Hello,

Let's open an official discussion about this.


First of all, what happens on Gentoo. The PyPy's build process creates
an executable named 'pypy-c'. We install it with this name, and symlink
as /usr/bin/pypy-cX.Y for each PyPy version. We don't do any 'common'
wrapper for PyPy.

Upstream's packaging scripts, however, rename the 'pypy-c' executable
to 'pypy'. All other distros I have checked (Arch, Debian, Fedora,
Ubuntu) install a single /usr/bin/pypy for them. They don't support
multiple versions though.

At the point, it seems reasonable to drop our '-c' addition and just
use 'pypyX.Y' (+ 'pypy') instead. We could do this starting with 2.1
but...


Since v2.1, PyPy has been 'split' into PyPy and PyPy3, the former being
Python2 variant and the latter Python3. Both share the same versions
(that is, there's PyPy 2.1 and PyPy3 2.1).

From what floppym checked, the build process builds plain 'pypy-c'
as well. Alike regular PyPy, upstream's scripts rename it to 'pypy'.

Since PyPy3 is still in beta, only Arch Linux has packages for it. It
installs a single /usr/bin/pypy3 (what a surprise).


This raises the question: how should we name our PyPy executables?
While I really like the 'pypyX.Y' idea, I don't really want to see
'pypy32.1' :). We could go for 'pypy-X.Y' and 'pypy3-X.Y' but that would
be inconsistent with CPython (pythonX.Y) and Jython (jythonX.Y0).
A somehow ugly alternative would be to use 'pypy-cX.Y' and 'pypy3-cX.Y'
but that is quite a custom invention.

Your thoughts?

-- 
Best regards,
Michał Górny