From Craig.Dedo@mixcom.com Wed Apr 3 15:46:16 1996 Received: from mixcom.mixcom.com (mixcom.mixcom.com [198.137.186.100]) by dkuug.dk (8.6.12/8.6.12) with ESMTP id PAA29857 for ; Wed, 3 Apr 1996 15:46:06 +0200 Received: from 156.46.43.40 by mixcom.mixcom.com (8.6.12/2.2) id HAA26952; Wed, 3 Apr 1996 07:44:40 -0600 Message-Id: <199604031344.HAA26952@mixcom.mixcom.com> MIME-Version: 1.0 Content-Type: text/plain Content-Transfer-Encoding: 7bit Date: Wed, 03 Apr 96 07:43:08 -0600 From: Craig Dedo Subject: Default Data Type - 2 Different Definitions? To: WG5 Mailing List X-Mailer: SPRY Mail Version: 04.00.06.17 Dear Members of the WG5 Mailing List: Here is an item that I am throwing out for discussion. While thinking about the problem that Keith Bierman mentioned in his January 29 e-mail titled "Fodder for Meeting" (x3j3.1996-37), I realized that we may have a problem with the term "default". In using the concept of "default" kind of data type, we MAY actually be using two differing and incompatible definitions. 1. The "default" kind for a data type is the kind of the data type which is the most frequent and suitable for the application which the programmer is writing. 2. The "default" kind for a data type is the kind of data type which is most natural for the hardware architecture which the Fortran compiler is compiling source code for. These two definitions need not be the same. Many Fortran compilers offer a compiler option which allows the user to choose the default kind of a data type. The user usually chooses a kind type which occurs most frequently in the application. However, this is not necessarily the kind which is most natural for the hardware. For example, a user may encounter only a few large integers in the problem but many small ones. Such a user would choose 2 byte integers as the default integer kind. However, if the processor is 32 bits, it naturally processes data 4 bytes at a time. The Fortran compiler for such a processor may implicitly or explicitly consider 4 byte integers to be the default. Thus, there is an incompatibility between what the user specifies and the Fortran compiler assumes. I believe that the conflict in these two definitions is at the heart of the problems that we experience when we specify that certain data objects in Fortran should be of a default kind. Therefore, in our work on Fortran 2000, we should clearly define both usages of the concept of default kind and rigorously distinguish which one we are using in any given situation. What do others think? ---------- Sincerely, Craig T. Dedo Internet: Craig.Dedo@mixcom.com Elmbrook Computer Services Voice Phone: (414) 783-5869 17130 W. Burleigh Place Brookfield, WI 53005 Disclaimer: These opinions are mine alone. USA They do NOT represent any organization. "They that can give up essential liberty to obtain a little temporary safety deserve neither liberty nor safety." -- Benjamin Franklin (1759)