On Fri, Sep 28, 2018 at 12:42 PM, Paolo Carlini <paolo.carl...@oracle.com> wrote: > Hi, > > the primary issue here is a rather straightforward error-recovery, where, > after a sensible (but see below) error message we ICE in > get_underlying_template - called by convert_template_argument - because we > don't handle correctly error_mark_node as TREE_TYPE, as set in > grokdeclarator. While working on it I noticed that the error message emitted > by grokdeclarator seems suboptimal, because, per 10.1.3/2, we correctly > handle the alias declaration like a typedef but, I believe, we don't want to > just say *typedef* in the error message: luckily the infrastructure to be > more accurate is fully available, because we need it to set > TYPE_DECL_ALIAS_P anyway.
OK, though as long as you're messing with this error you might want to improve its location as well. Jason