Re: MinGW build of 2.6.5: Still having auto-import problems ;-(

[Toralf Lund <toralf procaptura com>]

> > I'm now trying to build gtkmm 2.6.5 via g++ set up as cross compiler
> > with i386-mingw32 output (now that I've got 2.2.12  to work after
> > tweaking a bit - I want to support 2.0 *and* 2.4 APIs.) Unfortunately, I
> > get the problem described in
> > http://bugzilla.gnome.org/show_bug.cgi?id=309030 - despite the fact that
> > 2.6.5 is supposed to contain a fix for that. It seems like the problem
> > is that the GTKMM_API will *not* contain "__declspec(dllimport)" [ ... ]
> > #else
> >
> > - notice how GLIBMM_DLL won't be defined in the "MinGW" case. I tried to
> > rebuild glibmm after inserting
> >    
>
> Defining GLIBMM_DLL (hence __declspec(dllimport/dllexport)) is not
> required as all symbols are exported at link time when using the
> -Wl,--export-all-symbols option.
That's actually not quite true, I think. "dllimport" is required in some 
cases *even when --export-all-symbols is being used* - as far as I can 
tell from the tests I've done, and also the comments from others in the 
above mentioned bug report (I think this is essentially what it is all 
about.) Maybe there aren't any such cases in glibmm, but there certainly 
are in gtkmm (this is also what the bug report is all about...)

Also, I'm a bit confused about the meaning of the macros here. I thought 
GLIBMM_DLL meant just "a dll is being built" or "a dll exists" (for the 
glibmm library), in which case it ought to be defined regardless of 
whether you want special __declspec directives or not (there should be 
additional tests deciding if that's necessary.) I may well be wrong, 
though, or perhaps I wasn't even trying to build a dll (can't check on 
the machine I'm on now.)

> Now, another important option to use at
> link time is -no-undefined (see build_shared/Makefile_build.am_fragment)
> and I don't see it in your g++ command line options. As this option is
> automatically added to LDFLAGS at configure time when the host is either
> *-mingw32-* or *-cygwin-*, I guess something you did not use the
> --host=i686-pc-mingw32 option with configure or there is an error in the
> configure.in file.
>
> Anyway, either try again with --host=i686-pc-mingw32 or manually add
> -no-undefined to LDFLAGS in glib/glibmm/Makefile.
>  
Hmmm. I *think* I passed the correct host settings, but you never 
know... Again, I'm not able to test right now. Also, I don't necessarily 
want to change the glibmm setup anyway - maybe it's a gtkmm patch I 
need. It's a bit surprising that this would be necessary, though, based 
on the fact that bug 309030 is marked as resolved, but I suppose that 
also goes for a change to the glibmm setup. Maybe a different glibmm 
version was used by others who tested this?

- Toralf