http://gcc.gnu.org/bugzilla/show_bug.cgi?id=46484
Tobias Burnus <burnus at gcc dot gnu.org> changed:
What |Removed |Added
----------------------------------------------------------------------------
Priority|P3 |P4
Target Milestone|--- |4.6.0
Summary|Should reject |[4.5/4.6 Regression] Should
|ALLOCATED(non-variable |reject
|expression ) |ALLOCATED(non-variable
| |expression )
--- Comment #2 from Tobias Burnus <burnus at gcc dot gnu.org> 2010-11-15
15:25:48 UTC ---
(In reply to comment #1)
> Longer test case is gfortran.dg/allocatable_scalar_5.f90 (cf. 46485).
PR 46485
Mark as 4.5/4.6 regression. It is not true regression as 4.3/4.4 rejected it
with "just" with
Internal Error at (1):
gfc_variable_attr(): Expression isn't a variable
but still ...
The reason for the failure is that check.c's variable_check accepts:
|| (e->expr_type == EXPR_FUNCTION
which at a glance does not make sense. Currently, I can only imagine a Fortran
2008 case, which allows pointer-returning functions in place of variables. Cf.
PR 40054 (and PR 46100).
I traced the line back to:
Rev. 81764: "Merge tree-ssa-20020619-branch into mainline." (2004-05-13)
Thus, it seems as if this line was never valid.
* * *
Draft patch: The following seems to be sensible and some incomplete regtesting
suggests that it probably works:
diff --git a/gcc/fortran/check.c b/gcc/fortran/check.c
index 51ea877..cc36fea 100644
--- a/gcc/fortran/check.c
+++ b/gcc/fortran/check.c
@@ -491,10 +491,8 @@ variable_check (gfc_expr *e, int n)
return FAILURE;
}
- if ((e->expr_type == EXPR_VARIABLE
+ if (e->expr_type == EXPR_VARIABLE
&& e->symtree->n.sym->attr.flavor != FL_PARAMETER)
- || (e->expr_type == EXPR_FUNCTION
- && e->symtree->n.sym->result == e->symtree->n.sym))
return SUCCESS;
gfc_error ("'%s' argument of '%s' intrinsic at %L must be a variable",
