[Gcl-devel] clc and pathnames

[Dennis Decker Jensen]

Hi Camm

The previous weekend I made the script for common-lisp-controller
(clc).  However, I cannot test it until pathnames, physical as
well as logical, are implemented in GCL.  Just about every aspect
(by nature) of clc depends on it.

I've noticed you've had some problems regarding pathnames
versus namestrings.  Once ANSI pathnames are in, you should not
need to fuzz around with `.', `..' and such things in strings.
Instead of substituting or concatenating strings, think of
merging pathnames.  Use pathnames everywhere references to files
are used -- instead of strings, and rely on ANSI CL functions
to do the converting to namestrings only in the last possible
instance where they absolutely are needed.

While skimming the docs I found out that pathnames unfortunately
are not self-contained.  The directory component of pathnames
allows some syntax with semantic behaviour, which relies on data
in the actual filesystem underneath (symbolic links, shortcuts
or what ever the platform calls them), but then again that
should not be necessary to deal with until namestrings are used
(in the last possible instance, as late as possible).  It also
means that one cannot reconstruct the original pathname from
the constructed namestring.  Hmmm...  Need to do more digging.

Sorry for the longish letter.  I tend to speculate too much
out in the open.

This means that CL-controller will have to wait for pathnames
in 2.7.X

Yours

Dennis Decker Jensen

P.S. Leaving aside pathnames, a lot of the cl-packages also
relies on UFFI -- Universal Foreign Function Interface to work.
AFAICT it can be implemented on top of a native FFI.  Although it
is far from perfect (so people on c.l.c. keep telling) it is
well supported, so maybe this also should be on a wish list
or something?  BTW. I find GCL's connection to C compelling in
this area, even though I haven't programmed in C for four years
now.  Perhaps FFI is so simple in GCL that UFFI is not needed?
Obviously I don't know...