Re: [Sbcl-devel] 0.6.4 and FreeBSD

[Daniel B. <da...@te...>]

"Raymond A. Wiker" <rw...@c2...> writes:

> 	Given the fact that the definitions in unix.lisp are difficult
> to get right, I have started thinking about generating this file
> automatically (or at least semi-automatically), using a program to
> read through header files and generate the appropriate lisp
> definitions. Any thoughts on this?

I needed to do this for my sockets library, and eventually decided on
an approach that takes a data file full of sexprs specifying which
values we'd like to know about, then writes out, compiles and executes
a small C program that gets them.  It copes with constants, #defines,
enums, and structure offsets.  Because it uses the system's C compiler
it's as near as possible guaranteed to arrive at the same answer -
parsing C from Lisp code really doesn't soud like a fun idea
especially when you start looking at the hairy conditional #include
mess that is /usr/include on Glibc-based systems.

CVS repository (look at INSTALL, constants.ccon, defs-to-lisp.lisp and
the Makefile)

    http://loaclhost.telent.net/cgi-bin/cvsweb/telent/sockets/

:sockets information and links to download from 

    http://ww.telent.net/lisp/sockets.html

My mk-defsystem-for-sbcl patch, if anybody else needs one, at 

    http://ww.telent.net/lisp/defsystem-v1.12.diff

> 	mk:defsystem has quite a few feature-tests on :cltl2, which is
> predefined for CMUCL. Should this also be put on the features-list for
> SBCL, or would it be better to change defsystem.lisp to use :ansi-cl
> or :sbcl directly?

My guess: the latter.  The language specificed by cltl2 is not
actually compatible with ansi cl, according to the hyperspec.

-dan