Re: Change to gmodule

[Owen Taylor <otaylor redhat com>]

"Gary V.Vaughan" <gary oranda demon co uk> writes:

> On Tuesday 29 May 2001 12:06 pm, Tim Janik wrote:
> > On Tue, 29 May 2001, Padraig O'Briain wrote:
> > > 2) The code in version 1.33 which checked whether the file_name ended in
> > > G_MODULE_SUFFIX or ".la" and if not, appended a suffix before calling
> > > _g_module_open() is not in version 1.35.
> >
> > hm, that you need to specify libgail.so:libtestobject.so is odd, module
> > loading works here without path and without suffix, a suffix even.
> > please try out the new CVS version.
> 
> I expect my patch to make gmodule a thin wrapper for libltdl would fix this 
> problem... now that libtool-1.4 is out, is there anything holding up 
> acceptance?

To give my point view (and only my point of view), I don't think
making gmodule a wrapper for libltdl makes sense. As downsides:

 - It adds another shared library to be loaded at runtime
 - It adds another runtime dependency
 - It adds a dependency on something we don't control.

The upsides are obviously:

 - libltdl may work better now than gmodule does
 - We don't have to maintain dynamic loading code
 - there may be increased memory sharing with non-glib packages
   (say aspell) that use libltdl directly.

But none of these advantages come from using gmodule as a wrapper
around libltdl, they simply come from using libltdl. 

If libltdl is really nice enough that we would want to make gmodule a
wrapper around it, then we should simply encourage people to use
libltdl and deprecate gmodule; if libltdl has a significantly worse
interface than gmodule, than its not going to be easy to wrap
a nicer interface around it.

Regards,
                                        Owen