> On 10 Feb 2017, at 15:50, Pavel Grunt <pgrunt@xxxxxxxxxx> wrote: > > On Thu, 2017-02-09 at 18:46 +0100, Christophe de Dinechin wrote: >> I tend to agree. I don’t see how any legitimate GError * (from >> g_error_new or the like) would have a NULL message. So this is >> really defensive coding. >> > nothing prevents calling g_error_new with a NULL message :) I know it > is not our case Actually, digging deeper, I think this is not valid. See https://github.com/GNOME/glib/blob/master/glib/gerror.c#L406. There is a g_warn_if_fail in that case. Also, the error message is the result of a strdup, and is allocated. And at https://github.com/GNOME/glib/blob/master/glib/gerror.c#L491, the call to g_free is unconditional. So it looks to me like the internal assumption in all these routines is that message cannot be NULL. That being said, nothing prevents you from doing something like GError *bad_error = malloc(27); In that case, we are pretty much screwed. So in light of the g_warn_if_fail, I think that my second patch should be rejected. Christophe _______________________________________________ virt-tools-list mailing list virt-tools-list@xxxxxxxxxx https://www.redhat.com/mailman/listinfo/virt-tools-list
