[Bug libfortran/50105] New: Possibly: [4.3/4.4/4.5/4.6/4.7 Regression] I/O with g6.5 - wrong number of "**" shown

[Bug libfortran/50105] New: Possibly: [4.3/4.4/4.5/4.6/4.7 Regression] I/O with g6.5 - wrong number of "**" shown

[burnus at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=50105

             Bug #: 50105
           Summary: Possibly: [4.3/4.4/4.5/4.6/4.7 Regression] I/O with
                    g6.5 - wrong number of "**" shown
    Classification: Unclassified
           Product: gcc
           Version: 4.7.0
            Status: UNCONFIRMED
          Keywords: wrong-code
          Severity: normal
          Priority: P3
         Component: libfortran
        AssignedTo: unassigned@gcc.gnu.org
        ReportedBy: burnus@gcc.gnu.org
                CC: jvdelisle@gcc.gnu.org, thenlich@users.sourceforge.net


Bob Corbett (formally SUN, now Oracle) wonders at
http://j3-fortran.org/pipermail/j3/2011-August/004573.html
what the following program should print:

     PROGRAM MAIN
       PRINT '(G6.5)', 1.0
     END


Crayftn 7.1.4.111, gfortran 4.3 and 4.7, ifort 11.1, openf95, sunf95, and
pathf95 produce the following output (when piped through "od -a"):

0000000   *   *   *   *   *   *  nl
0000007


While g95, gfortran 4.1, NAG f95 5.1 and PGI 11.5 produce:

0000000   *   *  sp  sp  sp  sp  nl
0000007


Bob thinks that only the latter is correct.

[Bug libfortran/50105] Possibly: [4.3/4.4/4.5/4.6/4.7 Regression] I/O with g6.5 - wrong number of "**" shown

[burnus at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=50105

--- Comment #1 from Tobias Burnus <burnus at gcc dot gnu.org> 2011-08-17 06:22:20 UTC ---
Actually, I think that this bug is invalid - but wait for the answers at the J3
mailing list. The reason I think that I think that 6 asterisks should be
produced are stated below. However, it would not be the first time that I
missed something.


We have (refs = F2008):

  R1007 data-edit-desc  is  ...
                        or  G w [ . d [ E e ] ]

such that for G6.5, one has "w = 6".


And thus, I expect 6 "*" as: if the "characters produced exceeds the field
width, the processor shall fill the entire field of width w with asterisks"


Full quote:

10.7.2 Numeric editing, 10.7.2.1 General rules states:

"(5) On output, if an exponent exceeds its specified or implied width using the
E, EN, ES, D, or G edit descriptor, or the number of characters produced
exceeds the field width, the processor shall fill the entire field of width w
with asterisks. However, the processor shall not produce asterisks if the field
width is not exceeded when optional characters are omitted."

[Bug libfortran/50105] Possibly: [4.3/4.4/4.5/4.6/4.7 Regression] I/O with g6.5 - wrong number of "**" shown

[burnus at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=50105

--- Comment #2 from Tobias Burnus <burnus at gcc dot gnu.org> 2011-08-17 06:40:00 UTC ---
Bob's point is that for 1.0, the following rule (10.7.5.2.2) applies:

"Magnitude of Internal Value" is 1.0 and thus the "Equivalent Conversion" is
F2.5,4(' '). Thus, one has "**" for F followed by 4 spaces.

Cf. http://j3-fortran.org/pipermail/j3/2011-August/004575.html

[Bug libfortran/50105] Possibly: [4.3/4.4/4.5/4.6/4.7 Regression] I/O with g6.5 - wrong number of "**" shown

[thenlich at users dot sourceforge.net]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=50105

--- Comment #3 from Thomas Henlich <thenlich at users dot sourceforge.net> 2011-08-17 07:36:54 UTC ---
Maybe we can trace back the change in GFortran between 4.1 and 4.3 and find out
why it was changed?

[Bug libfortran/50105] Possibly: [4.6/4.7 Regression] I/O with g6.5 - wrong number of "**" shown

[burnus at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=50105

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
            Summary|Possibly:                   |Possibly: [4.6/4.7
                   |[4.3/4.4/4.5/4.6/4.7        |Regression] I/O with g6.5 -
                   |Regression] I/O with g6.5 - |wrong number of "**" shown
                   |wrong number of "**" shown  |

--- Comment #4 from Tobias Burnus <burnus at gcc dot gnu.org> 2011-08-17 09:12:38 UTC ---
(In reply to comment #3)
> Maybe we can trace back the change in GFortran between 4.1 and 4.3 and find out
> why it was changed?

I think it was a mistake of mine: I was actually testing with GCC 4.7 as I
misloaded the module. (I use "module" to load different shell environment
settings.) The change seems to be much more recent:

GCC 4.7 uses in libgfortran/io/write_float.def's OUTPUT_FLOAT_FMT_G:

  if (nb > 0 && !dtp->u.p.g0_no_blanks)\
    ...
      if (result == FAILURE)\
        pad = '*';\

This seems to have been added 2011-01-28 for PR 47285 in patch
http://gcc.gnu.org/ml/fortran/2011-01/msg00226.html and thus it should should
only affect GCC 4.6.x and the 4.7 trunk.

Assuming the standards guys agree that the old behaviour is correct, it is
probably sufficient to delete the two quoted lines.

[Bug libfortran/50105] Possibly: [4.6/4.7 Regression] I/O with g6.5 - wrong number of "**" shown

[burnus at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=50105

--- Comment #5 from Tobias Burnus <burnus at gcc dot gnu.org> 2011-08-17 13:05:07 UTC ---
To continue tracking: Bill Long (Cray) thinks that "******" (i.e. the current
gfortran behaviour) is correct as:

"The key point here, which does not show up in email font, is that, in the
standard, the "w" in the phrase "entire field of width w"  is in syntax font. 
For G format, this ties to the line of R1007 noted above.  The value of w is
clearly 6 here. That's what appears in the program."

Cf. http://j3-fortran.org/pipermail/j3/2011-August/004580.html

[Bug libfortran/50105] Possibly: I/O with g6.5 - wrong number of "**" shown

[burnus at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=50105

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
            Summary|Possibly: [4.6/4.7          |Possibly: I/O with g6.5 -
                   |Regression] I/O with g6.5 - |wrong number of "**" shown
                   |wrong number of "**" shown  |

--- Comment #6 from Tobias Burnus <burnus at gcc dot gnu.org> 2011-08-18 08:26:28 UTC ---
While Bob disagrees with Bill:
  http://j3-fortran.org/pipermail/j3/2011-August/004581.html

I removed now the regression marker; I think it will end up being an
interpretation request and my feeling is that the end result will be "******",
which a user would naively expect and which matches the current implementation
of most compilers (cf. comment 0).

[Bug libfortran/50105] [4.6/4.7 Regression] I/O with g6.5 - wrong number of "**" shown

[burnus at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=50105

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
   Target Milestone|---                         |4.6.2
            Summary|Possibly: I/O with g6.5 -   |[4.6/4.7 Regression] I/O
                   |wrong number of "**" shown  |with g6.5 - wrong number of
                   |                            |"**" shown

--- Comment #7 from Tobias Burnus <burnus at gcc dot gnu.org> 2011-08-19 06:15:56 UTC ---
I find the analysis of Malcolm (as usual) convincing, he thinks the correct
result is "**    ".

Thus, I marked it again as regression - and removed the "possibly" from the
summary.

Malcolm wrote at  http://j3-fortran.org/pipermail/j3/2011-August/004594.html
| The wording is quite tricky, you need to read it very carefully.
| 
| It says (slight restructuring to make it clearer)
|   "On output,
|    (a) if an exponent exceeds its specified or implied width using the ...
|         G edit descriptor,
| or
|    (b) the number of characters produced exceeds the field width,
|        [you get stars]"
|
| The bit that explicitly mentions G ***only applies to the exponent width
| being exceeded***.
|
| That is not the case here.
|
| What about the number of characters produced?  Well, as Robert says, the
| F2.5 descriptor produces two characters (both stars) and the 4(' ')
| produces another 4, making a total of 6, within the width of G6.5.
|
| Perhaps this is merely bad wording, but it goes back a long way (F77).


And indeed not only gfortran < 4.6.0, g95, NAG and PGI but also g77 produces
"**     ".

(While with gfortran >= 4.6.0, ifort, crayftn/open64/pathf95/sunf95 produce
"******".)


(We might still see an interpretation request confirming Malcolm's
interpretation - or changing the standard such that "******" is correct; only
the future will tell ...)

[Bug libfortran/50105] [4.6/4.7 Regression] I/O with g6.5 - wrong number of "**" shown

[burnus at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=50105

--- Comment #8 from Tobias Burnus <burnus at gcc dot gnu.org> 2011-08-19 10:34:13 UTC ---
A simple (partial) revert of the patch of PR 47285
(http://gcc.gnu.org/ml/fortran/2011-01/msg00226.html) will lead to wrong
results as one does not honour the following.

  write(*,'(G5.5E5)') -10000.

will then print "*    " instead of the correct "*****" which is printed
currently. However, due to "w-(e+2) = 5-(5+2) > 0" [where the symbols are:
Gw.dEe], the whole field should be filled with asterisks, cf. below.


I think best is to do a full revert, replace the following hack in the patch
+  if (e > 4)\
+    e = 4;\

by a simply check for w-4 > 0 or w-(e+2) > 0 and - if failing - fill it with
"*" and return. Only when succeeding, move to the FMT_F. In the latter padding
section, one can then always pad with ' '. I have not yet checked whether the
  +  nb = nb >= w ? 0 : nb;\
is then still required or not.


>From the F2008 standard:

"if an exponent exceeds its specified or implied width using the ... G edit
descriptor," [the field should be filled with "*"].

One has for "Gw.d" n = 4 and for "Gw.d.Ee" it is "n = e+2". The requirement is
"w-n > 0". In that case, one needs to fill the whole width w (of "Gw.d") with
asterisks.

If "w-n > 0", one has "F(w-n).(dd),n(' ')" with some "dd" depending on rounding
and magnitude of the value. In that case, only (w-n) is - if needed - filled
with asterisks.

(For the calculations, see 10.7.5.2.2 (below the first table); see also
10.7.2.1 in comment 1 with the interpretation in comment 7.)

[Bug libfortran/50105] [4.6/4.7 Regression] I/O with g6.5 - wrong number of "**" shown

[burnus at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=50105

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
           Priority|P3                          |P5
           Severity|normal                      |minor

--- Comment #9 from Tobias Burnus <burnus at gcc dot gnu.org> 2011-08-19 18:14:46 UTC ---
For completeness: FORTRAN 77 had a similar wording at
  http://www.fortran.com/fortran/F77_std/rjcnf-13.html#sh-13.5.9
  http://www.fortran.com/fortran/F77_std/rjcnf-13.html#sh-13.5.9.2.3

Bill is not convinced about Malcolm's answer:
  http://j3-fortran.org/pipermail/j3/2011-August/004606.html
If I read it correctly, Bill thinks that one can read the standard such that
"**    " is correct, but thinks that the standard is ambiguous and also allows
"******" which - according to his count - most compilers do and a user would
expect.


I think we will have at the end an interpretation request - thus, one might
leave this item open until after the J3 meeting or better after the balloting
of the J3 meeting edits. The next meeting is October 10-14, 2011.

[Bug libfortran/50105] [4.6/4.7 Regression] I/O with g6.5 - wrong number of "**" shown

[jvdelisle at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=50105

--- Comment #10 from Jerry DeLisle <jvdelisle at gcc dot gnu.org> 2011-08-22 03:37:30 UTC ---
I just returned from travel and will have a look at this.

[Bug libfortran/50105] [4.6/4.7 Regression] I/O with g6.5 - wrong number of "**" shown

[jakub at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=50105

Jakub Jelinek <jakub at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
   Target Milestone|4.6.2                       |4.6.3

--- Comment #11 from Jakub Jelinek <jakub at gcc dot gnu.org> 2011-10-26 17:13:23 UTC ---
GCC 4.6.2 is being released.

[Bug libfortran/50105] [4.6/4.7 Regression] I/O with g6.5 - wrong number of "**" shown

[jvdelisle at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=50105

--- Comment #12 from Jerry DeLisle <jvdelisle at gcc dot gnu.org> 2012-01-10 00:40:55 UTC ---
Did any interpretation requests go in on this and did we get an answer back?

I am leaning toward current 4.7 is OK and 4.6 has the regression.

[Bug libfortran/50105] [4.6/4.7 Regression] I/O with g6.5 - wrong number of "**" shown

[burnus at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=50105

--- Comment #13 from Tobias Burnus <burnus at gcc dot gnu.org> 2012-01-10 08:45:21 UTC ---
(In reply to comment #12)
> Did any interpretation requests go in on this and did we get an answer back?

No, but I just wrote one:
  http://j3-fortran.org/pipermail/j3/2012-January/004976.html

The next J3 meeting will be #197, February 13-17, 2012, Las Vegas NV USA

The IR should pop up soon at:
  http://j3-fortran.org/doc/meeting/197/


> I am leaning toward current 4.7 is OK and 4.6 has the regression.

Sorry, that twists my mind. 4.6.0 *and* 4.7 produce "******". (While g77 and
GCC 4.1 to 4.5 produce "**    ".)

Any reason - in terms of the standard - why you "am leaning toward current
4.7"?

I believe that Bob Corbett and Malcolm Cohen are right and "**    " is the
correct output - even if more compilers have six asterisks:

"**    ": g77, gfortran < 4.6, g95, NAG f95, PGI
"******": gfortran >= 4.6.0, ifort, crayftn, open64, pathf95, Sun/Oracle f95,
IBM xlf

[Bug libfortran/50105] [4.6/4.7 Regression] I/O with g6.5 - wrong number of "**" shown

[jvdelisle at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=50105

--- Comment #14 from Jerry DeLisle <jvdelisle at gcc dot gnu.org> 2012-01-10 19:32:46 UTC ---
It twists my mind too, I meant 4.5, not 4.6, I don't have a working 4.6 here at
the moment.  Thanks for doing the interp request.  I hope this turns into a do
nothing solution.

[Bug libfortran/50105] [4.6/4.7 Regression] I/O with g6.5 - wrong number of "**" shown

[burnus at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=50105

--- Comment #15 from Tobias Burnus <burnus at gcc dot gnu.org> 2012-01-12 09:07:02 UTC ---
(In reply to comment #13)
> (In reply to comment #12)
> > Did any interpretation requests go in on this and did we get an answer back?

Thanks to Van and Dan, it's now on the server for the #197 meeting as 12-102;
cf. http://j3-fortran.org/doc/meeting/197/12-102.txt / check for updates
(12-102, 12-102r1 etc.) also at http://j3-fortran.org/doc/year/12/

[Bug libfortran/50105] [4.6/4.7 Regression] I/O with g6.5 - wrong number of "**" shown

[tkoenig at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=50105

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|UNCONFIRMED                 |WAITING
   Last reconfirmed|                            |2012-02-03
                 CC|                            |tkoenig at gcc dot gnu.org
     Ever Confirmed|0                           |1

--- Comment #16 from Thomas Koenig <tkoenig at gcc dot gnu.org> 2012-02-03 19:21:15 UTC ---
Let's make put this on WAITING until the IR is finished.

[Bug libfortran/50105] [4.6/4.7 Regression] I/O with g6.5 - wrong number of "**" shown

[burnus at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=50105

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

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

--- Comment #17 from Tobias Burnus <burnus at gcc dot gnu.org> 2012-02-18 08:06:04 UTC ---
Revised IR after J3 meeting
  http://j3-fortran.org/doc/meeting/197/12-102r1.txt
namely:

ANSWER:

10.7.2.3.2 says
  "the field occupies w positions, the fractional part of which
   consists of d digits"
and this does not make sense if w<d.  Therefore, according to the
first sentence of 1.5 Conformance, the program is not standard
conforming.

There are therefore no requirements on the processor for this
program, and it may produce either output, or even a runtime error.

 * * *

We should be able to diagnose this at compile time.

 * * *

Disclaimer: WG5 might still revise the J3 answer; there will be also an J3
email voting on the IR.

[Bug libfortran/50105] [4.6/4.7 Regression] I/O with g6.5 - wrong number of "**" shown

[dominiq at lps dot ens.fr]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=50105

--- Comment #18 from Dominique d'Humieres <dominiq at lps dot ens.fr> 2012-02-18 17:03:18 UTC ---
> There are therefore no requirements on the processor for this
> program, and it may produce either output, or even a runtime error.

If accepted, this means that it is not a regression. IMO any amount of * is
better than a runtime error (the code may output important data after it).

[Bug libfortran/50105] [4.6/4.7 Regression] I/O with g6.5 - wrong number of "**" shown

[jakub at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=50105

Jakub Jelinek <jakub at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
   Target Milestone|4.6.3                       |4.6.4

--- Comment #19 from Jakub Jelinek <jakub at gcc dot gnu.org> 2012-03-01 14:38:12 UTC ---
GCC 4.6.3 is being released.

[Bug libfortran/50105] [4.6/4.7/4.8 Regression] I/O with g6.5 - wrong number of "**" shown

[burnus at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=50105

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

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

--- Comment #20 from Tobias Burnus <burnus at gcc dot gnu.org> 2013-01-02 17:24:12 UTC ---
(In reply to comment #17)
> 10.7.2.3.2 says
>   "the field occupies w positions, the fractional part of which
>    consists of d digits"
> and this does not make sense if w<d.  Therefore, according to the
> first sentence of 1.5 Conformance, the program is not standard
> conforming.
> 
> There are therefore no requirements on the processor for this
> program, and it may produce either output, or even a runtime error.


> Disclaimer: WG5 might still revise the J3 answer; there will be also an J3
> email voting on the IR.

Latest result version ftp://ftp.nag.co.uk/sc22wg5/N1901-N1950/N1932.txt (search
for F08/0063)

Result was: YES WITH COMMENTS, with 7 YES votes, 1 YES WITH COMMENT and 1 NO
vote (or two? At the top it lists one, below it has two comments claiming a
"NO".) ftp://ftp.nag.co.uk/sc22wg5/N1901-N1950/N1944.txt

Additionally, it is listed in the Corrigendum 2 draft as being contained in the
document, though no edits have been done:
ftp://ftp.nag.co.uk/sc22wg5/N1901-N1950/N1948.pdf

[Bug libfortran/50105] [4.6/4.7/4.8 Regression] I/O with g6.5 - wrong number of "**" shown

[jvdelisle at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=50105

--- Comment #21 from Jerry DeLisle <jvdelisle at gcc dot gnu.org> 2013-01-02 20:46:02 UTC ---
Based on the interp, current gfortran is acceptable and we can close this PR.

[Bug libfortran/50105] [4.6/4.7/4.8 Regression] I/O with g6.5 - wrong number of "**" shown

[anlauf at gmx dot de]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=50105

Harald Anlauf <anlauf at gmx dot de> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |anlauf at gmx dot de

--- Comment #22 from Harald Anlauf <anlauf at gmx dot de> 2013-01-02 21:40:03 UTC ---
(In reply to comment #17)
> We should be able to diagnose this at compile time.

It might make sense to link PR45129 (and possibly PR28397)
to this one as part of a test suite.

[Bug libfortran/50105] [4.6/4.7/4.8 Regression] I/O with g6.5 - wrong number of "**" shown

[burnus at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=50105

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

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

--- Comment #23 from Tobias Burnus <burnus at gcc dot gnu.org> 2013-02-15 07:39:21 UTC ---
(In reply to comment #22)
> It might make sense to link PR45129 (and possibly PR28397)
> to this one as part of a test suite.

I did so now – and closed this bug as INVALID.