Re: [Chicken-users] New egg: pyffi

[John Cowan]

Ivan Raikov scripsit:

> Pyffi uses the Python C interface to convert Python values to Scheme
> values and vice versa, and to invoke Python procedures and methods,
> and automatically convert their arguments and return values to and
> from Python and Scheme representations. It supports Python numeric
> values, strings, lists, tuples, and dictionaries; all other types are
> represented in Chicken as pointers. I have not made any attempts to
> optimize the value conversion code, and I have only tested pyffi with
> very simple code and small data structures.

This is great stuff.  Two points about Unicode (inevitably):

1) UTF-32 representations are definitely unsigned, so a u32vector is
better than an s32vector.  It doesn't matter in practice, since the
largest possible UTF-32 value is #x10FFFF.

2) I would argue that for Chicken, 32vectors are a bad choice; it
makes it annoying within Chicken to treate Unicode strings as strings.
There is a good package for treating Chicken strings as UTF-8, and IMHO
both Python strs and Python unicodes should be represented on the Chicken
side as strings, converting the Python unicode to UTF-8.

This of course leads to the question of when a Chicken string should be
passed to Python as a str, and when it should be passed as a unicode.
A simple approach such as using a pair like '(UTF-8 . "foobar") to pass
a unicode vs. simply "foobar" to pass a str would not be optimal, but
would suffice.  Looking at the content is not enough, since some Python
code requires unicodes even if the content is all-ASCII.

It's a question whether this representation should also be returned to
Chicken, or not to bother.

-- 
The experiences of the past show                John Cowan
that there has always been a discrepancy        address@hidden
between plans and performance.                  http://www.ccil.org/~cowan
        --Emperor Hirohito, August 1945