Language Shootout

Bengt Richter bokr at accessone.com
Thu Jul 12 09:58:39 CEST 2001


On Wed, 11 Jul 2001 23:52:59 -0400, "Tim Peters" <tim.one at home.com> wrote:
[...]
>
>The rub is that while this kind of thing is fun and easy to program in
>Python, it's a total pain in the ass to program in C, and indeed leads to
>code bloat.  Since Python longs use base 2**15 internally, converting to
>power-of-2 bases is both simple and fast in C, so we settled for that.
>
That's interesting. Are you in effect saying there can't be compiled
python bytecode in python, even if there could be both speed and
space advantage for a given function?

So long as you weren't trying to lift yourself by not-yet-existent
boot straps, why not? That sounds a little funny, but ITYKWIM ;-)

BTW, looking at the way you can call the interpreter from C for embedding,
I would think you could write a translator for a useful subset of python
that would output a mix of plain C and Py_whatever calls to create C source
that could be compiled and linked into CPython. This would bypass the PITA
part and leave the fun part, ISTM ;-) Could you alter the byte code compiler
to emit this?

Done that way, strL wouldn't amount to that much C (or would it?), and you
could have 100% C code that way (i.e., not even byte codes as C constants
to feed the python interpreter).




More information about the Python-list mailing list