[Bug fortran/110723] New: ICE with allocatable character lhs and parenthesized array with vector subscript

[Bug fortran/110723] New: ICE with allocatable character lhs and parenthesized array with vector subscript

[anlauf at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=110723

            Bug ID: 110723
           Summary: ICE with allocatable character lhs and parenthesized
                    array with vector subscript
           Product: gcc
           Version: unknown
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: fortran
          Assignee: unassigned at gcc dot gnu.org
          Reporter: anlauf at gcc dot gnu.org
  Target Milestone: ---

Found while analyzing pr95947:

program p
  implicit none
  character(len=1)              :: m(3) = ['a','b','c']
  character(len=:), allocatable :: n(:)
  n =   m([2])
  n =   m([2,3])
  n =  (m([2]))   ! ICE
  n =  (m([2,3])) ! ICE
end

We end up with EXPR_OP in the assignment for the parenthesized variants,
leading to bad gimple.

Tentative fix:

diff --git a/gcc/fortran/trans-expr.cc b/gcc/fortran/trans-expr.cc
index ef3e6d08f78..63d39c2516e 100644
--- a/gcc/fortran/trans-expr.cc
+++ b/gcc/fortran/trans-expr.cc
@@ -11954,6 +11954,12 @@ gfc_trans_assignment_1 (gfc_expr * expr1, gfc_expr *
expr2, bool init_flag,

   rss = NULL;

+  /* Strip left-over parentheses around rhs variables (not expressions).  */
+  while (expr2->expr_type == EXPR_OP
+        && expr2->value.op.op == INTRINSIC_PARENTHESES
+        && expr2->value.op.op1->expr_type == EXPR_VARIABLE)
+    expr2 = expr2->value.op.op1;
+
   if (expr2->expr_type != EXPR_VARIABLE
       && expr2->expr_type != EXPR_CONSTANT
       && (expr2->ts.type == BT_CLASS || gfc_may_be_finalized (expr2->ts)))

Not sure if this is the right place.

[Bug fortran/110723] ICE with allocatable character lhs and parenthesized array with vector subscript

[anlauf at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=110723

--- Comment #1 from anlauf at gcc dot gnu.org ---
(In reply to anlauf from comment #0)
> Not sure if this is the right place.

Actually, the following still fails:

  n =  (m([2])//"")   ! ICE

:-(

Generally stripping parentheses generates a failure in
gfortran.dg/reassoc_2.f90, which is bad...