python-dev-request@python.org writes:

I'd like to get the necessary C extension in before 2.0 freeze, if possible. I've copied its author. Again, the motivation here is to make shelving transactional, with useful read-many/write-once guarantees. Thousands of CGI programmers would thank us for this.

When we last discussed this subject, there was general support for the functionality, but a couple of people went "bletch!" about SWIG-generated code (there was unhappiness about pointers being treated as strings).

Somebody said something about having SWIG patches to address this. Is this the only real issue with SWIG-generated code? If so, we can pursue two paths:

Well, as the guilty party on the SWIG front, I can say that the current development version of SWIG is using CObjects instead of strings (well, actually I lie---you have to compile the wrappers with -DSWIG_COBJECT_TYPES to turn that feature on). Just as a general aside on this topic, I did a number of experiments comparing the performance of using CObjects vs.the gross string-pointer hack about 6 months ago. Strangely enough, there was virtually no-difference in runtime performance and if recall correctly, the string hack might have even been just a little bit faster. Go figure :-). Overall, the main difference between SWIG1.3 and SWIG1.1 is in runtime performance of the wrappers as well as various changes to reduce the amount of wrapper code. However, 1.3 is also very much an alpha release right now---if you're going to use that, make sure you thoroughly test everything. On the subject of the Berkeley DB module, I would definitely like to see a module for this. If there is anything I can do to either modify the behavior of SWIG or to build an extension module by hand, let me know. Cheers, Dave