Re: MinGW status

[Ralf Wildenhues]

* Bob Friesenhahn wrote on Sun, Oct 03, 2004 at 12:00:28AM CEST:
> On Sat, 2 Oct 2004, Ralf Wildenhues wrote:
> 
> >* Bob Friesenhahn wrote on Sat, Oct 02, 2004 at 09:51:57PM CEST:
> >>This build script represents the current state of building CVS libltdl
> >>under MinGW.  Symbols show up as unresolved.  For example 'lt__malloc'
> >>is reported as unresolved, but nm shows that there is a '_lt__malloc'
> >>symbol in lt_alloc.o.
> >
> >Might I suggest you try the patch below?  As I have no MinGW available I
> >cannot try it myself, sorry.  It might need some adapting.
> 
> Adapting?  My foot!  The patch worked! :-))

Yeah, but probably one would want this instead (undef LT_SCOPE again
afterwards), see below.  Not that I know whether argz.h is used outside
of Libtool or what this requirement is for.

> >*snip*
> >>//scooby/bfriesen/src/gnu/libtool/libltdl/lt__private.h:121: warning: 
> >>array
> >>`lt__error_strings' assumed to have one element
> >*snip*
> >
> >This is bad news indeed (but a different matter).  It essentially means
> >that we are breaking binary compatibility as soon as we add another
> >warning, right? (The array size is part of the ABI).  Very unfortunate.
> 
> At first glance it does not appear to be a problem since the interface 
> is merely an array of string pointers and the index into the array 
> should be generated by libltdl itself so the array index should always 
> be in bounds.  Am I wrong?

Well, I'm not an expert on this subject.  In Drepper's `How To Write
Shared Libraries', chapter 3.1, he mentions that ``On platforms which
require copy relocations to handle access to variables defined in DSOs
in the main application (such as IA-32) the size of the variable must
not change at all.  Otherwise variables might increase in size.''
I don't what exactly is meant here, and consequently I don't know if
it applies to this case.

> >BTW, the #define hacks make us get away with the namespace w.r.t argz.h,
> >but not w.r.t slist.h.  I think that needs to be addressed, right?
> 
> Don't know.  What I do know is that 'make' for libtool has now 
> succeeded under MinGW.
> 
> As far as how libltdl's build goes, I do not like it at all that 
> libltdl is unnecessary introducing another installed library.  I don't 
> see a need for it.  I also don't like it that libltdl's Makefile.am is 
> referencing Automake's internal variables.

Which ones are Automake internal and not published interface?

> We should be able to reference the necessary source file's in the 
> loaders subdirectory so that no extra convenience library is 
> necessary.  This approach risks the possibility that Automake might 
> demand that the dependent package include a `compile' script though. 
> According to libtool's own documentation, this should work for 
> Automake 1.4 and later.

At the moment, I do not feel qualified to comment on this and the extra
library requirement you mention above.

Regards,
Ralf

Updated patch:

2004-10-02  Ralf Wildenhues <address@hidden>

        * libltdl/argz_.h, libltdl/lt__alloc.h: export functions
        portably.


Index: libltdl/argz_.h
===================================================================
RCS file: /cvsroot/libtool/libtool/libltdl/argz_.h,v
retrieving revision 1.1
diff -u -r1.1 argz_.h
--- libltdl/argz_.h     22 Apr 2004 22:47:20 -0000      1.1
+++ libltdl/argz_.h     3 Oct 2004 09:21:11 -0000
@@ -35,23 +35,30 @@
 
 #if defined(LTDL)
 #  include "lt__glibc.h"
+#  include "lt_system.h"
+#else
+#  define LT_SCOPE
 #endif
 
 #if defined(_cplusplus)
 extern "C" {
 #endif
 
-error_t        argz_append     (char **pargz, size_t *pargz_len,
-                        const char *buf, size_t buf_len);
-error_t        argz_create_sep (const char *str, int delim,
-                        char **pargz, size_t *pargz_len);
-error_t        argz_insert     (char **pargz, size_t *pargz_len,
-                        char *before, const char *entry);
-char * argz_next       (char *argz, size_t argz_len, const char *entry);
-void   argz_stringify  (char *argz, size_t argz_len, int sep);
+LT_SCOPE error_t argz_append   (char **pargz, size_t *pargz_len,
+                                const char *buf, size_t buf_len);
+LT_SCOPE error_t argz_create_sep(const char *str, int delim,
+                                char **pargz, size_t *pargz_len);
+LT_SCOPE error_t argz_insert   (char **pargz, size_t *pargz_len,
+                                char *before, const char *entry);
+LT_SCOPE char *         argz_next      (char *argz, size_t argz_len, const 
char *entry);
+LT_SCOPE void   argz_stringify (char *argz, size_t argz_len, int sep);
 
 #if defined(_cplusplus)
 }
 #endif
 
+#if !defined(LTDL)
+#  undef LT_SCOPE
+#endif
+
 #endif /*!defined(LT__ARGZ_H)*/
Index: libltdl/lt__alloc.h
===================================================================
RCS file: /cvsroot/libtool/libtool/libltdl/lt__alloc.h,v
retrieving revision 1.4
diff -u -r1.4 lt__alloc.h
--- libltdl/lt__alloc.h 15 Jul 2004 12:14:47 -0000      1.4
+++ libltdl/lt__alloc.h 2 Oct 2004 20:35:09 -0000
@@ -43,14 +43,14 @@
                                                        } LT_STMT_END

/* If set, this function is called when memory allocation has failed.  */
-extern void (*lt__alloc_die) (void);
+LT_SCOPE void (*lt__alloc_die) (void);

-void *lt__malloc (size_t n);
-void *lt__zalloc (size_t n);
-void *lt__realloc (void *mem, size_t n);
-void *lt__memdup (void const *mem, size_t n);
+LT_SCOPE void *lt__malloc (size_t n);
+LT_SCOPE void *lt__zalloc (size_t n);
+LT_SCOPE void *lt__realloc (void *mem, size_t n);
+LT_SCOPE void *lt__memdup (void const *mem, size_t n);

-char *lt__strdup (const char *string);
+LT_SCOPE char *lt__strdup (const char *string);

LT_END_C_DECLS