On Wed, Jan 14, 2015 at 8:26 AM, Patrick Palka <patr...@parcs.ath.cx> wrote: > On Wed, Jan 14, 2015 at 8:05 AM, Jason Merrill <ja...@redhat.com> wrote: >> On 01/13/2015 10:54 PM, Patrick Palka wrote: >>> >>> + type = error_mark_node; >>> + goto out; >> >> >> Why exit early in the explicit instantiation cases? Doesn't it work to give >> the error and continue? >> >> Jason >> > > Yes it does. I changed it to an early exit in the last minute for no > good reason. But an earlier version of the patch that continued > instead of exiting worked correctly.
Actually I recall having issues with not exiting early in the "an explicit instantiation may not have a definition" case. But in the "an explicit instantiation must be preceded by template" case one could safely continue (even though it goes on to be processed as a forward declaration of a template specialization instead of a mistyped explicit instantiation...)
