[Bug fortran/46641] New: Specification-expr checks and results with C_SIZEOF and SIZEOF

[Bug fortran/46641] New: Specification-expr checks and results with C_SIZEOF and SIZEOF

[burnus at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=46641

           Summary: Specification-expr checks and results with C_SIZEOF
                    and SIZEOF
           Product: gcc
           Version: 4.6.0
            Status: UNCONFIRMED
          Keywords: wrong-code
          Severity: normal
          Priority: P3
         Component: fortran
        AssignedTo: unassigned@gcc.gnu.org
        ReportedBy: burnus@gcc.gnu.org


Kind of follow up to PR 46638; based on
http://groups.google.com/group/comp.lang.fortran/browse_thread/thread/1b907e3b7b6f3461


There is something fishy about the following program. In terms of validity, the
question is whether
   C_SIZEOF (structure-constructor ())
is a valid initialization expression -- gfortran accepts it.

Besides the validity issue, there is something fishy. The "string" has a length
of 16 -- using either C_SIZEOF or the literal 16 in the declaration. However,
if one uses the literal, it works while using C_SIZEOF the TRANSFER seems to
override some memory it shouldn't - causing a segfault.

Using "sizeof()" instead of "c_sizeof()", the either version is rejected by the
Cray compiler - but both version work with ifort. Neither compiler likes the
C_sizeof version (not implemented in Cray, not sure about ifort).

gfortran prints for sizeof the error "must have constant character length" for
both versions.


module utils
   use ISO_C_BINDING
   implicit none
   type, bind(C) :: CPUID_type
      integer(C_INT32_T) eax
      integer(C_INT32_T) ebx
      integer(C_INT32_T) edx
      integer(C_INT32_T) ecx
   end type CPUID_type
   type(CPUID_type) :: test_type
end module utils

program mxcsr_test
   use ISO_C_BINDING
   use utils
   implicit none
   integer(C_INT32_T) reg
   integer(C_INT64_T) ts1, ts2
   integer(C_INT16_T) sw, cw
   integer(C_INT32_T) eax, ecx
   type(CPUID_type) result
!  Works:
!    character(16) string
!  Compiles but fails:
     character(SIZEOF(CPUID_type(0,0,0,0))) string
!  Rejected with: must have constant character length
!    character(SIZEOF(test_type)) string

   result = CPUID_TYPE(int(Z'0000000A'),int(Z'756E6547'), &
      int(Z'49656E69'),int(Z'6C65746E'))
   print *, len(string) ! OK: 16
   string = transfer(result,string)
   write(*,'(a,a)') 'Brand name: ',string(5:16)
  write(*,'(4(z8.8:1x))') result
end program mxcsr_test

[Bug fortran/46641] Specification-expr checks and results with C_SIZEOF and SIZEOF

[burnus at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=46641

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
         Depends on|                            |36437, 45824

--- Comment #1 from Tobias Burnus <burnus at gcc dot gnu.org> 2010-11-24 14:53:27 UTC ---
(In reply to comment #0)
> In terms of validity, the question is whether
>    C_SIZEOF (structure-constructor ())
> is a valid initialization expression -- gfortran accepts it.

I marked the PR now as regression as gfortran 4.5 accepts it - but does not
have the segfault issue at run time.

That issue seems to be related to the compile-time simplification - or rather
the lack of. Both 4.5 and 4.6 produce the dump:

  character(kind=1) str[1:.str];

TODO:
a) Fix the issue above, cf. PR 36437

b) decide which of the various c_sizeof/sizeof is valid and thus should be
accepted or rejected; look also at the the thread (cf. comment 0) and in
particular at
http://groups.google.com/group/comp.lang.fortran/msg/8b36b578793b7037 , cf.
also PR 45824

[Bug fortran/46641] Specification-expr checks and results with C_SIZEOF and SIZEOF

[kargl at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=46641

kargl at gcc dot gnu.org changed:

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

--- Comment #2 from kargl at gcc dot gnu.org 2010-11-24 17:53:29 UTC ---
>From F2008, 13.1:

   Intrinsic procedures are divided into seven classes: inquiry
   functions, ...

Then one finds:

   15.2.3.7  C SIZEOF (X)

   Description. Size of X in bytes.

   Class. Inquiry function.

Then from 7.1.11:

   A specification inquiry is a reference to
   ...
     (4) the function C_SIZEOF from the intrinsic module ISO_C_BINDING
        (15.2.3.7), or

Then in "7.1.12 Constant Expression"

     (4) a specification inquiry where each designator or function
         argument is
         (a)   a constant expression or

So, c_sizeof() should be accepted by gfortran with -std=f2008 or gnu,
need to check f2003 whether -std=f2003 should accept it, and reject
for -std=f95.

[Bug fortran/46641] Specification-expr checks and results with C_SIZEOF and SIZEOF

[burnus at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=46641

--- Comment #3 from Tobias Burnus <burnus at gcc dot gnu.org> 2010-11-27 21:57:01 UTC ---
The issue with specification issues is now fixed by PR 46678 (almost, not yet
committed). The TRANSFER issue was fixed as PR 46638.

Thus, remaining is only using C_SIZEOF - and by completeness - SIZEOF in
initialization expressions.

I think the latter is really tracked in PR 36437. Related is PR 45824.

[Bug fortran/46641] Specification-expr checks and results with C_SIZEOF and SIZEOF

[dominiq at lps dot ens.fr]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=46641

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|UNCONFIRMED                 |WAITING
   Last reconfirmed|                            |2013-06-20
     Ever confirmed|0                           |1

--- Comment #4 from Dominique d'Humieres <dominiq at lps dot ens.fr> ---
!  Rejected with: must have constant character length
!    character(SIZEOF(test_type)) string

This is the case up to revision 197010 (2013-03-23), but the code is accepted
after revision 197550 (2013-04-07). Can this PR be closed as FIXED?

[Bug fortran/46641] Specification-expr checks and results with C_SIZEOF and SIZEOF

[dominiq at lps dot ens.fr]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=46641

--- Comment #5 from Dominique d'Humieres <dominiq at lps dot ens.fr> ---
> This is the case up to revision 197010 (2013-03-23), but the code is 
> accepted after revision 197550 (2013-04-07). Can this PR be closed as FIXED?

Likely revision 197053.

[Bug fortran/46641] Specification-expr checks and results with C_SIZEOF and SIZEOF

[dominiq at lps dot ens.fr]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=46641

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

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

--- Comment #6 from Dominique d'Humieres <dominiq at lps dot ens.fr> ---
No feedback since almost six months. Closing as FIXED. Please open a new PR if
I have missed any remaining issue.

[Bug fortran/46641] Specification-expr checks and results with C_SIZEOF and SIZEOF

[dominiq at lps dot ens.fr]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=46641

Bug 46641 depends on bug 45824, which changed state.

Bug 45824 Summary: Update expr.c's check_inquiry for C_SIZEOF, compiler_version/_options, etc.
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=45824

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