[Bug fortran/43931] New: coarrays: Wrong code (segfault) with allocatable coarrays

[Bug fortran/43931] New: coarrays: Wrong code (segfault) with allocatable coarrays

[burnus at gcc dot gnu dot org]

The following program segfaults at run time. gdb does not show a backtrace and
valgrind just has:

==27936== Jump to the invalid address stated on the next line
==27936==    at 0x8: ???
==27936==  Address 0x8 is not stack'd, malloc'd or (recently) free'd

The program is minimal. Removing either the "CALL" or the "ALLOCATE" or "four"
causes non-segfaulting code.

program test
  implicit none
  call one()
contains
  subroutine one()
    integer, allocatable :: a(:)[:,:,:]
    allocate(a(1)[-4:9,8,4:*])
  end subroutine one
  subroutine four(C)
    integer, allocatable :: C(:)[:]
 end subroutine four
end program test


-- 
           Summary: coarrays: Wrong code (segfault) with allocatable
                    coarrays
           Product: gcc
           Version: 4.6.0
            Status: UNCONFIRMED
          Keywords: wrong-code
          Severity: normal
          Priority: P3
         Component: fortran
        AssignedTo: unassigned at gcc dot gnu dot org
        ReportedBy: burnus at gcc dot gnu dot org


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=43931

[Bug fortran/43931] coarrays: Wrong code (segfault) with allocatable coarrays

[burnus at gcc dot gnu dot org]

------- Comment #1 from burnus at gcc dot gnu dot org  2010-04-29 13:49 -------
Created an attachment (id=20515)
 --> (http://gcc.gnu.org/bugzilla/attachment.cgi?id=20515&action=view)
142t.optimized dump


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=43931

[Bug fortran/43931] coarrays: Wrong code (segfault) with allocatable coarrays

[burnus at gcc dot gnu dot org]

------- Comment #2 from burnus at gcc dot gnu dot org  2010-04-29 13:49 -------
Created an attachment (id=20516)
 --> (http://gcc.gnu.org/bugzilla/attachment.cgi?id=20516&action=view)
Assembler (x86-64-linux)


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=43931

[Bug fortran/43931] coarrays: Wrong code (segfault) with allocatable coarrays

[burnus at gcc dot gnu dot org]

------- Comment #3 from burnus at gcc dot gnu dot org  2010-04-29 14:34 -------
The issue is that
  a.dim[]
is too small, which was traced by Alexander Monakov (thanks!).

This is in turn due to:
  gfc_get_array_descriptor_base (int dimen, int codimen, bool restricted)
  {
    int idx = 2 * (dimen - 1) + restricted;
    [...]
    gfc_array_descriptor_base[idx] = fat_type;

Possible patch (untested) below.

However, as coarrays do not need to agree on the coranks (except for
allocatable ones), coarrays with different corank are compatible. Ditto for
passing a coarray to a non-coarray. Thus, the patch below will cause problems
for
  integer :: coarray(1)[2,*]
  call foo(coarray)
to
  subroutine foo(coarray)
  integer :: coarray(1)[*]
or to
  subroutine foo(array)
  integer :: array(1)
as the coranks don't agree and thus not the type. I have to think about how to
solve this best. For allocatable coarrays there is no problem, as there the
coranks have to agree.


Index: trans-types.c
===================================================================
--- trans-types.c       (revision 158895)
+++ trans-types.c       (working copy)
@@ -1542,7 +1542,7 @@
 {
   tree fat_type, fieldlist, decl, arraytype;
   char name[16 + 2*GFC_RANK_DIGITS + 1 + 1];
-  int idx = 2 * (dimen - 1) + restricted;
+  int idx = 2 * (codimen + dimen - 1) + restricted;

   gcc_assert (dimen >= 1 && codimen + dimen <= GFC_MAX_DIMENSIONS);
   if (gfc_array_descriptor_base[idx])
@@ -1551,8 +1551,7 @@
   /* Build the type node.  */
   fat_type = make_node (RECORD_TYPE);

-  sprintf (name, "array_descriptor" GFC_RANK_PRINTF_FORMAT "_"
-          GFC_RANK_PRINTF_FORMAT, dimen, codimen);
+  sprintf (name, "array_descriptor" GFC_RANK_PRINTF_FORMAT, dimen + codimen);
   TYPE_NAME (fat_type) = get_identifier (name);

   /* Add the data member as the first element of the descriptor.  */
@@ -1629,8 +1628,7 @@
     type_name = IDENTIFIER_POINTER (tmp);
   else
     type_name = "unknown";
-  sprintf (name, "array" GFC_RANK_PRINTF_FORMAT "_"
-          GFC_RANK_PRINTF_FORMAT "_%.*s", dimen, codimen,
+  sprintf (name, "array" GFC_RANK_PRINTF_FORMAT "_%.*s", dimen + codimen,
           GFC_MAX_SYMBOL_LEN, type_name);
   TYPE_NAME (fat_type) = get_identifier (name);


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=43931

[Bug fortran/43931] coarrays: Wrong code (segfault) with allocatable coarrays

[burnus at gcc dot gnu dot org]

------- Comment #4 from burnus at gcc dot gnu dot org  2010-04-30 18:31 -------
Subject: Bug 43931

Author: burnus
Date: Fri Apr 30 18:30:53 2010
New Revision: 158941

URL: http://gcc.gnu.org/viewcvs?root=gcc&view=rev&rev=158941
Log:
2010-04-30  Tobias Burnus  Mburnus@net-b.de>

        PR fortran/18918
        PR fortran/43931
        *  trans-types.c (gfc_get_array_descriptor_base): Fix index
        calculation for array descriptor types.

2010-04-30  Tobias Burnus  Mburnus@net-b.de>

        PR fortran/18918
        PR fortran/43931
        * gfortran.dg/coarray_13.f90: New test.


Added:
    trunk/gcc/testsuite/gfortran.dg/coarray_13.f90
Modified:
    trunk/gcc/fortran/ChangeLog
    trunk/gcc/fortran/trans-types.c
    trunk/gcc/testsuite/ChangeLog


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=43931

[Bug fortran/43931] coarrays: Wrong code (segfault) with allocatable coarrays

[burnus at gcc dot gnu dot org]

------- Comment #5 from burnus at gcc dot gnu dot org  2010-04-30 18:36 -------
Close as FIXED.

With the other issues (passing as actual arguments) I will deal - but not as
part of this PR. See PR 18918 for the general coarray tracking.


-- 

burnus at gcc dot gnu dot org changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|UNCONFIRMED                 |RESOLVED
         Resolution|                            |FIXED


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=43931