[Pythonmac-SIG] PPC vs 68K

Just van Rossum just@letterror.com
Thu, 21 May 1998 02:18:17 +0200


At 4:53 PM -0700 5/20/98, Rob Managan wrote:
>Just to stir the soup a little!

Yessir!

>I am finishing up the latest version of the Numeric extension (also known
>as NumPy). The PPC version works fine but the CFM68K version is
>problematic. Seems to work fine from the console. The test file dies on the
>first line; something like:
>
>array([1,2,3])
>
>The same line works fine on the console.

Strange. Something like this crashes from the console, too b.t.w.:

>>> a = Numeric.array([0.0, 0.1, 0.2, 0.3])
>>> a
kaboom!
(the same with integer works ok.)

Jack once built a static 68k interpreter with NumPy, which runs fine.
So it is pretty hairy cfm68k stuff, I presume :-(.

>So my question is: How much interest is there in the CFM68K version??? I
>know Jack has said that the CFM68K version has other troubles. How many
>would use a 68K version?

I would not personally use it much, but I'm working on an app which will
use NumPy extensively, and would like people with 68k machines to be
able to use it... But: it seems quite a nasty bug, so if it's going to take
several people weeks to fix it, it might not be worth it. I'd be happy
to test, though, the good ol' Quadra is not far away... (Or if you could
help me to set up for debugging, I could step with the MW debugger,
I don't really know how to do that for shared libs :-( )

>How's that for stirring the soup up!

Very good indeed. Thanks!

On a related note: I recently tried to compile calldll for cfm68k, and
it partially works. Trouble here is that I have no clue about calling
conventions, let alone 68k specifics in that area. Some calls work,
some crash. Perhaps you or others know about this subject so we could
collectively try to get it to work?
Same question remains, though: how many people need it?

Just