[lxml-dev] Python 2.4.1 and threading
Hi there, I think I just discovered that lxml 1.1.2 doesn't work clearly with Python 2.4.1 either. It appears to work with Python 2.4.3 and 4, but when I compile it for Python 2.4.1, it segfaults when you get an error during parsing. When I take lxml trunk and compile it without threading support, it does all work with Python 2.4.1. This leads me to suspect threading support is again the issue. Stefan, perhaps you can turn off threading support not only in Python 2.3 but also in (at least) Python 2.4.1. Should we be going for a new release? Perhaps the world is ready for a lxml 1.2. We haven't done a lot of changes except for the setup.py stuff (that is, I can't see any mentioned in the CHANGES.txt..), but I think those changes might warrant a new version number. Regards, Martijn
Martijn Faassen wrote:
I think I just discovered that lxml 1.1.2 doesn't work clearly with Python 2.4.1 either. It appears to work with Python 2.4.3 and 4, but when I compile it for Python 2.4.1, it segfaults when you get an error during parsing.
When I take lxml trunk and compile it without threading support, it does all work with Python 2.4.1. This leads me to suspect threading support is again the issue.
Hmm, ok, I saw a couple of differences in the PyGILState_* API functions between 2.3.6 and 2.4.4, so maybe they make a difference for us. It's still possible that we're doing something wrong in lxml, but having it work with newer versions lets me suspect that it's a race condition that was solved in later Python versions).
Stefan, perhaps you can turn off threading support not only in Python 2.3 but also in (at least) Python 2.4.1.
Ok, no problem. It's just a plain version number comparison.
Should we be going for a new release? Perhaps the world is ready for a lxml 1.2. We haven't done a lot of changes except for the setup.py stuff (that is, I can't see any mentioned in the CHANGES.txt..), but I think those changes might warrant a new version number.
There are a couple of things that I expected to make it into 1.2, especially the xmlReconciliateNs() replacement. But that one definitely needs more testing before a release. There's also the integration of ElementInclude.py that should be easier to integrate. I don't think it's a good time to release "right now" or even next week, but I agree that having a simpler-to-hack build process can become an opener and should get a second-level version number to show that there may be things to do to get it back working. Stefan
participants (2)
-
Martijn Faassen
-
Stefan Behnel