This has been bugging me while reading the documentation.  A lot of the
examples have code like (taken from chap. 19):

int 
main (int argc, char *argv[]) 
{
  GstElement *pipeline, *src, *demux;

  gst_init (&argc, &argv);

  pipeline = gst_pipeline_new ("pipeline");
  g_return_val_if_fail (pipeline != NULL, -1);


According to
http://developer.gnome.org/doc/API/2.0/glib/glib-Error-Reporting.html:
"If the programmer has screwed up, then you should use g_warning(),
g_return_if_fail(), g_assert(), g_error(), or some similar facility.
(Incidentally, remember that the g_error() function should only be used for
programming errors, it should not be used to print any error reportable via
GError.)"

In other words, g_return_val_if_fail() should be used for programmer checks
(ie, at the beginning of functions, to ensure an argument that was passed
was not NULL), not runtime errors.  gst_pipeline_new() (well,
gst_element_factory_make) may return NULL; this is a valid return value,
according to the API docs.  It is not a programmer error, but rather, a
runtime error (running out of memory, unable to load a plugin/module, etc).
 Things like that would normally use GError (if they desire details of the
failure).

To encourage usage of g_return_val_if_fail() for such situations is a very
bad idea; programmers will later on have to unlearn techiques they've been
taught from the gst docs.  In place of g_return_val_if_fail(), i'd suggest
something similar to
if (pipeline == NULL) {
    g_printerr("could not allocate pipeline!");
    return -1;
}

I'll submit patches if you agree.