[Pyrex] Questions about your approach

[Greg Ewing]

Paul Biggar wrote:
> Did pyrex originally compile standard python, or was it annotated with
> types from the start?

It had type declarations from the start. It was conceived
as a bridge between Python and C, with automatic conversion
between Python and C data wherever it was obvious what
needed to be done. For that, I needed information about
the C types.

> Has you run benchmarks on the compiled applications?

Not very much. I once timed the primes.pyx example as about
100 times faster than a Python version of the same thing.
I created a Pyrex version of my Plex scanner, and managed to
get it to scan about 30 times faster.

However, speed isn't really the main point of Pyrex -- the
point is to interface between Python and C, either external
or written in the C-subset of Pyrex. The C parts run about
as fast as you would expect if they were written in straight
C, which isn't very surprising. The Python parts don't need
to be particularly fast, as they're only intended for getting
the data in and out of a form that the C parts can use.

> Are there any papers describing python?

You mean Pyrex? Not currently.

> I wonder if you have any comments on your approach vs. that of the
> defunct python2c?

They're aimed at different problems, I think. Python2c seems
to have been an attempt to compile pure Python code. Pyrex
is for interfacing between Python and C, so it's competing
more with the likes of SWIG and Boost Python. It happens
to have much the same effect as python2c if you apply it to
pure Python code, but that's pretty much accidental.

-- 
Greg