[Bug fortran/36257] New: SPREAD gives wrong results with CHARACTER variables

[Bug fortran/36257] New: SPREAD gives wrong results with CHARACTER variables

[fxcoudert at gcc dot gnu dot org]

This one turned up during my testing of SPREAD on wide character strings, but
it turns out that it already happens on default character variables. It looks
like the front-end doesn't fill the return array descriptor well.

$ cat k2.f90
  implicit none
  character(len=5), dimension(3,3), parameter :: &
    p = reshape(["", "", "", "", "", "", "", "", ""], [3,3])
  character(len=5), dimension(3,3) :: m1

  m1 = p
  if (any (spread (p, 1, 2) /= spread (m1, 1, 2))) call abort

end
$ gfortran k2.f90 && ./a.out 
Fortran runtime error: rank mismatch in spread()

The most amazing is that it's specific to character variables (ie it doesn't
happen with integer, for example).


-- 
           Summary: SPREAD gives wrong results with CHARACTER variables
           Product: gcc
           Version: 4.4.0
            Status: UNCONFIRMED
          Keywords: wrong-code
          Severity: normal
          Priority: P3
         Component: fortran
        AssignedTo: unassigned at gcc dot gnu dot org
        ReportedBy: fxcoudert at gcc dot gnu dot org


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=36257

[Bug fortran/36257] SPREAD gives wrong results with CHARACTER variables

[pault at gcc dot gnu dot org]

------- Comment #1 from pault at gcc dot gnu dot org  2008-05-18 10:03 -------
Confirmed.

This is quite revealing...

  implicit none
  character(len=5), dimension(3,3), parameter :: &
    p = reshape(["a", "b", "c", "d", "e", "f", "g", "h", "i"], [3,3])
  character(len=5), dimension(3,3) :: m1

  m1 = p

  print *, shape (spread (p, 1, 2))
  print *, spread (p, 1, 2)
  print *, spread (p, 1, 2)//"z"
!  print *, spread (p, 1, 2) /= spread (m1, 1, 2)
!  if (any (spread (p, 1, 2) /= spread (m1, 1, 2))) call abort

end

The first two prints are OK but the third is completely screwed up.  The
front-end is not providing the right shape, to judge by the array descriptors
provided for the temporaries.  BTW characters are treated separately in the
library - they all go to spread_internal.

Thanks for that one! Maybe Thomas can pinpoint the problem right away?

Cheers

Paul 


-- 

pault at gcc dot gnu dot org changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|UNCONFIRMED                 |NEW
     Ever Confirmed|0                           |1
   Last reconfirmed|0000-00-00 00:00:00         |2008-05-18 10:03:00
               date|                            |


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=36257

[Bug fortran/36257] SPREAD gives wrong results with array CHARACTER parameters

[tkoenig at gcc dot gnu dot org]

[-- Warning: decoded text below may be mangled, UTF-8 assumed --]
[-- Attachment #1: Type: text/plain, Size: 2948 bytes --]



------- Comment #2 from tkoenig at gcc dot gnu dot org  2008-05-18 10:42 -------
The problem is with the array parameter.

$ cat spread.f90
  implicit none
  character(len=5), dimension(3,3), parameter :: &
    p = reshape(["a", "b", "c", "d", "e", "f", "g", "h", "i"], [3,3])
  character(len=5), dimension(3,3) :: m1
  m1 = p
  print '(18A)', spread (p, 1, 2)//"z"
  print '(18A)', spread (m1, 1, 2)//"z"
end
$ gfortran -fdump-tree-original -g spread.f90
$ gdb ./a.out
Breakpoint 1 (_gfortran_spread_char) pending.
(gdb) r
Starting program: /tmp/a.out
Breakpoint 2 at 0xb7f69f13: file
../../../../gcc/trunk/libgfortran/intrinsics/spread_generic.c, line 421.
Pending breakpoint "_gfortran_spread_char" resolved

Breakpoint 2, *_gfortran_spread_char (ret=0xbfe7d7b8, ret_length=5,
source=0xbfe7d80c,
    along=0x8048dd4, pncopies=0x8048dd0, source_length=5)
    at ../../../../gcc/trunk/libgfortran/intrinsics/spread_generic.c:422
422       spread_internal (ret, source, along, pncopies, source_length);
(gdb) p *source
$1 = {data = 0xbfe7d827 "a    b    c    ç¿\001", offset = 0, dtype = 369, dim =
{{
      stride = 1, lbound = 0, ubound = 2}, {stride = 1627389952, lbound =
538976288,
      ubound = 538976354}, {stride = 538993440, lbound = -1075372000, ubound =
1}, {
      stride = 0, lbound = -1075324808, ubound = -1208419248}, {stride =
-1075324752,
      lbound = -1209229152, ubound = -1075317768}, {stride = -1075324652,
      lbound = -1075324808, ubound = 134515769}, {stride = 1, lbound =
-1075324652,
      ubound = -1075324792}}}

This is seriously flawed.  The data pointer is wrong, as is the dtype.

Using m1 is ok:

(gdb) c
Continuing.
a    za    zb    zb    zc    zc    z

Breakpoint 2, *_gfortran_spread_char (ret=0xbfe7d758, ret_length=5,
source=0xbfe7d7b8,
    along=0x8048dd4, pncopies=0x8048dd0, source_length=5)
    at ../../../../gcc/trunk/libgfortran/intrinsics/spread_generic.c:422
422       spread_internal (ret, source, along, pncopies, source_length);
(gdb) p *source
$2 = {data = 0xbfe7d78b "a    b    c    d    e    f    g    h    i   
\213×ç¿üÿÿÿr\001",
  offset = 4294967292, dtype = 370, dim = {{stride = 1, lbound = 1, ubound =
3}, {
      stride = 3, lbound = 1, ubound = 3}, {stride = -1208504320, lbound =
543414304,
      ubound = 2048925728}, {stride = -1075324928, lbound = 0, ubound = 0},
{stride = 1,
      lbound = 0, ubound = 2}, {stride = 134516032, lbound = 134516037,
      ubound = 134516042}, {stride = -1075324889, lbound = 0, ubound = 369}}}


-- 

tkoenig at gcc dot gnu dot org changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
            Summary|SPREAD gives wrong results  |SPREAD gives wrong results
                   |with CHARACTER variables    |with array CHARACTER
                   |                            |parameters


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=36257

[Bug fortran/36257] SPREAD gives wrong results with array CHARACTER parameters

[fxcoudert at gcc dot gnu dot org]

------- Comment #3 from fxcoudert at gcc dot gnu dot org  2008-05-18 14:55 -------
I have some more information, but it still doesn't make sense. I'm comparing
the codepaths taken in the front-end by:

  implicit none
  character(len=5), dimension(3,3), parameter :: &
    p = reshape(["a", "b", "c", "d", "e", "f", "g", "h", "i"], [3,3])
  character(len=5), dimension(3,3) :: m1
  m1 = p
  print '(18A)', spread (m1, 1, 2)//"z"
  end

and:

  implicit none
  character(len=5), dimension(3,3), parameter :: &
    p = reshape(["a", "b", "c", "d", "e", "f", "g", "h", "i"], [3,3])
  print '(18A)', spread (p, 1, 2)//"z"
  end

In both cases, we end up twice in gfc_resolve_spread, but while for a
non-parameter array, we always give the return type its correct rank (f->rank =
3), for the paramater array, we give it a rank of 2 the second time. That's
because in  check_charlen_present(), the source->rank gets modified from 2 to
1). I don't understand why we do this (someone assumed that we can only have
array parameters with rank 1?)

So, commenting that piece of code there:

Index: iresolve.c
===================================================================
--- iresolve.c  (revision 135395)
+++ iresolve.c  (working copy)
@@ -78,7 +78,7 @@ check_charlen_present (gfc_expr *source)
     {
       source->ts.cl->length =
        gfc_int_expr (source->value.constructor->expr->value.character.length);
-      source->rank = 1;
+      /*source->rank = 1;*/
     }
 }

make that testcase work, but I wonder what else it breaks :)


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=36257

[Bug fortran/36257] SPREAD gives wrong results with array CHARACTER parameters

[fxcoudert at gcc dot gnu dot org]

------- Comment #4 from fxcoudert at gcc dot gnu dot org  2008-05-20 13:33 -------
(In reply to comment #3)
> I wonder what else it breaks :)

Apparently, nothing. It was apparently introduced by Paul and I during the
efforts to fix PR31608, and is probably just copied from the lines above.
Anyway, the testsuite regtests fine on x86_64-linux with it, and CP2K still
compiles. As I don't see a logic for that line, let's just remove it!


-- 

fxcoudert at gcc dot gnu dot org changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
         AssignedTo|unassigned at gcc dot gnu   |fxcoudert at gcc dot gnu dot
                   |dot org                     |org
             Status|NEW                         |ASSIGNED
   Last reconfirmed|2008-05-18 10:03:00         |2008-05-20 13:33:47
               date|                            |


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=36257

[Bug fortran/36257] SPREAD gives wrong results with array CHARACTER parameters

[fxcoudert at gcc dot gnu dot org]

------- Comment #5 from fxcoudert at gcc dot gnu dot org  2008-05-21 16:02 -------
Subject: Bug 36257

Author: fxcoudert
Date: Wed May 21 16:01:08 2008
New Revision: 135723

URL: http://gcc.gnu.org/viewcvs?root=gcc&view=rev&rev=135723
Log:
        PR fortran/36257

        * gfortran.dg/char_length_12.f90: New test.

        * iresolve.c (check_charlen_present): Don't force the rank to 1.

Added:
    trunk/gcc/testsuite/gfortran.dg/char_length_12.f90
Modified:
    trunk/gcc/fortran/ChangeLog
    trunk/gcc/fortran/iresolve.c
    trunk/gcc/testsuite/ChangeLog


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=36257

[Bug fortran/36257] SPREAD gives wrong results with array CHARACTER parameters

[fxcoudert at gcc dot gnu dot org]

------- Comment #6 from fxcoudert at gcc dot gnu dot org  2008-05-21 16:12 -------
Fixed.


-- 

fxcoudert at gcc dot gnu dot org changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|ASSIGNED                    |RESOLVED
         Resolution|                            |FIXED
   Target Milestone|---                         |4.4.0


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=36257