ANN: 0.0.0

Aahz Maruch aahz at
Sun May 20 04:40:23 CEST 2001

In article <3B071E0F.578EA073 at>,
Steve Williams  <stevewilliams at> wrote:
>One seldom inverts a matrix in a payroll program.
>Decimal arithmetic is not about speed, it's about doing the right thing.

Those statements are true as far as they go, but as you've seen time and
time again if you pay any attention to this newsgroup, floating point
binary numbers contain gotchas for *everyone* (go on, type 1.1 into the
interpreter for Python 2.x).  And many people who need accurate decimal
arithmetic need a reasonable amount of speed, including the NumPy people.

I should note that I'm coming at this bass-ackward.  I've gained a bit
of a reputation as a thread expert, and one of the things that's
bothered me about Python is that almost the entire focus on sidestepping
the Global Interpreter Lock has been on the I/O side.  The eventual goal
of this project is to essentially provide a replacement for the Python
long that releases GIL when it's doing the heavy crunching -- which
should provide a performance boost on multi-CPU machines.

Now, I could try doing that to longobject.c directly, but I generally
find it easier to write code from scratch than to modify other people's
code, especially in languages that I don't know well (like C).  And the
weird thing is, I'm having *fun* doing this.  Normally I actually don't
like programming all that much.
                      --- Aahz  <*>  (Copyright 2001 by aahz at

Androgynous poly kinky vanilla queer het Pythonista
Hugs and backrubs -- I break Rule 6

Cole's Law: thinly sliced cabbage

More information about the Python-list mailing list