[Bug libfortran/47434] New: Wrong field width for NaN with (F0.n) formatting

[Bug libfortran/47434] New: Wrong field width for NaN with (F0.n) formatting

[thenlich at users dot sourceforge.net]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=47434

           Summary: Wrong field width for NaN with (F0.n) formatting
           Product: gcc
           Version: 4.6.0
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: libfortran
        AssignedTo: unassigned@gcc.gnu.org
        ReportedBy: thenlich@users.sourceforge.net


When formatting NaN with a F edit descriptor where the field width is zero, the
string "<blank>NaN" is written (field width = 4).

The expected string would be "NaN" (field width = 3, the smallest positive
actual field width that does not result in a field filled with asterisks).

For example: 

program testnan
    real :: n = 0.0
    n = 0.0 / n
    print "(F0.2)", n
    print "(F3.2)", n
end program testnan

Output:
<blank>NaN
NaN

10.7.2.1
(6) On output, with I, B, O, Z, F, and G editing, the specified value of the
field width w may be zero. In such cases, the processor selects the smallest
positive actual field width that does not result in a field filled with
asterisks. The specified value of w shall not be zero on input.

10.7.2.3.2 F editing
For an internal value that is an IEEE NaN, the output field consists of blanks,
if necessary, followed by the letters 'NaN' and optionally followed by one to w
- 5 alphanumeric processor-dependent characters enclosed in parentheses, right
justified within the field. If w is less than 3, the field is filled with
asterisks.

[Bug libfortran/47434] Wrong field width for NaN with (F0.n) formatting

[jvdelisle at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=47434

Jerry DeLisle <jvdelisle at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|UNCONFIRMED                 |ASSIGNED
   Last reconfirmed|                            |2011.01.24 13:12:19
                 CC|                            |jvdelisle at gcc dot
                   |                            |gnu.org
         AssignedTo|unassigned at gcc dot       |jvdelisle at gcc dot
                   |gnu.org                     |gnu.org
     Ever Confirmed|0                           |1

--- Comment #1 from Jerry DeLisle <jvdelisle at gcc dot gnu.org> 2011-01-24 13:12:19 UTC ---
I will investigate.

[Bug libfortran/47434] Wrong field width for NaN with (F0.n) formatting

[thenlich at users dot sourceforge.net]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=47434

--- Comment #2 from Thomas Henlich <thenlich at users dot sourceforge.net> 2011-01-24 14:37:29 UTC ---
A similar issue occurs with the values +Infinity and 0.0:

program testnan
    real :: i, n = 0.0, m = tiny(0.0)
    i = 1.0 / n
    print "(F0.2)", i
    print "(F3.2)", i

    print "(F0.2)", n
    print "(F3.2)", n

    print "(F0.2)", m
    print "(F3.2)", m
end program testnan

Output:
+Inf
Inf
0.00
.00
.00
.00

The expected chosen field width is 3 in all cases, exactly for the same reason
demanded by 10.7.2.1(6)

[Bug libfortran/47434] Wrong field width for NaN with (F0.n) formatting

[jvdelisle at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=47434

--- Comment #3 from Jerry DeLisle <jvdelisle at gcc dot gnu.org> 2011-01-27 12:47:04 UTC ---
I agree to fix the NaN case. I have the patch already.

Regarding the other cases:

The copy of the F2008 standard I have says in 10.7.2.3.2 F editing:

"When w is zero, the processor selects the field width."

I also see this:

"Leading zeros are not permitted except for an optional zero immediately to the
left of the decimal symbol if the magnitude of the value in the output field is
less than one."

Similarly for the '+' sign on Inf.  Until now we have chosen to emit the
optional characters.

[Bug libfortran/47434] Wrong field width for NaN with (F0.n) formatting

[burnus at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=47434

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

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

--- Comment #4 from Tobias Burnus <burnus at gcc dot gnu.org> 2011-01-27 13:08:22 UTC ---
(In reply to comment #3)
> Regarding the other cases:
> Similarly for the '+' sign on Inf.  Until now we have chosen to emit the
> optional characters.

Comparison with other compilers for the example of comment 3.

a) Crayftn 7.2.4
  Infinity
Inf
      0.00
.00
      0.00
.00

b) ifort 11.1, PathScale 3.2.99, Open64 4.2.2.1, 
Infinity
Inf
0.00
.00
0.00
.00

c) Sun Fortran 95 8.3, pgf90 10.5-0
Inf
Inf
0.00
.00
0.00
.00

d) g95 4.03 (0.93)
+Inf
Inf
.00
.00
.00
.00

e) xlf95/AIX V12.1
INF
INF
0.00
.00
0.00
.00

f) NAG 5.1
        <<< Yes, that's an zero-length string
Inf
.00
.00
.00
.00

g) g77: Does not like w = 0 and prints "inf" (small "i"), and ".00" for the
other lines.

[Bug libfortran/47434] Wrong field width for NaN with (F0.n) formatting

[thenlich at users dot sourceforge.net]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=47434

--- Comment #5 from Thomas Henlich <thenlich at users dot sourceforge.net> 2011-01-27 13:25:15 UTC ---
(In reply to comment #3)
> The copy of the F2008 standard I have says in 10.7.2.3.2 F editing:
> 
> "When w is zero, the processor selects the field width."

My interpretation is this:

Section 10.7.2.1 (General rules) applies to these cases, which demands
specifically:

"... the processor selects the smallest positive actual field width that does
not result in a field filled with asterisks. ..."

The statement from 10.7.2.3.2 "When w is zero, the processor selects the field
width." does only refer to this. IMHO it does not suggest the possibility to
include optional characters (a leading decimal zero or plus sign) which make
the field width larger than required.

In the examples, since w=3 ("F3.2") does not result in a field filled with
asterisks, and is the smallest possible such value, the processor must select a
field width of 3.

Also, it is inconsistent that the following commands should result in a
different output:

print "(F0.2)", 0.0
print "(F0.2)", tiny(0.0)

0.00
.00

[Bug libfortran/47434] Wrong field width for NaN with (F0.n) formatting

[burnus at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=47434

--- Comment #6 from Tobias Burnus <burnus at gcc dot gnu.org> 2011-01-27 13:30:30 UTC ---
(In reply to comment #4)
> (In reply to comment #3)
> > Regarding the other cases:
> > Similarly for the '+' sign on Inf.  Until now we have chosen to emit the
> > optional characters.

There seems to be a bug with regards to S:

  print "(s,F0.2)", inf

still prints "+Inf". ("The edit descriptors SS, SP, and S set the sign mode
corresponding to the SIGN= specifier values SUPPRESS, PLUS, and
PROCESSOR_DEFINED, respectively.")


"(6) On output, with I, B, O, Z, F, and G editing, the specified value of the
field width w may be zero. In such cases, the processor selects the smallest
positive actual field width that does not result in a field filled with
asterisks" (10.7.2.1 General rules)

I have to agree with Thomas here: ".00" and "Inf" are the shorted possible
strings, unless SP (sign="PLUS") is used; then "+.00" and "+Inf" are the
shortest possible strings.

In that sense, none of my 10 compiler gets it right ...

[Bug libfortran/47434] Wrong field width for NaN with (F0.n) formatting

[jvdelisle at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=47434

--- Comment #7 from Jerry DeLisle <jvdelisle at gcc dot gnu.org> 2011-01-27 16:35:13 UTC ---
I remember the discussion when we first implemented this and we wanted the + on
Inf to be consistent with -Inf. 

Oh well, if folks really feel strongly about this, I can change it all.  Hmm, I
think I will add the optional NaN type field so now the smallest I can fit will
be quite a bit wider then 3.  (Just kidding ;) )

[Bug libfortran/47434] Wrong field width for NaN with (F0.n) formatting

[burnus at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=47434

--- Comment #8 from Tobias Burnus <burnus at gcc dot gnu.org> 2011-01-27 17:21:24 UTC ---
(In reply to comment #7)
> I remember the discussion when we first implemented this and we wanted the + on
> Inf to be consistent with -Inf. 

Well, the following is in any case wrong:

  real :: inf
  inf = 0
  inf = 1/inf
  print "(sp,F0.2)", inf  ! SIGN=PLUS
  print "(ss,F0.2)", inf  ! SIGN=SUPPRESS
  print "(s, F0.2)", inf  ! SIGN=PROCESSOR_DEFINED
  end

The same result for "F4.2": Three times "+Inf". (Ditto for g95.)

For SIGN=SUPPRESS it should just print "Inf", whether "Inf" or "+Inf" is
correct for PROCESSOR_DEFINED is another question. For consistency with the
other compilers - and for consistency with 10.7.2.1(6), one should probably
remove the "+".

For "F4.2" also the other compilers do not seem to print a + with
sign='PROCESSOR_DEFINED' though I am not sure that I like
Intel's/pathscale's/open64's "Infi" for S and SS better. NAG and Sunf95 have "
Inf" for S/SS.


Regarding "0.00" vs ".00" with "F0.2": I do not have any real preference;
following 10.7.2.1(6) I am slightly inclined that ".00" is better.

[Bug libfortran/47434] Wrong field width for NaN with (F0.n) formatting

[thenlich at users dot sourceforge.net]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=47434

--- Comment #9 from Thomas Henlich <thenlich at users dot sourceforge.net> 2011-01-27 17:55:13 UTC ---
Maybe this should be a command option, possibly defaulting to the current
behaviour unless -std=xx is given.

[Bug libfortran/47434] Wrong field width for NaN with (F0.n) formatting

[jvdelisle at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=47434

--- Comment #10 from Jerry DeLisle <jvdelisle at gcc dot gnu.org> 2011-01-29 17:31:07 UTC ---
Author: jvdelisle
Date: Sat Jan 29 17:31:04 2011
New Revision: 169390

URL: http://gcc.gnu.org/viewcvs?root=gcc&view=rev&rev=169390
Log:
2011-01-29  Jerry DeLisle  <jvdelisle@gcc.gnu.org>

    PR libgfortran/47434
    * io/write_float.def (write_infnan): Use calculate_sign to determine
    if the sign should be given and check field widths accordingly.

Modified:
    trunk/libgfortran/ChangeLog
    trunk/libgfortran/io/write_float.def

[Bug libfortran/47434] Wrong field width for NaN with (F0.n) formatting

[jvdelisle at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=47434

--- Comment #11 from Jerry DeLisle <jvdelisle at gcc dot gnu.org> 2011-01-29 17:33:39 UTC ---
Author: jvdelisle
Date: Sat Jan 29 17:33:36 2011
New Revision: 169391

URL: http://gcc.gnu.org/viewcvs?root=gcc&view=rev&rev=169391
Log:
2011-01-29  Jerry DeLisle  <jvdelisle@gcc.gnu.org>

    PR libgfortran/47434
    * gfortran.dg/read_infnan_1.f90: Update test.
    * gfortran.dg/module_nan.f90: Update test.
    * gfortran.dg/char4_iunit_1.f03: Update test.
    * gfortran.dg/large_real_kind_1.f90: Update test.
    * gfortran.dg/real_const_3.f90: Update test.
    * gfortran.fortran-torture/execute/nan_inf_fmt.f90: Update test.

Modified:
    trunk/gcc/testsuite/ChangeLog
    trunk/gcc/testsuite/gfortran.dg/char4_iunit_1.f03
    trunk/gcc/testsuite/gfortran.dg/large_real_kind_1.f90
    trunk/gcc/testsuite/gfortran.dg/module_nan.f90
    trunk/gcc/testsuite/gfortran.dg/read_infnan_1.f90
    trunk/gcc/testsuite/gfortran.dg/real_const_3.f90
    trunk/gcc/testsuite/gfortran.fortran-torture/execute/nan_inf_fmt.f90

[Bug libfortran/47434] Wrong field width for NaN with (F0.n) formatting

[jvdelisle at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=47434

Jerry DeLisle <jvdelisle at gcc dot gnu.org> changed:

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

--- Comment #12 from Jerry DeLisle <jvdelisle at gcc dot gnu.org> 2011-01-29 17:59:31 UTC ---
fixed and closing

[Bug libfortran/47434] Wrong field width for NaN with (F0.n) formatting

[dnovillo at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=47434

--- Comment #13 from Diego Novillo <dnovillo at gcc dot gnu.org> 2011-02-02 18:02:34 UTC ---
Author: dnovillo
Date: Wed Feb  2 18:02:28 2011
New Revision: 169686

URL: http://gcc.gnu.org/viewcvs?root=gcc&view=rev&rev=169686
Log:
2011-01-29  Jerry DeLisle  <jvdelisle@gcc.gnu.org>

    PR libgfortran/47434
    * io/write_float.def (write_infnan): Use calculate_sign to determine
    if the sign should be given and check field widths accordingly.

Modified:
    branches/google/integration/libgfortran/ChangeLog
    branches/google/integration/libgfortran/io/write_float.def

[Bug libfortran/47434] Wrong field width for NaN with (F0.n) formatting

[dnovillo at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=47434

--- Comment #14 from Diego Novillo <dnovillo at gcc dot gnu.org> 2011-02-02 18:02:43 UTC ---
Author: dnovillo
Date: Wed Feb  2 18:02:38 2011
New Revision: 169687

URL: http://gcc.gnu.org/viewcvs?root=gcc&view=rev&rev=169687
Log:
2011-01-29  Jerry DeLisle  <jvdelisle@gcc.gnu.org>

    PR libgfortran/47434
    * gfortran.dg/read_infnan_1.f90: Update test.
    * gfortran.dg/module_nan.f90: Update test.
    * gfortran.dg/char4_iunit_1.f03: Update test.
    * gfortran.dg/large_real_kind_1.f90: Update test.
    * gfortran.dg/real_const_3.f90: Update test.
    * gfortran.fortran-torture/execute/nan_inf_fmt.f90: Update test.

Modified:
    branches/google/integration/gcc/testsuite/ChangeLog
    branches/google/integration/gcc/testsuite/gfortran.dg/char4_iunit_1.f03
    branches/google/integration/gcc/testsuite/gfortran.dg/large_real_kind_1.f90
    branches/google/integration/gcc/testsuite/gfortran.dg/module_nan.f90
    branches/google/integration/gcc/testsuite/gfortran.dg/read_infnan_1.f90
    branches/google/integration/gcc/testsuite/gfortran.dg/real_const_3.f90
   
branches/google/integration/gcc/testsuite/gfortran.fortran-torture/execute/nan_inf_fmt.f90