------- Comment #1 from kargl at gcc dot gnu dot org 2009-10-19 22:06 ------- (In reply to comment #0) > In gfc_match_allocate one finds a call to gfc_resolve_expr, which resolves the > optional SOURCE expression. > > By definition, resolving should be done at resolution stage, and not already > when matching the ALLOCATE statement. The canonical way to do it would be to > add something like 'gfc_resolve_expr (code->expr3)' in resolve_code, where > already code->expr1 and code->expr2 are being resolved (the SOURCE expr is > being stored in code->expr3). > > So, unless there are very good reasons for the resolution to happen in > gfc_match_allocate, the call to gfc_resolve_expr should be moved to resolve.c > (together with the various checks for the SOURCE expression that follow it). > > I think the questionable code was committed as part of r151023 by Steve, when > he implemented the SOURCE tag in ALLOCATE statements.
Yes, I added the call to gfc_resolve_expr. Unfortnately, I can't remember why :( other than tmp->ts.type is not properly set in some situation. If you move this to resolve.c, you'll need to move some or all of the error checking involving tmp->ts.type. Perhaps, your changes to support CLASS and allocatable scalars has removed the need for gfc_resolve_expr here. -- http://gcc.gnu.org/bugzilla/show_bug.cgi?id=41758
