[Bug fortran/59906] New: [4.7/4.8 Regression] error: size of variable '<anonymous>' is too large

[Bug fortran/59906] New: [4.7/4.8 Regression] error: size of variable '<anonymous>' is too large

[anlauf at gmx dot de]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=59906

            Bug ID: 59906
           Summary: [4.7/4.8 Regression] error: size of variable
                    '<anonymous>' is too large
           Product: gcc
           Version: 4.8.2
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: fortran
          Assignee: unassigned at gcc dot gnu.org
          Reporter: anlauf at gmx dot de

Hi,

a colleague just reported a regression for the code below.
It compiles fine with 4.3.4 and 4.5.0, but fails with 4.7.4 and 4.8.2.
Cannot check trunk right now.

% gfortran48 robin_gfortran_bug.f90
robin_gfortran_bug.f90: In function 'y':
robin_gfortran_bug.f90:17:0: error: size of variable '<anonymous>' is too large
     print*,str_cmp((/'aaa','bbb'/), str_aux)
 ^

% gfortran47 robin_gfortran_bug.f90      
robin_gfortran_bug.f90: In function 'y':
robin_gfortran_bug.f90:17:0: error: size of variable '<anonymous>' is too large


% cat robin_gfortran_bug.f90
program x
  implicit none
  call y('abcdef')
contains 

  subroutine y(str)
    character(len=*), intent(in) :: str

    character(len=len_trim(str)) :: str_aux

    str_aux = str

    ! Compiles
!   print*,str_cmp((/'aaa','bbb'/), str)

    ! Does not compile:
    print*,str_cmp((/'aaa','bbb'/), str_aux)

  end subroutine y

  elemental logical function str_cmp(str1, str2)
    character(len=*), intent(in) :: str1
    character(len=*), intent(in) :: str2
    str_cmp = (str1 == str2)
  end function str_cmp

end program x

[Bug fortran/59906] [4.7/4.8 Regression] error: size of variable '<anonymous>' is too large

[anlauf at gmx dot de]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=59906

Harald Anlauf <anlauf at gmx dot de> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
      Known to work|                            |4.1.2
      Known to fail|                            |4.4.5, 4.5.2, 4.6.3

--- Comment #1 from Harald Anlauf <anlauf at gmx dot de> ---
It's not entirely clear when the regression occured.

I see no failure with

GNU Fortran (SUSE Linux) 4.5.0 20100604 [gcc-4_5-branch revision 160292]

but the failure occurs with 4.4.5 and 4.5.2 on a Debian based system.

[Bug fortran/59906] [4.7/4.8 Regression] error: size of variable '<anonymous>' is too large

[hjl.tools at gmail dot com]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=59906

H.J. Lu <hjl.tools at gmail dot com> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|UNCONFIRMED                 |NEW
   Last reconfirmed|                            |2014-01-22
            Version|4.8.2                       |4.7.3
   Target Milestone|---                         |4.7.4
     Ever confirmed|0                           |1

--- Comment #2 from H.J. Lu <hjl.tools at gmail dot com> ---
It is caused by r158683.

[Bug fortran/59906] [4.7/4.8/4.9 Regression] error: size of variable '<anonymous>' is too large

[anlauf at gmx dot de]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=59906

--- Comment #3 from Harald Anlauf <anlauf at gmx dot de> ---
I ran the testcase under gdb to see where the error is thrown.

Starting program: /opt/gcc/4.9/lib/gcc/i686-pc-linux-gnu/4.9.0/f951
robin_gfortran_bug.f90 -quiet -dumpbase robin_gfortran_bug.f90 -mtune=generic
-march=pentiumpro -auxbase robin_gfortran_bug -o /tmp/ccIAIgJb.s

Breakpoint 2, assign_temp (type_or_decl=0xb7951f00, memory_required=0, 
    dont_promote=1) at ../../trunk/gcc/function.c:931
931       if (DECL_P (type_or_decl))
(gdb) bt
#0  assign_temp (type_or_decl=0xb7951f00, memory_required=0, dont_promote=1)
    at ../../trunk/gcc/function.c:931
#1  0x0832a884 in expand_return (retval=<optimized out>)
    at ../../trunk/gcc/cfgexpand.c:3110
#2  expand_gimple_stmt_1 (stmt=0xb78cf268) at ../../trunk/gcc/cfgexpand.c:3187
#3  expand_gimple_stmt (stmt=0xb78cf268) at ../../trunk/gcc/cfgexpand.c:3309
#4  0x0832c7bf in expand_gimple_basic_block (bb=0xb79577c0, 
    disable_tail_calls=false) at ../../trunk/gcc/cfgexpand.c:5149
#5  0x0833092e in gimple_expand_cfg () at ../../trunk/gcc/cfgexpand.c:5715
#6  (anonymous namespace)::pass_expand::execute (this=0x9190488)
    at ../../trunk/gcc/cfgexpand.c:5935
#7  0x085ffcf6 in execute_one_pass (pass=0x9190488)
    at ../../trunk/gcc/passes.c:2229
#8  0x08600026 in execute_pass_list (pass=0x9190488)
    at ../../trunk/gcc/passes.c:2282
#9  0x08356b8a in expand_function (node=0xb789198c)
    at ../../trunk/gcc/cgraphunit.c:1764
#10 0x08358ff5 in output_in_order () at ../../trunk/gcc/cgraphunit.c:1996
#11 compile () at ../../trunk/gcc/cgraphunit.c:2237
#12 0x08359575 in finalize_compilation_unit ()
    at ../../trunk/gcc/cgraphunit.c:2319
#13 0x0857056b in write_global_declarations ()
    at ../../trunk/gcc/langhooks.c:323
#14 0x086b29ad in compile_file () at ../../trunk/gcc/toplev.c:562
#15 0x086b490c in do_compile () at ../../trunk/gcc/toplev.c:1914
#16 toplev_main (argc=11, argv=0xbfffeb84) at ../../trunk/gcc/toplev.c:1990
#17 0x08195bfb in main (argc=11, argv=0xbfffeb84) at ../../trunk/gcc/main.c:36

It appears that the error is generated rather late.  Are there any
intermediate files it makes sense to look at, to see the effect of
slight variations of the input?  (E.g. where the "anonymous namespace"
comes from.)

[Bug fortran/59906] [4.7/4.8/4.9 Regression] error: size of variable '<anonymous>' is too large

[pault at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=59906

Paul Thomas <pault at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
           Assignee|unassigned at gcc dot gnu.org      |pault at gcc dot gnu.org

--- Comment #4 from Paul Thomas <pault at gcc dot gnu.org> ---
(In reply to Harald Anlauf from comment #3)
> I ran the testcase under gdb to see where the error is thrown.
> 
....snip....

> It appears that the error is generated rather late.  Are there any
> intermediate files it makes sense to look at, to see the effect of
> slight variations of the input?  (E.g. where the "anonymous namespace"
> comes from.)

It would be very useful if you would try to compile it with
-ftree-dump-original and post the result around the scalarization loop for 
    ! Does not compile:
    print*,str_cmp((/'aaa','bbb'/), str_aux)

In order to isolate all the code associated with a particular line, I cast it
into a contained subroutine which is then called instead of the offending line.

(I am not in a position to do it myself until tonight.)

HJ is certainly correct that r158683 is the cause - Thanks for the heads-up,
HJ.  It amazes me that this bug has not been triggered in 3 years and 9 months!

Cheers

Paul

[Bug fortran/59906] [4.7/4.8/4.9 Regression] error: size of variable '<anonymous>' is too large

[rguenth at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=59906

Richard Biener <rguenth at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
           Priority|P3                          |P4

[Bug fortran/59906] [4.7/4.8/4.9 Regression] error: size of variable '<anonymous>' is too large

[anlauf at gmx dot de]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=59906

--- Comment #5 from Harald Anlauf <anlauf at gmx dot de> ---
Created attachment 32004
  --> http://gcc.gnu.org/bugzilla/attachment.cgi?id=32004&action=edit
-fdump-tree-original of failing variant

[Bug fortran/59906] [4.7/4.8/4.9 Regression] error: size of variable '<anonymous>' is too large

[anlauf at gmx dot de]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=59906

--- Comment #6 from Harald Anlauf <anlauf at gmx dot de> ---
Created attachment 32005
  --> http://gcc.gnu.org/bugzilla/attachment.cgi?id=32005&action=edit
-fdump-tree-original of working variant

[Bug fortran/59906] [4.7/4.8/4.9 Regression] error: size of variable '<anonymous>' is too large

[anlauf at gmx dot de]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=59906

--- Comment #7 from Harald Anlauf <anlauf at gmx dot de> ---
(In reply to Paul Thomas from comment #4)
> It would be very useful if you would try to compile it with
> -ftree-dump-original and post the result around the scalarization loop for 
>     ! Does not compile:
>     print*,str_cmp((/'aaa','bbb'/), str_aux)

The diff between compiling and failing variant is:

--- robin_gfortran_bug.f90.003t.original.compiles       2014-01-31
18:44:56.000000000 +0100
+++ robin_gfortran_bug.f90.003t.original.fails  2014-01-31 18:44:32.000000000
+0100
@@ -55,15 +55,15 @@
     struct __st_parameter_dt dt_parm.0;

     dt_parm.0.common.filename = &"robin_gfortran_bug.f90"[1]{lb: 1 sz: 1};
-    dt_parm.0.common.line = 14;
+    dt_parm.0.common.line = 17;
     dt_parm.0.common.flags = 128;
     dt_parm.0.common.unit = 6;
     _gfortran_st_write (&dt_parm.0);
     {
-      character(kind=1)[1:_str] & restrict D.2272;
+      character(kind=1) D.2272[1:.str_aux];
       static character(kind=1)[1:3] * A.1[2] = {&"aaa"[1]{lb: 1 sz: 1},
&"bbb"[1]{lb: 1 sz: 1}};

-      D.2272 = str;
+      D.2272 = str_aux;
       {
         integer(kind=4) S.2;

@@ -74,7 +74,7 @@
             {
               logical(kind=4) D.2274;

-              D.2274 = str_cmp (A.1[S.2], &D.2272, 3, _str);
+              D.2274 = str_cmp (A.1[S.2], &D.2272, 3, .str_aux);
               _gfortran_transfer_logical_write (&dt_parm.0, &D.2274, 4);
             }
             S.2 = S.2 + 1;

[Bug fortran/59906] [4.7/4.8/4.9 Regression] error: size of variable '<anonymous>' is too large

[pault at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=59906

--- Comment #8 from Paul Thomas <pault at gcc dot gnu.org> ---
Created attachment 32010
  --> http://gcc.gnu.org/bugzilla/attachment.cgi?id=32010&action=edit
Tentative patch for the PR

The line that compiled did not yield correct code.  The testcase in the patch
runs correctly.

The first regtest failed in elemental_scalar_args_2.f90.  To fix this, the
exclusion of functions in the patched part of trans-expr.c had to be added. 
This surprises me because it looks to me as if gfc_conv_string_parameter should
do the job itself. I will investigate.

In the meantime, a second regtest is underway.

Cheers

Paul

[Bug fortran/59906] [4.7/4.8/4.9 Regression] error: size of variable '<anonymous>' is too large

[pault at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=59906

--- Comment #9 from Paul Thomas <pault at gcc dot gnu.org> ---
Author: pault
Date: Sat Feb  1 18:50:41 2014
New Revision: 207389

URL: http://gcc.gnu.org/viewcvs?rev=207389&root=gcc&view=rev
Log:
2014-02-01  Paul Thomas  <pault@gcc.gnu.org>

    PR fortran/59906
    * trans-stmt.c (gfc_add_loop_ss_code): In the case of character
    SS_REFERENCE, use gfc_conv_string_parameter to ensure that a
    pointer to the string is stored.
    * trans-expr.c (gfc_conv_expr_reference): Likewise, use
    gfc_conv_string_parameter to ensure that a pointer to is passed
    to the elemental function.

2014-02-01  Paul Thomas  <pault@gcc.gnu.org>

    PR fortran/59906
    * gfortran.dg/elemental_subroutine_9.f90 : New test

Added:
    trunk/gcc/testsuite/gfortran.dg/elemental_subroutine_9.f90
Modified:
    trunk/gcc/fortran/ChangeLog
    trunk/gcc/fortran/trans-array.c
    trunk/gcc/fortran/trans-expr.c
    trunk/gcc/testsuite/ChangeLog

[Bug fortran/59906] [4.7/4.8 Regression] error: size of variable '<anonymous>' is too large

[dominiq at lps dot ens.fr]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=59906

Dominique d'Humieres <dominiq at lps dot ens.fr> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |fmartinez at gmv dot com

--- Comment #10 from Dominique d'Humieres <dominiq at lps dot ens.fr> ---
*** Bug 60066 has been marked as a duplicate of this bug. ***

[Bug fortran/59906] [4.7/4.8 Regression] error: size of variable '<anonymous>' is too large

[pault at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=59906

--- Comment #11 from Paul Thomas <pault at gcc dot gnu.org> ---
Author: pault
Date: Fri Feb  7 21:15:37 2014
New Revision: 207613

URL: http://gcc.gnu.org/viewcvs?rev=207613&root=gcc&view=rev
Log:
2014-02-07  Paul Thomas  <pault@gcc.gnu.org>

    PR fortran/59906
    * trans-stmt.c (gfc_add_loop_ss_code): In the case of character
    SS_REFERENCE, use gfc_conv_string_parameter to ensure that a
    pointer to the string is stored.
    * trans-expr.c (gfc_conv_expr_reference): Likewise, use
    gfc_conv_string_parameter to ensure that a pointer to is passed
    to the elemental function.

2014-02-07  Paul Thomas  <pault@gcc.gnu.org>

    PR fortran/59906
    * gfortran.dg/elemental_subroutine_9.f90 : New test

Added:
   
branches/gcc-4_8-branch/gcc/testsuite/gfortran.dg/elemental_subroutine_9.f90
Modified:
    branches/gcc-4_8-branch/gcc/fortran/ChangeLog
    branches/gcc-4_8-branch/gcc/fortran/trans-array.c
    branches/gcc-4_8-branch/gcc/fortran/trans-expr.c
    branches/gcc-4_8-branch/gcc/testsuite/ChangeLog

[Bug fortran/59906] [4.7/4.8 Regression] error: size of variable '<anonymous>' is too large

[pault at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=59906

--- Comment #12 from Paul Thomas <pault at gcc dot gnu.org> ---
Author: pault
Date: Fri Feb  7 23:29:44 2014
New Revision: 207617

URL: http://gcc.gnu.org/viewcvs?rev=207617&root=gcc&view=rev
Log:
2014-02-08  Paul Thomas  <pault@gcc.gnu.org>

    PR fortran/59906
    * trans-array.c (gfc_add_loop_ss_code): In the case of character
    SS_REFERENCE, use gfc_conv_string_parameter to ensure that a
    pointer to the string is stored.
    * trans-expr.c (gfc_conv_expr_reference): Likewise, use
    gfc_conv_string_parameter to ensure that a pointer to is passed
    to the elemental function.

2014-02-08  Paul Thomas  <pault@gcc.gnu.org>

    PR fortran/59906
    * gfortran.dg/elemental_subroutine_9.f90 : New test

Added:
   
branches/gcc-4_7-branch/gcc/testsuite/gfortran.dg/elemental_subroutine_9.f90
Modified:
    branches/gcc-4_7-branch/gcc/fortran/ChangeLog
    branches/gcc-4_7-branch/gcc/fortran/trans-array.c
    branches/gcc-4_7-branch/gcc/fortran/trans-expr.c
    branches/gcc-4_7-branch/gcc/testsuite/ChangeLog

[Bug fortran/59906] [4.7/4.8 Regression] error: size of variable '<anonymous>' is too large

[pault at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=59906

Paul Thomas <pault at gcc dot gnu.org> changed:

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

--- Comment #13 from Paul Thomas <pault at gcc dot gnu.org> ---
Fixed on 4.7, 4.8 and 4.9

Thanks for the report

Paul