Re: [Clg-devel] CLG on Mac OSX

[Thomas D. <th...@de...>]

Hi, and many thanks for your help.

The OSX/Darwin version of nm prints out entries for undefined
symbols, and doesn't seem to have any equivalent of the
--defined-only option.  (For example output, see
http://www.derkholm.net/thomas/nm-example.txt -- let me know
if you want to see output using different options).  The undefined
symbols are causing the loop in %find-types-in-library to
terminate prematurely.  There's also another issue: there's an
extra '_' prepended to each symbol name.  There's a little patch
which solves these two issues below.

Once this is applied, I can get much further through the build
process.  Unforturtunately, I don't quite have a working system
yet: I'm getting the following while compiling the Cairo support:

; compiling (EXPORT-FROM-FILE #P"CLG:CAIRO;CAIRO.LISP")
; file: /Users/thomas/Software/clg/cairo/export.lisp
; in: EXPORT-FROM-FILE #P"CLG:CAIRO;CAIRO.LISP"
;     (AUTOEXPORT:EXPORT-FROM-FILE #P"CLG:CAIRO;CAIRO.LISP")
; 
; caught ERROR:
;   (during macroexpansion of (EXPORT-FROM-FILE #P"CLG:CAIRO;CAIRO.LISP"))
;   error opening #P"CLG:CAIRO;CAIRO.LISP.NEWEST": No such file or directory


I'll keep investigating.

       Thomas.


--- glib/gtype.lisp.orig        2006-08-30 18:15:11.000000000 +0100
+++ glib/gtype.lisp     2006-08-30 18:19:17.000000000 +0100
@@ -208,7 +208,8 @@
           (run-program
            "/usr/bin/nm" 
            #+clisp :arguments
-           (list #-darwin"--defined-only" #-darwin"-D" "-g" #+darwin"-f" (namestring (truename pathname)))
+           (list #-darwin"--defined-only" #-darwin"-D" "-g" #+darwin"-f" #+darwin"-s" 
+                 #+darwin"__TEXT" #+darwin"__text" (namestring (truename pathname)))
            :output :stream :wait nil)))
       (unwind-protect
          (loop 
@@ -219,7 +220,7 @@
                                     nil))
                              (pos (position #\Space line :from-end t)))
                         (when (and line #+darwin(char= (char line (1- pos)) #\T))
-                          (subseq line (1+ pos))))
+                          (subseq line #-darwin(1+ pos) #+darwin(+ pos 2))))
           while symbol
           when (and
                 (> (length symbol) 9)



On Wed, Aug 30, 2006 at 01:34:24PM +0200, Espen S Johnsen wrote:
> Thomas Down <th...@de...> writes:
> 
> > I'm trying to install CLG on my Mac, under SBCL 0.9.16 (also tested
> > on 0.9.14).  In case it matters, I've got an Intel processor -- but
> > I think this issue is probably a general OSX/Darwin problem.  I'm using
> > a recent checkout from the CLG CVS repository.
> 
> Thanks. Your feedback is very welcome.
> 
>  
> > I've updated asdf-extensions.lisp to understand library names ending
> > in .dylib rather than .so, and also fixed a few hard-coded library
> > names elsewhere in the lisp code. 
> 
> I have updated the code to not relay on hardcoded library extensions.
> 
> >  When linking alien.dylib, I get a linker error:
> > 
> >   /usr/bin/ld: Undefined symbols:
> >   _g_closure_add_finalize_notifier
> >   _g_closure_new_simple
> >   _g_closure_set_meta_marshal
> >   collect2: ld returned 1 exit status
> > 
> > If I relink this against libgobject-2.0.dylib,
> 
> It wasn't really necessary to call these functions from C so I have
> removed the code doing this. Although the problem may reappear later in
> the build process.
> 
> > all appears to be well until:
> > 
> >   debugger invoked on a UNDEFINED-FUNCTION:
> >     The function |g_value_get_type| is undefined.
> >   [...]
> >   (SB-KERNEL:%COERCE-CALLABLE-TO-FUN |g_value_get_type|)
> > 
> > The problem is occurring while loading "gparam.fasl", offending line
> > seems to be:
> > 
> >    (REGISTER-TYPE GVALUE |g_value_get_type|)
> 
> Until the long-awaited full introspection becomes available i GLib, clg
> uses nm to find all types in a library by searching for symbols ending
> with _get_type. The cause of the problem above is that nm doesn't take
> the same options in OSX as the GNU nm. I have checked in some untested
> code that may solve this. It would be nice if you could test it again
> and report back to the list.
> 
> -- 
> Espen