On Tue, Nov 19, 2013 at 5:25 PM, Christophe Fergeau <cferg...@redhat.com> wrote: > On Tue, Nov 19, 2013 at 05:12:39PM +0100, Marc-André Lureau wrote: >> On Tue, Nov 19, 2013 at 5:09 PM, Christophe Fergeau <cferg...@redhat.com> >> wrote: >> > On Tue, Nov 19, 2013 at 11:05:13AM -0500, Marc-André Lureau wrote: >> >> >> >> >> >> ----- Original Message ----- >> >> > On Tue, Nov 19, 2013 at 04:14:22PM +0100, Marc-André Lureau wrote: >> >> > > Just like any other C function >> >> > >> >> > NACK, I just added this annotation as not checking its return value will >> >> > cause very hard to track down failures (no usable backtraces), so >> >> > better to >> >> > force it to be checked, especially as this is only used internally. >> >> >> >> This is really different from any other C API. >> > >> > GList, GSList make extensive use of it, and it's also used in a several >> > other >> > places in glib. >> >> Because they return object that contains allocated data, or are >> critical to there mode of working. > > _init() succeeding is critical to the coroutine being properly initialized, > and once again, not checking it can cause hard to diagnose failures, so > this function is special. > >> Any other function that returns an error code doesn't have >> G_GNUC_WARN_UNUSED_RESULT. >> >> I'll modify the patch to abort, and return void. > > Hrm, all of this would have been more useful during the review of "Check > coroutine_init() return value" as this is an alternate approach I > experimented with, but then I preferred to try to avoid adding more abort() > to spice-gtk.
spice-gtk follows glib/gtk principle of API and failure. It's not too late to fix this (and it's not too late to admit that G_GNUC_WARN_UNUSED_RESULT. shouldn't be used for error codes ;) -- Marc-André Lureau _______________________________________________ Spice-devel mailing list Spice-devel@lists.freedesktop.org http://lists.freedesktop.org/mailman/listinfo/spice-devel
