Re: [PATCH] fortran/28213 -- Fix Hollerith in IO list

[Feng Wang <wf_cs at yahoo dot com>]

--- Steve Kargl <sgk@troutmask.apl.washington.edu> wrote:

> On Tue, Jul 04, 2006 at 08:21:54AM -0700, Steve Kargl wrote:
> > On Tue, Jul 04, 2006 at 10:26:55PM +0800, Feng Wang wrote:
> >> 
> >> --- Jerry DeLisle <jvdelisle@verizon.net>????:
> >> 
> >>> Steve Kargl wrote:
> >> 
> > 
> >>From Fortran 66, 5.1.1.6, page 10.
> > 
> >   "This type of constant may be written only in the argument list
> >    of a CALL statement and in the data initialization statement."
> > 
> > There is an ambiguity with a Hollerith constant in an output statement.
> > 
> >       program z
> >       data i/3horz/
> >       data x/3horz/
> >       print *, i, x
> > c What value should the following list-directed print statement print?
> > c     print *, 3horz
> >       end
> > 
> > Furthermore, see section A2, "Conflicts with ANSI X3.9-1966,"
> > of the Fortran 77 standard.  Item (3) says "Hollerith constants
> > and Hollerith data are not permitted in this standard."  However,
> > Appendix C, address the use of Hollerith in a Fortran 77 
> > program.  In particular, section C6 address IO with the A edit
> > descriptor, but my interpretation (which is sometimes known to
> > be too strict and/or wrong) is that a "print '(A3)', 3horz' is
> > not permitted.

Hollerith constant is a more compiler implementation dependency feature. The
legacy specification is less meaningful than the practice codes. This typeless
literal constant was created to be used as character. So I prefer to accept it
to be printed as character type.

> > 
> > Finally, the behavior I implemented is consistent with NAG's
> > Fortran 95 compiler.

ifort is consistent with mine :-)

> atlas:kargl[205] cat z.f
>       program z
>       data i/3horz/
>       data x/3horz/
>       print *, i, x
>       print *, 3horz
>       print '(A3)', i
>       end

[wf@WFPC bug]$ ifort z.f
[wf@WFPC bug]$ ./a.out 
   544895599  2.1213687E-19
 orz
orz

And IBM XL Fortran V10.1 online documents say:
(http://publib.boulder.ibm.com/infocenter/lnxpcomp/v8v101/index.jsp?topic=/com.ibm.xlf101l.doc/xlflr/typeless_lit_constants.htm)

When you use a typeless constant in any other context, the constant assumes the
default integer type, with the exception of Hollerith constants. Hollerith
constants assume a character data type when used in the following situations: 
1. An H edit descriptor 
2. A relational operation with both operands being Hollerith constants 
3. An input/output list
 
> 
> Sigh.  g77 does something completely different from your patch and
> from mine.
> 
> atlas:kargl[203] g77 -o z z.f
> atlas:kargl[204] ./z
>  544895599  2.12136866E-19
>  544895599
> orz
> 
> g77 converts the Hollerith constant to an integer in list-directed
> output.  This means we probably need to implement the g77 behavior
> for backwards compatibility.
> 

The patch enclosed implements g77 behavior, but I still think it is better to
print as character type.



Best Regards,
Feng Wang

--
Creative Compiler Research Group,
National University of Defense Technology, China.


	

	
		
___________________________________________________________ 
Mp3疯狂搜-新歌热歌高速下
http://music.yahoo.com.cn/?source=mail_mailbox_footer

Index: io.c
===================================================================
--- io.c	£¨ÐÞ¶©°æ 115177£©
+++ io.c	£¨¹¤×÷¿½±´£©
@@ -26,6 +26,7 @@ Software Foundation, 51 Franklin Street,
 #include "gfortran.h"
 #include "match.h"
 #include "parse.h"
+#include "intrinsic.h"
 
 gfc_st_label format_asterisk =
   {0, NULL, NULL, -1, ST_LABEL_FORMAT, ST_LABEL_FORMAT, NULL,
@@ -2132,6 +2133,11 @@ match_io_element (io_kind k, gfc_code **
       gfc_free_expr (expr);
       return MATCH_ERROR;
     }
+  if ((k == M_WRITE || k == M_PRINT) && expr->expr_type == EXPR_CONSTANT
+	&& expr->ts.type == BT_HOLLERITH)
+    {
+      expr = gfc_convert_constant (expr, BT_INTEGER, gfc_default_integer_kind);
+    }
 
   cp = gfc_get_code ();
   cp->op = EXEC_TRANSFER;