Re: Change to gmodule

On Mon, 28 May 2001, Joel Becker wrote:

> On Mon, May 28, 2001 at 08:24:23PM +0200, Tim Janik wrote:
> > thanks for spotting, should be fixed in CVS now.
> 	On the subject of GModule, the error reporting is *horrible*.  I
> have an app using GModule loaded libraries, and any time there is a
> problem the g_module_error() is *always* "Cannot open libtool library".
> Looking at the source, this is because GModule tries various methods to
> open the requested module.  It tries dlopen, then reading a libtool
> archive, then appending .so and trying dlopen(), then appending .la and
> trying libtool again.  My module had a missing symbol, so dlopen()
> failed.  However, the g_module_error() with the dlerr was immediately
> overwritten by the subsequent tries with libtool-style parsing.  I
> eventually had to instrument gmodule-dl.c and recompile/install glib to
> get a meaningful error message.
> 	Tim, I'm not sure how you want to handle this.  The fallback
> methods make sense, but they shouldn't squash the original error.  Maybe
> keep the *first* error, not the last?  What do you think?

joel, this is exactly the situation that my first patch adressed.
we only do dlopen() _once_ now and preserve the error, choosing the
correct method for loading is done before dlopen() now, based on access().

> Joel


[Date Prev][Date Next]   [Thread Prev][Thread Next]   [Thread Index] [Date Index] [Author Index]