Re: fortran/10197: direct acces files not unformatted by default

[Tim Prince <timothyprince@sbcglobal.net>]

The following reply was made to PR fortran/10197; it has been noted by GNATS.

From: Tim Prince <timothyprince@sbcglobal.net>
To: bdavis9659@comcast.net, gcc-gnats@gcc.gnu.org
Cc:  
Subject: Re: fortran/10197: direct acces files not unformatted by default
Date: Mon, 24 Mar 2003 06:15:26 -0800

 On Sunday 23 March 2003 16:34, bdavis9659@comcast.net wrote:
 > >Number:         10197
 > >Category:       fortran
 > >Synopsis:       direct acces files not unformatted by default
 
 > OPEN(..ACCESS='DIRECT'..) should open a file with FORM='UNFORMATTED'.
 >
 > It is being opened with FORM='FORMATTED', which causes a runtime error on
 > the first unformatted I/O.
 >
 > This is correct on g77 2.96, and not on the pre-release g77 3.3 or g77 3.4.
 >
 > At the risk of being proven wrong by a plethora of language lawyers:):)
 >
 > The FORTAN-77 standard, section 12.10.1 states; when discussing FORM=
 >
 > "If this specifier is omitted, a value of UNFORMATTED is assumed if the
 > file is being connected for direct access, and a value of FORMATTED is
 > assumed if the file is being connected for sequential access. "
 >
 > So, IMHO, this is how it should work.
 >
 > >How-To-Repeat:
 >
 > C demonstrate a direct access file is by default unformatted
 >        IMPLICIT NONE
 >        LOGICAL*4 ERROR /.FALSE./
 >        CHARACTER*12 FORM
 >        DATA FORM   / ''        /
 >        OPEN(UNIT=60,
 >      1      ACCESS='DIRECT',
 >      2      STATUS='SCRATCH',
 >      3      RECL=255)
 >        INQUIRE(UNIT=60,FORM=FORM)
 >        IF (FORM.EQ.'UNFORMATTED') THEN
 >           PRINT*,'FORM IS ',FORM,' WHICH IS CORRECT.'
 >        ELSE
 >           PRINT*,'FORM IS ',FORM,' S/B UNFORMATTED.'
 >           ERROR = .TRUE.
 >        ENDIF
 >        CLOSE(UNIT=60)
 >        IF (ERROR) CALL ABORT
 >        END
 >
 I "sent mail to interested parties," but, AFAICT, all the interested parties 
 selected by gnats-web bounced my mail.  I pointed out that the version of 
 open.c included in gcc-3.2.2 implements the standard, in accordance with 
 Bud's test.  I rebuilt gcc-3.3 with that version of open.c, and it passed 
 Bud's test.  So, gcc-3.3 has a regression from gcc-3.2.2, apparently caused 
 by an out of date libf2c.
 -- 
 Tim Prince