On 3 Jan 2014 08:20, "Vinay Sajip" <vinay_sajip@yahoo.co.uk> wrote:
I guess my question then is "why do you care?" If 2to3 is run at install time then it's a cost, but it's one-time and if you really care you can always create your own wheel of the translated code or something.
Because I've written a packaging tool that doesn't use setup.py but
instead uses a fully declarative format for building and installing packages, and I don't want that tool to run what can be a time-consuming step every time it installs certain packages, when that step isn't necessary. It's not a common use case, I grant you :-) Also getting into the metabuild system design. The idea of classifiers potentially having programmatic effects is one that hadn't previously occurred to me - I've always just viewed them as advisories for human readers. However, it would only apply when installing from an sdist (not a wheel), so I don't believe the extra complexity is worth it at this point - for the moment, setup.py remains the only "defined" build system. However, we should definitely make sure that it is covered in the development of the metabuild system. Cheers, Nick.
Regards,
Vinay Sajip