[Bug fortran/54934] New: Invalid debug/ array bounds w/-fno-range-check and 32-bit target

[Bug fortran/54934] New: Invalid debug/ array bounds w/-fno-range-check and 32-bit target

[jan.kratochvil at redhat dot com]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=54934

             Bug #: 54934
           Summary: Invalid debug/ array bounds w/-fno-range-check and
                    32-bit target
    Classification: Unclassified
           Product: gcc
           Version: 4.8.0
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: fortran
        AssignedTo: unassigned@gcc.gnu.org
        ReportedBy: jan.kratochvil@redhat.com
            Target: i386-unknown-linux-gnu


Siddhesh implemented support for 64-bit array bounds:
[PATCH] Expand fortran array bounds sizes to LONGEST
http://sourceware.org/ml/gdb-patches/2012-08/msg00562.html
+        dimension foo(4294967296:4294967297)
+        dimension bar(-4294967297:-4294967296)

This requires -fno-range-check for gfortran to compile it at all.

It works for x86_64 target but it produces wrong debug info for i386 target.

FAIL: gcc (GCC) 4.8.0 20121015 (experimental)

Is this a bug or not?  Should GDB support 64-bit Fortran array bounds?
Should such bounds require -fno-range-check?

[Bug fortran/54934] Invalid debug/ array bounds w/-fno-range-check and 32-bit target

[kargl at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=54934

kargl at gcc dot gnu.org changed:

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

--- Comment #1 from kargl at gcc dot gnu.org 2012-10-15 15:40:28 UTC ---
Jan,

It's a convoluted problem.  In gfortran, to get
the maximum range of an integer type you need
to use [-huge(i)-1:huge(i)] where 'i' is of the
type of interest.  This is a result of gfortran
seeing a number like -4294967297 as a unary minus
with an operand of 4294967297.  4294967297 without
a kind suffix is of default integer kind, which is
a 32-bit signed int, and 4294967297 is slightly 
larger than huge(1) = 2147483647.  Note, Fortran
does not have unsigned integers.  If you don't
want to or cannot use the -fno-range-check option,
you can do 

        integer(8), parameter :: n = 2 * (huge(1_4) + 1_8) + 1
        dimension foo(n-1:n)
        dimension bar(-n-1:-n)
        bar = 42
        foo = bar
        print *, n
        stop
        end

where I correctly initialize bar to some value; otherwise, the
code is invalid Fortran.  You could also decorate the integers
with a integer kind suffix

        dimension foo(4294967296_8:4294967297_8)
        dimension bar(-4294967297_8:-4294967296_8)
        bar = 42
        foo=bar
        stop
        end

Here, the _8 tells gfortran to use a signed 64-bit int.

Both code snippets above compile on my x86_64-*-freebsd.
I don't know want happens on an i686 target.

[Bug fortran/54934] Invalid debug/ array bounds w/-fno-range-check and 32-bit target

[jan.kratochvil at redhat dot com]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=54934

--- Comment #2 from Jan Kratochvil <jan.kratochvil at redhat dot com> 2012-10-15 18:41:59 UTC ---
Created attachment 28451
  --> http://gcc.gnu.org/bugzilla/attachment.cgi?id=28451
.f source for a reproducer

Submitting it as a PR as the testcase no longer requires -fno-range-check and
it produces valid debug/ output with -m64 but invalid with -m32.
gcc (GCC) 4.8.0 20121015 (experimental)

[Bug debug/54934] [fortran] Invalid >4GB array bounds with 32-bit target

[jan.kratochvil at redhat dot com]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=54934

--- Comment #3 from Jan Kratochvil <jan.kratochvil at redhat dot com> 2012-10-15 18:47:05 UTC ---
Created attachment 28452
  --> http://gcc.gnu.org/bugzilla/attachment.cgi?id=28452
.S from reproducer for -m32 output fixed by hand

.S file is a fix-up by hand to make -m32 working:
#if 0
original - not working
#else
patched - working
#endif

Or for original GCC output:
-m64 (working) -> -m32 (not working)

    <88>   DW_AT_lower_bound : 0xfffffffeffffffff
    <90>   DW_AT_upper_bound : 0xffffffff00000000
->
    <74>   DW_AT_lower_bound : 0xffffffff
    <78>   DW_AT_upper_bound : 0        

    <b5>   DW_AT_lower_bound : 0x100000000
    <bd>   DW_AT_upper_bound : 0x100000001
->
    <96>   DW_AT_lower_bound : 0
    <97>   DW_AT_upper_bound : 1