[Bug fortran/53685] New: surprising warns about transfer with explicit character range

[Bug fortran/53685] New: surprising warns about transfer with explicit character range

[ajmay81 at googlemail dot com]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=53685

             Bug #: 53685
           Summary: surprising warns about transfer with explicit
                    character range
    Classification: Unclassified
           Product: gcc
           Version: 4.7.1
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: fortran
        AssignedTo: unassigned@gcc.gnu.org
        ReportedBy: ajmay81@googlemail.com


Fortran code:

      subroutine test()
      implicit none
      character(len=4) :: record_type
      integer          :: i
      i=transfer(record_type,i) ! no warning
      i=transfer(record_type(1:4),i) ! warning
      return
      end

gfortran -c -Wsurprising test.f
test.f:6.17:

      i=transfer(record_type(1:4),i) ! warning                          
                 1
Warning: Intrinsic TRANSFER at (1) has partly undefined result: source size 0 <
result size 4

When the string length is explicitly given the compiler thinks it is length 0,
even though it is the same length as the previous instance.

Seen with 4.7.1 built from source.

[Bug fortran/53685] surprising warns about transfer with explicit character range

[burnus at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=53685

Tobias Burnus <burnus at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
           Keywords|                            |diagnostic
             Status|UNCONFIRMED                 |NEW
   Last reconfirmed|                            |2012-06-15
                 CC|                            |burnus at gcc dot gnu.org
     Ever Confirmed|0                           |1

--- Comment #1 from Tobias Burnus <burnus at gcc dot gnu.org> 2012-06-15 16:18:36 UTC ---
Confirmed.

The size is determined via target-memory.c's gfc_target_expr_size. There seems
to be two issues:

(a) A minor one that in the second case, the length of "record_type" is
"e->ts->u.cl->length == NULL" instead of 4 (as one could be simply calculated).

b) And the missing error handling for gfc_target_expr_size in
gfc_calculate_transfer_sizes: There, the result size is error checked, namely:
  if (result_elt_size == 0)
    return FAILURE;
But a similar line for source_size is lacking.


Regarding (a): The question is whether one shouldn't set expr->ts.u.cl->length
to expr->ref->next->...->u.ss.length for substrings. I don't know whether that
messes with the CL cleanup or other issues, but it'd make handling substrings
easier than checking whether there is a last_ref(expr)->u.ss.length.


Patch for the second issue.

--- a/gcc/fortran/check.c
+++ b/gcc/fortran/check.c
@@ -4003,2 +4003,4 @@ gfc_calculate_transfer_sizes (gfc_expr *source, gfc_expr
*mold, gfc_expr *size,
   *source_size = gfc_target_expr_size (source);
+  if (source_size == 0)
+    return FAILURE;

[Bug fortran/53685] surprising warns about transfer with explicit character range

[burnus at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=53685

--- Comment #2 from Tobias Burnus <burnus at gcc dot gnu.org> 2012-06-15 17:02:44 UTC ---
(In reply to comment #1)
> (a) A minor one that in the second case, the length of "record_type" is
> "e->ts->u.cl->length == NULL" instead of 4 (as one could be simply calculated).

That should be fixed by the following patch (untested). The question is why
e->ref->type == REF_SUBSTRING was excluded if (and only if) it was the first
reference?

The whole condition plus function call was added 2007-08-31 in the big patch
for PR fortran/31879, PR fortran/31197, PR fortran/31258 and PR fortran/32703:
http://gcc.gnu.org/viewcvs?view=revision&revision=127939

--- a/gcc/fortran/resolve.c
+++ b/gcc/fortran/resolve.c
@@ -6325,4 +6325,3 @@ gfc_resolve_expr (gfc_expr *e)

-      if (e->ts.type == BT_CHARACTER && e->ts.u.cl == NULL && e->ref
-         && e->ref->type != REF_SUBSTRING)
+      if (e->ts.type == BT_CHARACTER && e->ts.u.cl == NULL && e->ref)
        gfc_resolve_substring_charlen (e);

[Bug fortran/53685] surprising warns about transfer with explicit character range

[burnus at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=53685

--- Comment #3 from Tobias Burnus <burnus at gcc dot gnu.org> 2012-06-16 10:50:48 UTC ---
That patch of comment 2 fails for:

testsuite/gfortran.dg/string_3.f90:16.15:
  print *, len(s(2_8**32_8+3_8:1))
               1
Error: Result of LEN overflows its kind at (1)

It also fails for "if (len(t(8:4)) /= 0) call abort" in gfortran.dg/substr_2.f.
And for gfortran.dg/widechar_IO_4.f90 (rt, line 16),
gfortran.dg/widechar_compare_1.f90 (line 9), and
gfortran.dg/widechar_intrinsics_8.f90 (line 24).

[Bug fortran/53685] surprising warns about transfer with explicit character range

[burnus at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=53685

--- Comment #4 from Tobias Burnus <burnus at gcc dot gnu.org> 2012-06-17 19:30:44 UTC ---
Author: burnus
Date: Sun Jun 17 19:30:29 2012
New Revision: 188708

URL: http://gcc.gnu.org/viewcvs?root=gcc&view=rev&rev=188708
Log:
2012-06-17  Tobias Burnus  <burnus@net-b.de>

        PR fortran/53691
        PR fortran/53685
        * check.c (gfc_calculate_transfer_sizes): Return if
        SIZE= is not constant or source-size cannot be determined.

2012-06-17  Tobias Burnus  <burnus@net-b.de>

        PR fortran/53691
        PR fortran/53685
        * gfortran.dg/transfer_check_3.f90: New.


Added:
    trunk/gcc/testsuite/gfortran.dg/transfer_check_3.f90
Modified:
    trunk/gcc/fortran/ChangeLog
    trunk/gcc/fortran/check.c
    trunk/gcc/testsuite/ChangeLog

[Bug fortran/53685] surprising warns about transfer with explicit character range

[burnus at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=53685

--- Comment #5 from Tobias Burnus <burnus at gcc dot gnu.org> 2012-06-22 10:28:01 UTC ---
Author: burnus
Date: Fri Jun 22 10:27:56 2012
New Revision: 188882

URL: http://gcc.gnu.org/viewcvs?root=gcc&view=rev&rev=188882
Log:
2012-06-22  Tobias Burnus  <burnus@net-b.de>

        Backport from mainline
        2012-06-17  Tobias Burnus  <burnus@net-b.de>

        PR fortran/53691
        PR fortran/53685
        * check.c (gfc_calculate_transfer_sizes): Return if
        SIZE= is not constant or source-size cannot be determined.

2012-06-22  Tobias Burnus  <burnus@net-b.de>

        Backport from mainline
        2012-06-17  Tobias Burnus  <burnus@net-b.de>

        PR fortran/53691
        PR fortran/53685
        * gfortran.dg/transfer_check_3.f90: New.


Added:
    branches/gcc-4_7-branch/gcc/testsuite/gfortran.dg/transfer_check_3.f90
Modified:
    branches/gcc-4_7-branch/gcc/fortran/ChangeLog
    branches/gcc-4_7-branch/gcc/fortran/check.c
    branches/gcc-4_7-branch/gcc/testsuite/ChangeLog

[Bug fortran/53685] surprising warns about transfer with explicit character range

[ajmay81 at googlemail dot com]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=53685

--- Comment #6 from Andy May <ajmay81 at googlemail dot com> 2012-09-24 13:57:37 UTC ---
Thanks for fixing this, the GCC 4.7.1 shipping with openSUSE 12.2 does not show
this problem, and a build of GCC 4.7.2 from source doesn't either. Providing
this has also been pushed to trunk and is not causing problems, I think this
bug can be closed.

[Bug fortran/53685] surprising warns about transfer with explicit character range

[kloedej at knmi dot nl]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=53685

Jos de Kloe <kloedej at knmi dot nl> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |kloedej at knmi dot nl

--- Comment #7 from Jos de Kloe <kloedej at knmi dot nl> 2012-11-29 10:52:32 UTC ---
I see a very similar problem with a slightly modified piece of testcode on
version 4.7.2. Not sure if it is exactly the same, but probably related to the
already patched issue. My testcode:

program testtransfer

  integer, parameter :: r8_ = Selected_Real_Kind(15,307)  ! = real*8
  integer, parameter :: nbytes_r8_ = 8
  real(r8_) :: val
  character(len=1), dimension(16) :: byte_array

  byte_array(:) = ' '
  call mytest(byte_array,val)
  print *,"val: ",val
contains
  subroutine mytest(byte_array,val)
    character(len=1), dimension(16), intent(in) :: byte_array
    real(r8_),intent(out) :: val
    val = transfer(byte_array(1:nbytes_r8_),val)    
  end subroutine mytest
end program testtransfer

Results:

>gfortran -Wall  -o transfer_problem transfer_problem.F90
transfer_problem.F90:15.19:

    val = transfer(byte_array(1:nbytes_r8_),val)    
                   1
Warning: Intrinsic TRANSFER at (1) has partly undefined result: source size 1 <
result size 8
>

This occurs for this version:
>gfortran --version | head -1
GNU Fortran (GCC) 4.7.2 20120921 (Red Hat 4.7.2-2)
on my Fedora17 system. (package gcc-gfortran-4.7.2-2.fc17.x86_64)

The problem is not present in:
>gfortran --version | head -1
GNU Fortran (GCC) 4.6.3 20120306 (Red Hat 4.6.3-2)
on a Fedora16 system. (package gcc-gfortran-4.6.3-2.fc16.x86_64)

[Bug fortran/53685] surprising warns about transfer with explicit character range

[janus at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=53685

--- Comment #8 from janus at gcc dot gnu.org 2013-04-22 19:23:37 UTC ---
Comment 7 is fixed on 4.9 trunk with the following commit:

Author: janus
Date: Mon Apr 22 19:14:22 2013
New Revision: 198155

URL: http://gcc.gnu.org/viewcvs?rev=198155&root=gcc&view=rev
Log:
2013-04-22  Janus Weil  <janus@gcc.gnu.org>

    PR fortran/53685
    PR fortran/57022
    * check.c (gfc_calculate_transfer_sizes): Fix for array-valued SOURCE
    expressions.
    * simplify.c (gfc_simplify_sizeof,gfc_simplify_storage_size): Get rid
    of special treatment for EXPR_ARRAY.
    * target-memory.h (gfc_element_size): New prototype.
    * target-memory.c (size_array): Remove.
    (gfc_element_size): New function.
    (gfc_target_expr_size): Modified to always return the full size of the
    expression.


2013-04-22  Janus Weil  <janus@gcc.gnu.org>

    PR fortran/53685
    PR fortran/57022
    * gfortran.dg/transfer_check_4.f90: New.

Added:
    trunk/gcc/testsuite/gfortran.dg/transfer_check_4.f90
Modified:
    trunk/gcc/fortran/ChangeLog
    trunk/gcc/fortran/check.c
    trunk/gcc/fortran/simplify.c
    trunk/gcc/fortran/target-memory.c
    trunk/gcc/fortran/target-memory.h
    trunk/gcc/testsuite/ChangeLog

[Bug fortran/53685] surprising warns about transfer with explicit character range

[janus at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=53685

janus at gcc dot gnu.org changed:

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

--- Comment #9 from janus at gcc dot gnu.org 2013-04-26 22:50:35 UTC ---
(In reply to comment #8)
> Comment 7 is fixed on 4.9 trunk with the following commit:

which has also been backported to 4.8 and 4.7:

http://gcc.gnu.org/viewcvs?rev=198345&root=gcc&view=rev
http://gcc.gnu.org/viewcvs?rev=198348&root=gcc&view=rev


I think we can close this bug.