GCC Bugzilla will be upgraded from version 4.4.9 to 5.0rc3 on Saturday, April 25, starting around 17:00 UTC. The upgrade process should only last a few minutes. Check bug 64968 for details.
Bug 44735 - ICE on FORALL with character array pointer
ICE on FORALL with character array pointer
Status: NEW
Product: gcc
Classification: Unclassified
Component: fortran
4.6.0
: P3 normal
: ---
Assigned To: Not yet assigned to anyone
: ice-on-valid-code
Depends on:
Blocks: 19276 32834
  Show dependency treegraph
 
Reported: 2010-06-30 21:34 UTC by Joe Krahn
Modified: 2014-08-31 14:36 UTC (History)
3 users (show)

See Also:
Host:
Target:
Build:
Known to work:
Known to fail: 4.3.4, 4.4.0, 4.5.1, 4.6.0, 4.9.0
Last reconfirmed: 2013-03-17 15:00:00


Attachments
The beginings of a fix (8.28 KB, patch)
2010-08-18 18:55 UTC, Paul Thomas
Details | Diff

Note You need to log in before you can comment on or make changes to this bug.
Description Joe Krahn 2010-06-30 21:34:48 UTC
$ cat bug3.f90
subroutine bug
  character(len=10) :: F_string
  character(len=1), dimension(:), pointer :: p_chars
  forall (i=1:len(F_string))
    F_string(i:i) = p_chars(i)
  end forall
end subroutine bug

$ /usr/local/gfortran/bin/gfortran -c bug3.f90
bug3.f90: In function ‘bug’:
bug3.f90:1:0: internal compiler error: Segmentation fault
Please submit a full bug report,
with preprocessed source if appropriate.
See <http://gcc.gnu.org/bugs.html> for instructions.
Comment 1 Tobias Burnus 2010-07-01 06:21:38 UTC
Confirm.

Valgrind shows:

==11328== Invalid read of size 8
==11328==    at 0x6CBB86: fold_convert_loc (fold-const.c:1856)
==11328==    by 0x58E7A3: allocate_temp_for_forall_nest_1 (trans-stmt.c:2583)
==11328==    by 0x58F394: gfc_trans_assign_need_temp (trans-stmt.c:2702)
==11328==    by 0x59284C: gfc_trans_forall_1 (trans-stmt.c:3158)
==11328==    by 0x551DFF: trans_code (trans.c:1205)

Failing is the following line in fold-const.c
  tree orig = TREE_TYPE (arg);
which is the second argument in the following line of allocate_temp_for_forall_nest_1:
   unit = fold_convert (gfc_array_index_type, TYPE_SIZE_UNIT (type));

It is set in gfc_trans_assign_need_temp:

  if (expr1->ts.type == BT_CHARACTER && expr1->ts.u.cl->length)
    ...
  else
    type = gfc_typenode_for_spec (&expr1->ts);

For the example, expr1->ts.type == BT_CHARACTER but expr1->ts.u.cl->length == NULL.

The symbol itself has:
(gdb) p *expr1->symtree->n.sym->ts.u.cl->length->value.integer._mp_d
$8 = 10

And the size is given by the expression:
(gdb) p expr1->ref->u.ss.start->expr_type 
$19 = EXPR_VARIABLE
Comment 2 Paul Thomas 2010-08-18 18:55:31 UTC
Created attachment 21513 [details]
The beginings of a fix

This PR is going to drive me mad!

The immediate cause is a failure to get the TYPE_SIZE_UNIT to calculate the size of the temporary needed for the assignment.  This fixes that part of the problem but moves the focus to the assignment itself.  Something is badly wrong with the temporary's TREE_TYPE but I do not see what it is right now.

Let this be the record of where I got to....

Paul
Comment 3 Thomas Koenig 2010-12-29 22:18:34 UTC
Hi Paul,

The problem may well be where Joe pointed it out.

In

  if (expr1->ts.type == BT_CHARACTER && expr1->ts.u.cl->length)
    {
      if (!expr1->ts.u.cl->backend_decl)
	{
	  gfc_se tse;
	  gfc_init_se (&tse, NULL);
	  gfc_conv_expr (&tse, expr1->ts.u.cl->length);
	  expr1->ts.u.cl->backend_decl = tse.expr;
	}
      type = gfc_get_character_type_len (gfc_default_character_kind,
				         expr1->ts.u.cl->backend_decl);
    }

the fact that expr1 has a substring reference is not taken into account,
so it might be necessary to calculate the length of the string
from expr1->ref->u.ss.end - expr1->ref.u.ss.start + 1.  You may have
to take care not to calculate any functions twice, though.

Hope this helps a little bit.
Comment 4 Thomas Koenig 2013-03-17 15:24:44 UTC
Still fails:

ig25@linux-fd1f:~/Krempel/Forall> valgrind /home/ig25/lib/gcc/x86_64-unknown-linux-gnu/4.9.0/f951 bug.f90 
==24599== Memcheck, a memory error detector
==24599== Copyright (C) 2002-2011, and GNU GPL'd, by Julian Seward et al.
==24599== Using Valgrind-3.7.0 and LibVEX; rerun with -h for copyright info
==24599== Command: /home/ig25/lib/gcc/x86_64-unknown-linux-gnu/4.9.0/f951 bug.f90
==24599== 
 bug==24599== Invalid read of size 2
==24599==    at 0x786F01: fold_convert_loc(unsigned int, tree_node*, tree_node*) (tree.h:3797)
==24599==    by 0x5FE16F: allocate_temp_for_forall_nest_1(tree_node*, tree_node*, stmtblock_t*, tree_node**) (trans-stmt.c:3270)
==24599==    by 0x5FEA47: gfc_trans_assign_need_temp(gfc_expr*, gfc_expr*, tree_node*, bool, forall_info*, stmtblock_t*) (trans-stmt.c:3391)
==24599==    by 0x602DE0: gfc_trans_forall_1(gfc_code*, forall_info*) (trans-stmt.c:3868)
==24599==    by 0x5A6847: trans_code(gfc_code*, tree_node*) (trans.c:1569)
==24599==    by 0x5CD8AE: gfc_generate_function_code(gfc_namespace*) (trans-decl.c:5397)
==24599==    by 0x565800: gfc_parse_file() (parse.c:4468)
==24599==    by 0x5A20E5: gfc_be_parse_file() (f95-lang.c:189)
==24599==    by 0x94D725: compile_file() (toplev.c:543)
==24599==    by 0x94F5A9: toplev_main(int, char**) (toplev.c:1876)
==24599==    by 0x5A38454: (below main) (in /lib64/libc-2.15.so)
==24599==  Address 0x0 is not stack'd, malloc'd or (recently) free'd
Comment 5 Dominique d'Humieres 2014-08-29 09:44:36 UTC
This PR is fixed by the patch at https://gcc.gnu.org/ml/fortran/2014-08/msg00114.html. Test that can be executed:

subroutine bug
  character(len=10) :: F_string
  character(len=1), dimension(:), pointer :: p_chars
  allocate(p_chars(10))
  p_chars = ['a','b','c','d','e','f','g','h','i','j']
  forall (i=1:len(F_string))
    F_string(i:i) = p_chars(i)
  end forall
  print *, F_string
  deallocate(p_chars)
end subroutine bug
call bug()
end

[Book15] f90/bug% a.out
 abcdefghij
Comment 6 Tobias Burnus 2014-08-30 18:47:36 UTC
Author: burnus
Date: Sat Aug 30 18:47:04 2014
New Revision: 214757

URL: https://gcc.gnu.org/viewcvs?rev=214757&root=gcc&view=rev
Log:
2014-08-30  Tobias Burnus  <burnus@net-b.de>

        PR fortran/62278
        PR fortran/44735
        PR fortran/60593
        * dependency.c (gfc_check_dependency): Allow for optimizations
        in the pointer-alias check.


Modified:
    trunk/gcc/fortran/ChangeLog
    trunk/gcc/fortran/dependency.c
Comment 7 Thomas Koenig 2014-08-31 14:36:43 UTC
The fix looks good.

Commit the test case from comment #5 and close?