Graham Dumpleton
grahamd at dscpl.com.au
Tue Jan 18 18:42:44 EST 2005
Nicolas Lehuen wrote .. > Judging from > > http://www.apache.org/~jim/committers.html > > we are 6 with commiters right on httpd-python. Maybe we could do > something like a mod_python bug day soon and commit some of the > patches that were published on this mailing list ? I suggest all 6 > commiters register themselves on the JIRA bug repository > (http://issues.apache.org/jira/browse/MODPYTHON?report=select), so > that we can assign bugs to each others. Just for the record, here are my set of six bugs which I have posted about over the last few months. I don't follow what is in CVS so have no idea which ones anything has been done about. One is mentioned in the bug repository, but the listed fix for it will only work with more recent versions of Python, thus precluding people from using mod_python with Python 2.1. My later fix would have worked with Python 2.1 as well. The bugs as I have reported them are below. Anyone want to comment on whether they have been addressed. 1. Both mod_python 2.7.10 and 3.1.3 will not work with the standard version of Python which is shipped with MAC OS X. This is because the Python library when loaded thinks it is already initialised when it isn't. I supplied a fix for this in the prior posts. http://www.modpython.org/pipermail/mod_python/2004-August/015994.html http://www.modpython.org/pipermail/mod_python/2004-September/016507.html 2. The apache.import_module() module is not sufficiently thread lock protected. This means that it is possible for multiple threads to load distinct copies of a module at the same time. This one is in the bug repository as: http://issues.apache.org/jira/browse/MODPYTHON-2 The fix there will only work with Python 2.2 or later. My later post suggested alternate fix. Make sure followup is noted though as should use RLock and not Lock. http://www.modpython.org/pipermail/mod_python/2004-October/016644.html http://www.modpython.org/pipermail/mod_python/2004-October/016645.html 3. If using a multithreaded MPM, if on startup of process there is large number of accesses, it is possible that multiple interpreters can be created for the same named interpreter. Fix was supplied. http://www.modpython.org/pipermail/mod_python/2004-October/016634.html 4. The value for config_dir is not setup correctly by mod_python on Win32 platforms as it makes wrong assumption that Apache passes through paths in DOS style when it actually uses POSIX style. Fix was supplied. http://www.modpython.org/pipermail/mod_python/2004-November/016784.html 5. The interpreter name wasn't being set correctly when the directive PythonInterpPerDirectory was being used. Described problem but did not supply a fix. http://www.modpython.org/pipermail/mod_python/2004-November/016788.html 6. Not really a bug but an issue of sorts. That is, that once a option is set using PythonOption, there is no way to unset it, nor is there a way of setting an option to an empty string. http://www.modpython.org/pipermail/mod_python/2004-October/016562.html I think that was all the things I found. I supply the fixes for a couple of these along with Vampire since I think they can be reasonably serious if you are using multithreading. http://www.dscpl.com.au/projects/vampire/PATCHES The others I have workarounds in my code so as to avoid the problems. Anyway, though I better review these in cases they have fallen through the cracks. -- Graham Dumpleton (grahamd at dscpl.com.au)
|