Simon Josefsson <[EMAIL PROTECTED]> writes:
> (Since the error contained gss_name_t, I'm assuming you tried the
> gss-lsh branch, and I hope it is OK to provide support for it on this
> list.)
That's perfectly ok with me. I'm sorry I haven't yet had the time to
look into your gss code carefully and integrate it into my own tree.
> This point out a problem with the GABA header creation: how can I make
> the GABA stuff respect #if WITH_GSS? If someone doesn't have
> WITH_GSS, the gss header files won't be included, and gss_name_t won't
> be defined when the generated client_userauth.h is parsed. The types
> aren't used when WITH_GSS is 0, so they shouldn't be extracted by the
> GABA stuff. Perhaps GABA should run the file through CPP?
I've been considering doing some kind of preprocessing on GABA
expressions earlier, but so far I have been able to get by without
that. The easiest way is to put the code that depends on WITH_GSS in a
separate file, layouted like this:
#if HAVE_CONFIG_H
#include "config.h"
#endif
#if WITH_GSS
#include <gss.h>
#include "gss.c.x"
/* GABA:
(class
(name gss_foo)
(vars
(name . gss_name_t)))
*/
/* Various functions using gss_foo */
...
#endif /* WITH_GSS */
The file zlib.c is an example of that. I think it make sense also for
other reasons to isolate dependencies on special libraries into
separate files. A different and uglier way is to use dummy
definitions:
#if HAVE_CONFIG_H
#include "config.h"
#endif
#if WITH_GSS
#include <gss.h>
#else
/* Dummy definitions */
typdef int gss_name_t;
#endif
#include "gss.c.x"
/* GABA:
(class
(name gss_foo)
(vars
(name . gss_name_t)))
*/
#if WITH_GSS
/* Code that depends on real gss definitions */
...
#endif /* WITH_GSS */
As for adding proper cpp support to GABA, one way of doing that is
introducing a new keyword "condition". It would be used like
/* GABA:
(class
(name gss_foo)
(condition "WITH_GSS")
(vars
(name . gss_name_t)))
*/
When this expression is processed by gaba.scm, it should emit "#if
WITH_GSS"/#endif-guards around all the generated declarations and
definitions. Such a hack would go into the process-class function in
src/scm/gaba.scm (the code would need a little more restructuring to
get cpp conditions not only for expressions of type "class", but also for
"meta", "struct" and "expr"). An untested patch included below ;-)
Regards,
/Niels
Index: src/scm/gaba.scm
===================================================================
RCS file: /cvsroot/lsh/lsh/src/scm/gaba.scm,v
retrieving revision 1.15
diff -u -a -r1.15 gaba.scm
--- src/scm/gaba.scm 23 Feb 2003 18:08:58 -0000 1.15
+++ src/scm/gaba.scm 16 Aug 2003 10:42:24 -0000
@@ -121,7 +121,7 @@
(for-each (lambda (o)
(cond ((procedure? o) (o level))
((list? o) (apply out (+ 1 level) o))
- (else (display o))))
+ ((not (null? o)) (display o))))
args))
; This isn't very optimal
@@ -443,6 +443,7 @@
(define (process-class attributes)
(let* ((name (get 'name attributes cadr))
+ (condition (get 'condition attributes cadr))
(super (get 'super attributes cadr))
(vars (preprocess-vars name (get 'vars attributes cdr)))
(meta (get 'meta attributes cadr))
@@ -453,6 +454,7 @@
(free-function (make-free-function name vars)))
; (werror "baar\n")
(c-append (class-annotate name super meta)
+ (and condition (c-append "#if " condition "\n"))
"#ifndef GABA_DEFINE\n"
(make-instance-struct name super vars)
(if meta
@@ -467,7 +469,8 @@
(or free-function "")
(make-class name super mark-function free-function
meta methods)
- "#endif /* !GABA_DECLARE */\n\n"))))
+ "#endif /* !GABA_DECLARE */\n\n"
+ (and condition (c-append "#endif /* " condition " */\n")) ))))
(define (process-meta attributes)
(let ((name (get 'name attributes cadr))
_______________________________________________
lsh-bugs mailing list
[EMAIL PROTECTED]
http://lists.lysator.liu.se/mailman/listinfo/lsh-bugs
