(result, debug)
  File "/usr/lib64/python2.7/unittest/suite.py", line 115, in _wrapped_run
    test._wrapped_run(result, debug)
  File "/usr/lib64/python2.7/unittest/suite.py", line 115, in _wrapped_run
    test._wrapped_run(result, debug)
  File "/usr/lib64/python2.7/unittest/suite.py", line 115, in _wrapped_run
    test._wrapped_run(result, debug)
TypeError: _wrapped_run() takes exactly 2 arguments (3 given)

[Remy Blank]

The following patch fixes the issue for me:

trac/test.py

@@ -139 +139 @@
         self.tearDown()
         return result
 
-    def _wrapped_run(self, result):
+    def _wrapped_run(self, *args, **kwargs):
         "Python 2.7 / unittest2 compatibility - there must be a better way..."
         self.setUp()
         if hasattr(self, 'fixture'):
             for test in self._tests:
                 if hasattr(test, 'setFixture'):
                     test.setFixture(self.fixture)
-        unittest.TestSuite._wrapped_run(self, result)
+        unittest.TestSuite._wrapped_run(self, *args, **kwargs)
         self.tearDown()
-        return result
 
 class TestCaseSetup(unittest.TestCase):
     def setFixture(self, fixture):

[Christian Boos]

Fix confirmed on my side, test suite pass (only in-memory SQLite tested so far, "bigtest" in progress). I suppose that correspond to a last minute change on their side between rc2 and final (I had rc2 installed until this week-end, and I'm pretty sure I tested 0.12 with that one).

So this is just a unittest induced glitch, no need to modify anything in the release notes, Trac 0.12.1, as did 0.12 before, works perfectly well with Python 2.7.

[Christian Boos]

Scratch that… I specified python=27 but 2.6.5 was used due to a misconfiguration.

[Remy Blank]

Thanks for testing. I noticed another (unrelated) issue with Python 2.7 on Windows 7. Every second run of the test suite fails on creation of the testenv folder. After the failure, the folder disappears, so the next run passes.

Looking at the code, the corresponding line​ looks very suspect, as os.mkdir() isn't supposed to return anything. I haven't had time to investigate further.

[Christian Boos]

Ok, there's apparently a different problem when running the tests using the following command line:

$ python ./trac/test.py --skip-functional-tests
Traceback (most recent call last):
  File "./trac/test.py", line 427, in <module>
    unittest.main(defaultTest='suite')
  File "C:\Dev\Python27\lib\unittest\main.py", line 94, in __init__
    self.parseArgs(argv)
  File "C:\Dev\Python27\lib\unittest\main.py", line 149, in parseArgs
    self.createTests()
  File "C:\Dev\Python27\lib\unittest\main.py", line 158, in createTests
    self.module)
  File "C:\Dev\Python27\lib\unittest\loader.py", line 128, in loadTestsFromNames
    suites = [self.loadTestsFromName(name, module) for name in names]
  File "C:\Dev\Python27\lib\unittest\loader.py", line 113, in loadTestsFromName
    test = obj()
  File "./trac/test.py", line 392, in suite
    import trac.tests
ImportError: No module named tests

No idea why, yet.

But it works fine using python -m trac.test or even python setup.py test. This way, all unit tests pass, and the functional tests need the patch from comment:1.

Tested on Windows 7 (x64) with Python 2.7 (x86):

• OK for sqlite::memory: with the builtin sqlite
• OK for mysql: with ​MySQLdb 1.2.3
• OK for postgresql: with ​psycopg2-2.2.2 (well, I actually have 11 failures, but they are all due to the switch to 9.0.1, no real failure; fixes for the failing tests will come separately)
• FAIL for sqlite:test27.db (builtin sqlite)

  Initenv for 'C:\Workspace\src\trac\repos\trunk\testenv\trac' failed.
  Failed to create environment.
  [Error 183] Cannot create a file when that file already exists: u'C:\\Workspace\\src\\trac\\repos\\trunk\\testenv\\trac'
  

  Perhaps this is related to the issue you mentioned in comment:4? But it happens systematically for me (and 2.6.5 works fine for this new backend)

I didn't test the following:

SKIP: versioncontrol/tests/functional.py (no svn bindings)
SKIP: versioncontrol/tests/svn_fs.py (no svn bindings)
SKIP: tracopt/mimeview/tests/php.py (php cli binary, 'php', not found)

Python 2.7 bindings for Subversion, anyone?

[Remy Blank]

Patch from comment:1 applied in [10219].

[Remy Blank]

Here's the traceback for the issue I mentioned in comment:4:

Traceback (most recent call last):
  File "trac\test.py", line 441, in <module>
    unittest.main(defaultTest='suite')
  File "C:\Python27\Lib\unittest\main.py", line 95, in __init__
    self.runTests()
  File "C:\Python27\Lib\unittest\main.py", line 229, in runTests
    self.result = testRunner.run(self.test)
  File "C:\Python27\Lib\unittest\runner.py", line 151, in run
    test(result)
  File "C:\Python27\Lib\unittest\suite.py", line 65, in __call__
    return self.run(*args, **kwds)
  File "C:\Python27\Lib\unittest\suite.py", line 85, in run
    self._wrapped_run(result)
  File "C:\Python27\Lib\unittest\suite.py", line 115, in _wrapped_run
    test._wrapped_run(result, debug)
  File "C:\Python27\Lib\unittest\suite.py", line 115, in _wrapped_run
    test._wrapped_run(result, debug)
  File "Z:\src\trac\0.12-stable-2.7\trac\test.py", line 144, in _wrapped_run
    self.setUp()
  File "Z:\src\trac\0.12-stable-2.7\trac\tests\functional\__init__.py", line 121, in setUp
    self._testenv = self.env_class(dirname, port, baseurl)
  File "Z:\src\trac\0.12-stable-2.7\trac\tests\functional\testenv.py", line 58, in __init__
    self.create()
  File "Z:\src\trac\0.12-stable-2.7\trac\tests\functional\testenv.py", line 114, in create
    if os.mkdir(self.dirname):
WindowsError: [Error 5] Access is denied: 'Z:\\src\\trac\\0.12-stable-2.7\\testenv'

It seems this has nothing to do with Python 2.7, and everything with Windows (or VirtualBox). The "access denied" error comes from the testenv directory not being removed completely when the os.rmdir() call returns (possibly due to Windows Explorer), and the subsequent os.mkdir() being denied creation.

Start the following program on Windows 7 from the command-line, with no Windows Explorer windows open:

import os

path = 'testenv'

while True:
    try:
        os.rmdir(path)
    except OSError:
        pass
    os.mkdir(path)

Nothing happens, the directory is created and destroyed repeatedly. Now, open a Windows Explorer window and navigate to the directory where you launched the script, and expand that folder in the tree view on the left. Here (Windows 7 VM on VirtualBox), the script traps within a few seconds. The symptom is the same with both Python 2.6 and 2.7.

Adding even a very small pause (1 ms) before the os.mkdir() fixes the issue. So I suggest the following patch:

trac/tests/functional/testenv.py

@@ -55 +55 @@
         self.pid = None
         self.init()
         self.destroy()
+        time.sleep(0.1) # Avoid race condition on Windows
         self.create()
         locale.setlocale(locale.LC_ALL, '')
 

[Christian Boos]

• OK for postgresql: with ​psycopg2-2.2.2 (well, I actually have 11 failures, but they are all due to the switch to 9.0.1, no real failure; fixes for the failing tests will come separately)

In r10221 and r10223.

[Christian Boos]

Replying to rblank:

Here's the traceback for the issue I mentioned in comment:4:

I didn't observe that error, but:

The "access denied" error comes from the testenv directory not being removed completely when the os.rmdir() call returns (possibly due to Windows Explorer), and the subsequent os.mkdir() being denied creation.

I can see that as well with plain Windows 7, with the test script.

Adding even a very small pause (1 ms) before the os.mkdir() fixes the issue.

Should work fine, please commit.

Unfortunately, the error I mentioned in comment:5 for the sqlite: backend is still there, so it must be a different issue.

[Remy Blank]

Patch from comment:7 applied in [10225].

[Christian Boos]

Replying to cboos:

Unfortunately, the error I mentioned in comment:5 for the sqlite: backend is still there, so it must be a different issue.

Problem solved, I had an easy_install of 0.12.1 interfering.

Onto the make test issue… argh, was the same problem ;-)
All green on my side, now.

[Remy Blank]

Replying to cboos:

• FAIL for sqlite:test27.db (builtin sqlite)

  Initenv for 'C:\Workspace\src\trac\repos\trunk\testenv\trac' failed.
  Failed to create environment.
  [Error 183] Cannot create a file when that file already exists: u'C:\\Workspace\\src\\trac\\repos\\trunk\\testenv\\trac'
  

That one was due to the SQLite connector trying to os.makedirs() an existing directory. Running with sqlite:db\test27.db works great. I have fixed the issue in [10231], and removed the suspect code from comment:4 in [10232].