[Bug fortran/66193] New: ICE for initialisation of some non-zero-sized arrays

[Bug fortran/66193] New: ICE for initialisation of some non-zero-sized arrays

[gerhard.steinmetz.fortran@t-online.de]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=66193

            Bug ID: 66193
           Summary: ICE for initialisation of some non-zero-sized arrays
           Product: gcc
           Version: 5.1.1
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: fortran
          Assignee: unassigned at gcc dot gnu.org
          Reporter: gerhard.steinmetz.fortran@t-online.de
  Target Milestone: ---

This initialisation of an array (NOT zero sized, NOT a parameter)
   program p
      real :: z(2)
      z = 1 + [real :: 1, 2]
   end

yields (with gfortran 5.1.1 on SUSE Linux 13.2, 64 bit)
f951: internal compiler error: Segmentation fault


This variation ...
   program p
      real :: z(2) = 1 + [real :: 1, 2]
   end

or with an additional parameter attribute ...
   program p
      real, parameter :: z(2) = 1 + [real :: 1, 2]
   end

yields the same error message as above.
Kind regards.

[Bug fortran/66193] ICE for initialisation of some non-zero-sized arrays

[gerhard.steinmetz.fortran@t-online.de]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=66193

--- Comment #1 from Gerhard Steinmetz <gerhard.steinmetz.fortran@t-online.de> ---
For integer instead of real ...
   program p
      integer :: z(2)
      z = 1.2 + [integer :: 3.5, 4.5]
      print *, z
   end

it compiles with gfortran snippet.f90
but running ./a.out prints an unexpected
           1           1

---

On the other hand, something like this with a zero-sized array ...
   program p
      integer(8) :: z(2)
      z = 1 + [ integer(8) :: [ integer(4) :: ], 1, 2 ]
   end

gives an ...
f951: internal compiler error: Segmentation fault

[Bug fortran/66193] ICE for initialisation of some non-zero-sized arrays

[tkoenig at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=66193

Thomas Koenig <tkoenig at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |tkoenig at gcc dot gnu.org

--- Comment #2 from Thomas Koenig <tkoenig at gcc dot gnu.org> ---
Do you compile these test cases with -O?  What happens if you specify
-fno-frontend-optimize ?

[Bug fortran/66193] ICE for initialisation of some non-zero-sized arrays

[gerhard.steinmetz.fortran@t-online.de]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=66193

--- Comment #3 from Gerhard Steinmetz <gerhard.steinmetz.fortran@t-online.de> ---

Hmm, no observable difference with option -fno-frontend-optimize, sorry.

Of course I probed some combinations for several options.
One example for a more extensive "debug" run :

   gfortran  \
            -O0  -g  -msse3  \
            -fno-omit-frame-pointer  \
            -fdefault-real-8  \
            -fno-align-commons  \
            -std=f2008  \
            -pedantic  \
            -fcheck=all  \
            -fbacktrace  \
            -ftrapv  \
            -ffpe-trap=invalid,zero,overflow  \
            -fstack-protector  \
            -fstack-protector-all  \
            \
            -Wall  \
            -Wextra  \
            -Wsurprising  \
            -Warray-bounds  \
            -Warray-temporaries  \
            -Wcharacter-truncation  \
            -Wconversion-extra  \
            -Wfunction-elimination  \
            -Wrealloc-lhs  \
            -Wrealloc-lhs-all  \
            -Wtabs  \
            -Wunderflow  \
            -Wunused-dummy-argument  \
            \
            # ... some more
            # ... with and without -c

[Bug fortran/66193] ICE for initialisation of some non-zero-sized arrays

[kargl at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=66193

kargl at gcc dot gnu.org changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
           Priority|P3                          |P4
                 CC|                            |kargl at gcc dot gnu.org

--- Comment #4 from kargl at gcc dot gnu.org ---
Confirmed.  Here's a slightly expanded testcase where it
shows that the first 2 lines are failing.

   program p
      real :: z(2)
      z = 1 + [real :: 1, 2]        ! fails
      z = 1. + [real :: 1, 2]       ! fails
      z = 1. + [1, 2]                ! OK
      z = 1. + [real :: 1., 2.]      ! OK
      z = 1. + [real :: 1.d0, 2.d0]  ! OK
      z = 1 + [1, 2]                 ! OK
      z = 1 + [real :: 1., 2.]       ! OK
      z = 1 + [real :: 1.d0, 2.d0]   ! OK
      z = [1, 2] + 1                 ! OK
      z = [real :: 1., 2.] + 1       ! OK
      z = [real :: 1.d0, 2.d0] + 1   ! OK
      z = [1, 2] + 1.                ! OK
      z = [real :: 1., 2.] + 1.      ! OK
      z = [real :: 1.d0, 2.d0] + 1.  ! OK
      print *, z
   end

[Bug fortran/66193] ICE for initialisation of some non-zero-sized arrays

[tkoenig at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=66193

Thomas Koenig <tkoenig at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|UNCONFIRMED                 |NEW
   Last reconfirmed|                            |2015-05-19
     Ever confirmed|0                           |1

--- Comment #5 from Thomas Koenig <tkoenig at gcc dot gnu.org> ---
This does not fail:

ig25@linux-fd1f:~/Krempel/Constructor> cat gar.f90
   program p
      real :: z(2)
      z = [real :: 1, 2] + 1
      print *,z
   end
ig25@linux-fd1f:~/Krempel/Constructor> gfortran gar.f90 
ig25@linux-fd1f:~/Krempel/Constructor> ./a.out
   1.00000000       2.00000000

[Bug fortran/66193] ICE for initialisation of some non-zero-sized arrays

[tkoenig at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=66193

--- Comment #6 from Thomas Koenig <tkoenig at gcc dot gnu.org> ---
The problem manifests itself in reduce_binary_ca.  This function gets handed an
array constructor of type real whose elements are integer.  This makes no
sense.

(gdb) p *op1
$6 = {expr_type = EXPR_CONSTANT, ts = {type = BT_REAL, kind = 4, 
...

(gdb) p *op2
$7 = {expr_type = EXPR_ARRAY, ts = {type = BT_REAL, kind = 4, u = {derived =
0x0, cl = 0x0, pad = 0}, 

...

(gdb) p *(c->expr)
$9 = {expr_type = EXPR_CONSTANT, ts = {type = BT_INTEGER, kind = 4, u =
{derived = 0x0, cl = 0x0, 

So the real error is somewhere else.

[Bug fortran/66193] ICE for initialisation of some non-zero-sized arrays

[sgk at troutmask dot apl.washington.edu]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=66193

--- Comment #7 from Steve Kargl <sgk at troutmask dot apl.washington.edu> ---
On Tue, May 19, 2015 at 06:30:57PM +0000, tkoenig at gcc dot gnu.org wrote:
> The problem manifests itself in reduce_binary_ca.  This function
> gets handed an array constructor of type real whose elements are
> integer.  This makes no sense.
> 
> (gdb) p *op1
> $6 = {expr_type = EXPR_CONSTANT, ts = {type = BT_REAL, kind = 4, 
> ...
> 
> (gdb) p *op2
> $7 = {expr_type = EXPR_ARRAY, ts = {type = BT_REAL, kind = 4, u = {derived =
> 0x0, cl = 0x0, pad = 0}, 
> 
> ...
> 
> (gdb) p *(c->expr)
> $9 = {expr_type = EXPR_CONSTANT, ts = {type = BT_INTEGER, kind = 4, u =
> {derived = 0x0, cl = 0x0, 
> 
> So the real error is somewhere else.
> 

I suspect that we are missing an explicit type conversion step.
In array.c(gfc_match_array_constructor), we have code starting
at line 1141

  /* Size must be calculated at resolution time.  */
  if (seen_ts)
    {
      expr = gfc_get_array_expr (ts.type, ts.kind, &where);
      expr->ts = ts;
    }
  else
    expr = gfc_get_array_expr (BT_UNKNOWN, 0, &where);

I wonder if we should walk the array and do a conversion after
setting 'expr->ts = ts'.

[Bug fortran/66193] ICE for initialisation of some non-zero-sized arrays

[tkoenig at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=66193

--- Comment #8 from Thomas Koenig <tkoenig at gcc dot gnu.org> ---
This patch looks good:

--- arith.c     (Revision 223202)
+++ arith.c     (Arbeitskopie)
@@ -1390,6 +1390,12 @@ reduce_binary (arith (*eval) (gfc_expr *, gfc_expr
   if (op1->expr_type == EXPR_CONSTANT && op2->expr_type == EXPR_CONSTANT)
     return eval (op1, op2, result);

+  if (op1->expr_type == EXPR_ARRAY)
+    gfc_check_constructor_type (op1);
+
+  if (op2->expr_type == EXPR_ARRAY)
+    gfc_check_constructor_type (op2);
+
   if (op1->expr_type == EXPR_CONSTANT && op2->expr_type == EXPR_ARRAY)
     return reduce_binary_ca (eval, op1, op2, result);

[Bug fortran/66193] ICE for initialisation of some non-zero-sized arrays

[tkoenig at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=66193

--- Comment #9 from Thomas Koenig <tkoenig at gcc dot gnu.org> ---
(In reply to Thomas Koenig from comment #8)
> This patch looks good:

but fixes only the ICE, not the wrong-code issue in comment#1.

[Bug fortran/66193] ICE for initialisation of some non-zero-sized arrays

[gerhard.steinmetz.fortran@t-online.de]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=66193

--- Comment #10 from Gerhard Steinmetz <gerhard.steinmetz.fortran@t-online.de> ---
Perhaps it's better to make one factor larger.
Maybe the following will help.

$ cat zz1.f90
   program p
      real :: z(2)
      z = 10 + [real :: 1, 2]
      print *, z
   end
# you may check your patch

$ cat zz2.f90
   program p
      real :: z(2)
      z = 10. + [real :: 1, 2]
      print *, z
   end
# you may check your patch

---

$ cat zz3.f90
   program p
      real :: z(2)
      z = [real :: 1, 2] + 10
      print *, z
   end

$ gfortran zz3.f90
$ a.out
   1.00000000       2.00000000
# expected:
  11.00000000      12.00000000

$ cat zz4.f90
   program p
      real :: z(2)
      z = [real :: 1, 2] + 10.
      print *, z
   end

$ gfortran zz4.f90
$ a.out
   1.00000000       2.00000000
# expected:
  11.00000000      12.00000000

---

Or to use an other basic operation :

$ cat zz5.f90
   program p
      real :: z(2)
      z = -10 * [real :: 1, 2]
      print *, z
   end
# you may check your patch

$ cat zz6.f90
   program p
      real :: z(2)
      z = -10. * [real :: 1, 2]
      print *, z
   end
# you may check your patch

[Bug fortran/66193] ICE for initialisation of some non-zero-sized arrays

[gerhard.steinmetz.fortran@t-online.de]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=66193

--- Comment #11 from Gerhard Steinmetz <gerhard.steinmetz.fortran@t-online.de> ---
Created attachment 35726
  --> https://gcc.gnu.org/bugzilla/attachment.cgi?id=35726&action=edit
test case pr66193_1_real.f90

[Bug fortran/66193] ICE for initialisation of some non-zero-sized arrays

[gerhard.steinmetz.fortran@t-online.de]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=66193

--- Comment #12 from Gerhard Steinmetz <gerhard.steinmetz.fortran@t-online.de> ---
Created attachment 35727
  --> https://gcc.gnu.org/bugzilla/attachment.cgi?id=35727&action=edit
test case pr66193_2_integer.f90

[Bug fortran/66193] ICE for initialisation of some non-zero-sized arrays

[gerhard.steinmetz.fortran@t-online.de]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=66193

--- Comment #13 from Gerhard Steinmetz <gerhard.steinmetz.fortran@t-online.de> ---
Created attachment 35728
  --> https://gcc.gnu.org/bugzilla/attachment.cgi?id=35728&action=edit
test case pr66193_3_diverse.f90

[Bug fortran/66193] ICE for initialisation of some non-zero-sized arrays

[gerhard.steinmetz.fortran@t-online.de]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=66193

--- Comment #14 from Gerhard Steinmetz <gerhard.steinmetz.fortran@t-online.de> ---
Attached some prepared test files.
Added more test cases for integer and real.

Added new test case for logical :
  subroutine s2
    logical(8), parameter :: z1(2) = .true. .or. [ logical(8) :: [ logical(4)
:: ], .false., .false. ]
    logical(8) :: z2(2) = .true. .or. [ logical(8) :: [ logical(4) :: ],
.false., .false. ]
    logical(8) :: z3(2)
    z3 = .true. .or. [ logical(8) :: [ logical(4) :: ], .false., .false. ]
    if ( (.not. z1(1)) .and. (.not. z1(2)) ) call abort
    if ( (.not. z2(1)) .and. (.not. z2(2)) ) call abort
    if ( (.not. z3(1)) .and. (.not. z3(2)) ) call abort
  end subroutine s2

[Bug fortran/66193] ICE for initialisation of some non-zero-sized arrays

[gerhard.steinmetz.fortran@t-online.de]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=66193

--- Comment #15 from Gerhard Steinmetz <gerhard.steinmetz.fortran@t-online.de> ---
Created attachment 35745
  --> https://gcc.gnu.org/bugzilla/attachment.cgi?id=35745&action=edit
test case pr66193_3_diverse_new.f90

Better test, replaces pr66193_3_diverse.f90

[Bug fortran/66193] ICE for initialisation of some non-zero-sized arrays

[anlauf at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=66193

anlauf at gcc dot gnu.org changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |anlauf at gcc dot gnu.org

--- Comment #18 from anlauf at gcc dot gnu.org ---
Created attachment 52346
  --> https://gcc.gnu.org/bugzilla/attachment.cgi?id=52346&action=edit
WIP patch

This fixes at least some of the testcase given in this PR and regtests OK.
May need more testing and fine-tuning.

[Bug fortran/66193] ICE for initialisation of some non-zero-sized arrays

[anlauf at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=66193

anlauf at gcc dot gnu.org changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
           Assignee|unassigned at gcc dot gnu.org      |anlauf at gcc dot gnu.org
             Status|NEW                         |ASSIGNED

--- Comment #19 from anlauf at gcc dot gnu.org ---
Submitted: https://gcc.gnu.org/pipermail/fortran/2022-February/057520.html

[Bug fortran/66193] ICE for initialisation of some non-zero-sized arrays

[cvs-commit at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=66193

--- Comment #20 from CVS Commits <cvs-commit at gcc dot gnu.org> ---
The master branch has been updated by Harald Anlauf <anlauf@gcc.gnu.org>:

https://gcc.gnu.org/g:f3ffea93ef31c03ad8cdcb54e71ec868b57b264f

commit r12-7153-gf3ffea93ef31c03ad8cdcb54e71ec868b57b264f
Author: Harald Anlauf <anlauf@gmx.de>
Date:   Sun Feb 6 21:47:20 2022 +0100

    Fortran: try simplifications during reductions of array constructors

    gcc/fortran/ChangeLog:

            PR fortran/66193
            * arith.cc (reduce_binary_ac): When reducing binary expressions,
            try simplification.  Handle case of empty constructor.
            (reduce_binary_ca): Likewise.

    gcc/testsuite/ChangeLog:

            PR fortran/66193
            * gfortran.dg/array_constructor_55.f90: New test.

[Bug fortran/66193] ICE for initialisation of some non-zero-sized arrays

[cvs-commit at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=66193

--- Comment #21 from CVS Commits <cvs-commit at gcc dot gnu.org> ---
The releases/gcc-11 branch has been updated by Harald Anlauf
<anlauf@gcc.gnu.org>:

https://gcc.gnu.org/g:a5088b07e0a547cdacaa6db8ad762fd8c425b1f0

commit r11-9613-ga5088b07e0a547cdacaa6db8ad762fd8c425b1f0
Author: Harald Anlauf <anlauf@gmx.de>
Date:   Sun Feb 6 21:47:20 2022 +0100

    Fortran: try simplifications during reductions of array constructors

    gcc/fortran/ChangeLog:

            PR fortran/66193
            * arith.c (reduce_binary_ac): When reducing binary expressions,
            try simplification.  Handle case of empty constructor.
            (reduce_binary_ca): Likewise.

    gcc/testsuite/ChangeLog:

            PR fortran/66193
            * gfortran.dg/array_constructor_55.f90: New test.

    (cherry picked from commit f3ffea93ef31c03ad8cdcb54e71ec868b57b264f)

[Bug fortran/66193] ICE for initialisation of some non-zero-sized arrays

[anlauf at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=66193

anlauf at gcc dot gnu.org changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
   Target Milestone|---                         |11.3
             Status|ASSIGNED                    |RESOLVED
         Resolution|---                         |FIXED

--- Comment #22 from anlauf at gcc dot gnu.org ---
Fixed on mainline for gcc-12, and on 11-branch.  Closing.

Thanks for the report!