Dear All,
The fix to eliminate this ICE is trivial.
Bootstrapped and regtested on FC21/x86_64 - OK for 5 to 7 branches?
Cheers
Paul
2016-06-11 Paul Thomas <pa...@gcc.gnu.org>
PR fortran/70673
* frontend-passes.c (realloc_string_callback): Add a call to
gfc_dep_compare_expr.
2016-06-11 Paul Thomas <pa...@gcc.gnu.org>
PR fortran/70673
* gfortran.dg/pr70673.f90: New test.
Index: gcc/fortran/frontend-passes.c
===================================================================
*** gcc/fortran/frontend-passes.c (revision 237168)
--- gcc/fortran/frontend-passes.c (working copy)
*************** realloc_string_callback (gfc_code **c, i
*** 175,180 ****
--- 175,187 ----
if (!gfc_check_dependency (expr1, expr2, true))
return 0;
+ /* gfc_check_dependency doesn't always pick up identical expressions.
+ However, eliminating the above sends the compiler into an infinite
+ loop on valid expressions. Without this check, the gimplifier emits
+ an ICE for a = a, where a is deferred character length. */
+ if (!gfc_dep_compare_expr (expr1, expr2))
+ return 0;
+
current_code = c;
inserted_block = NULL;
changed_statement = NULL;
Index: gcc/testsuite/gfortran.dg/pr70673.f90
===================================================================
*** gcc/testsuite/gfortran.dg/pr70673.f90 (revision 0)
--- gcc/testsuite/gfortran.dg/pr70673.f90 (working copy)
***************
*** 0 ****
--- 1,25 ----
+ ! { dg-do run }
+ !
+ ! Test the fix for PR70673
+ !
+ ! Contributed by David Kinniburgh <davidgkinnibu...@yahoo.co.uk>
+ !
+ module m
+ contains
+ subroutine s(inp)
+ character(*), intent(in) :: inp
+ character(:), allocatable :: a
+ a = a ! This used to ICE.
+ a = inp
+ a = a ! This used to ICE too
+ if ((len (a) .ne. 5) .or. (a .ne. "hello")) call abort
+ a = a(2:3) ! Make sure that temporary creation is not broken.
+ if ((len (a) .ne. 2) .or. (a .ne. "el")) call abort
+ deallocate (a)
+ a = a ! This would ICE too.
+ end subroutine s
+ end module m
+
+ use m
+ call s("hello")
+ end
