picklable-itertools 0.1.0

Because the standard library pickle module (nor the excellent dill package) can’t serialize all of the itertools iterators, at least on Python 2 (at least some appear to be serializable on Python 3).

Because there are lots of instances where these things in itertools would simplify code, but can’t be used because serializability must be maintained across both Python 2 and Python 3. The in-development framework Blocks is our first consumer. We’d like to be able to serialize the entire state of a long-running program for later resumption. We can’t do this with non-picklable objects.

This should be a drop-in replacement. Pretty self-explanatory. Test against the standard library itertools or builtin implementation to verify behaviour matches. Where Python 2 and Python 3 differ in their naming, (filterfalse vs ifilterfalse, zip_longest vs. izip_longest) we provide both. We also provide names that were only available in the Python 2 incarnation of itertools (ifilter, izip), also available under their built-in names in Python 3 (filter, zip), for convenience. As new objects are added to the Python 3 itertools module, we intend to add them (accumulate, for example, appears only in Python 3, and a picklable implementation is contained in this package.)

Handle built-in types gracefully if possible. List iterators, etc. are not picklable on Python 2.x, so we provide an alternative implementation. File iterators are handled transparently as well. dict iterators and set iterators are currently not supported. picklable_itertools.xrange can be used as a drop-in replacement for Python 2 xrange/Python 3 range, with the benefit that the iterators produced by it will be picklable on both Python 2 and 3.

Premature optimization is the root of all evil. These things are implemented in Python, so speed is obviously not our primary concern. Several of the more advanced iterators are constructed by chaining simpler iterators together, which is not the most efficient thing to do but simplifies the code a lot. If it turns out that speed (or a shallower object graph) is necessary or desirable, these can always be reimplemented. Pull requests to this effect are welcome.