[Python-3000] Making 2to3 installable

[Guido van Rossum]

On Sun, Mar 16, 2008 at 4:22 PM, "Martin v. Löwis" <martin at v.loewis.de> wrote:
> >>  Wrt. invoking it from distutils: Why not?
>  >
>  > Because it's far from perfect. I'd expect the success rate of running
>  > 2to3 over a 2.x package to be close to zero. I think this is one case
>  > where it's better to let the developer run 2to3 and tweak the app
>  > until it actually works.
>
>  Ah. I still dream of people being able to use a single source for both
>  2.x and 3.x. They would maintain their source as 2.x, and convert to
>  3.0 at deployment time.
>
>  Of course, they have to change their source to make that actual work.
>  Most distutils projects probably won't work when all you do is run
>  2to3. However, if a project has prepared its source in a proper way
>  (for some definition of "proper"), it *should* be possible that all
>  remaining changes are done by 2to3.
>
>  People would try the process on their development machines, and change
>  the code until it actually runs under both versions.
>
>  I'll be using my sprinting time to find out whether that approach
>  can actually work.

That would be good. I agree with the ideal process, but I doubt that
it will realistically work exactly like that, so I expect that calling
2to3 upon install is just going to cost a lot of CPU time and then
produce a non-working install.

>  >>  Wrt. this covering all uses: Surely the ones that people
>  >>  would use the library for, no?
>  >
>  > I not understand. :-(
>
>  When people requested that 2to3 is a library, I think they
>  have exactly that use case in mind: programmatically convert
>  a source code base at deployment/build time. In those cases,
>  they can achieve the same thing with the command line tool.

I don't recall what the person who asked about this after my keynote
had in mind, but it sounds to me like converting code at install time
a minority use case -- especially while 2to3 is as slow as it is.

>  Of course, on Windows, it is probably still better to use
>  the library, since Python will find the library, but may
>  not find the command line tool.

Good point.

-- 
--Guido van Rossum (home page: http://www.python.org/~guido/)