As discussed on the mailing list, we should be able to generate the Python reference without building all of LLDB. To make that possible I create a dummy python package, which is then parsed by epydoc. The latter will complain that it couldn't import lldb, but that doesn't matter as far as generation of the docs is concerned.
Looks good to me. Thanks for looking into this.
On one hand, it would be nice to create avoid creating the fake package, and just run epydoc from the final destination, but on the other, the nice thing about this is that it will always run in the same environment and not depend on whether you happen to have _lldb.so built or not (plus, with all the complexity around frameworks and stuff, it might be hard to get __init__.py to be placed in the final destination without depending on any of the other build steps).