[cdavid]

It may just be anecdotal at this point, but several highprofiles libraries have given up on 2to3 entirely and use the 2/3 common codebase approach: Jinja, django, ply. For a library, that's the approach that I find more practical.

I have a hard time seeing how one can use 2to3 on an everyday basis: it makes testing, distribution, etc... that much more painful because it is so slow (2to3 takes as much time as compiling all the C code from numpy, for example). It also removes one of the big plus of a language like python (modify code - see the result right away).

[mcdonc]

Amen, 2to3 in practice for anything but the smallest library is effectively unusable due to speed. Even if it weren't, I'm not much of a fan of maintaining generated code. That's an imposition I'm not really willing to put up with.

[zzzeek]

Using a Jenkins server with continuous integration makes Py3K testing pretty painless.

I really don't want projects to go back to maintaining two source trees, that was a disaster.