RFC: Non-indirect-dispatch and failed loading of classes

Thu Sep 8 17:35:00 GMT 2005

Tom Tromey writes:
 > >> We do get an error, but it's not a fatal error and we die
 > >> later trying to lay out this class:
 > 
 > Andrew> OK, I see. The traditional way to fix things is to use
 > Andrew> error_mark_node, but it leads to uglified code with checks everywhere.
 > 
 > Andrew> I'm not sure how best to fix this one, but I'd try leaving
 > Andrew> layout_class_methods early, and if that doesn't work terminating the
 > Andrew> compilation.
 > 
 > There are a few bugs in bugzilla where we emit an error and then
 > proceed to crash or hang or something. The circular hierarchy bug
 > comes to mind.
I've never prioritized them, because an ICE after a correct error
message on incorrect code seems lik the most innocuous kind of
failure. On the other hand, I really care about incorrect code.
Andrew.