From J.Reid@letterbox.rl.ac.uk Wed Mar 26 19:14:15 1997 Received: from letterbox.rl.ac.uk (letterbox.rl.ac.uk [130.246.8.100]) by dkuug.dk (8.6.12/8.6.12) with SMTP id TAA21533 for ; Wed, 26 Mar 1997 19:14:09 +0100 Received: from jkr.cc.rl.ac.uk by letterbox.rl.ac.uk with SMTP (PP) id ; Wed, 26 Mar 1997 18:14:02 +0000 Received: by jkr.cc.rl.ac.uk (4.1/SMI-4.1) id AA01300; Wed, 26 Mar 97 18:14:00 GMT Date: Wed, 26 Mar 97 18:14:00 GMT From: jkr@letterbox.rl.ac.uk (John Reid) Message-Id: <9703261814.AA01300@jkr.cc.rl.ac.uk> To: SC22WG5@dkuug.dk Subject: Fortran Conditional Compilation I would like to tell you all that David Epstein, Dick Hendrickson and I have formed a volunteer editorial subgroup to polish the coco document. We appreciate that this may be wasted effort, but of course hope that WG5 will not decide to terminate the project. Note that the WG5 ballot date has been changed to 20 April at the request of DIN. I hope that our doing this work will encourage countries to vote YES. I am doing the typesetting and am putting the latest draft on my server. Please have a look: jkr.cc.rl.ac.uk (130.246.8.20) in the file pub/wg5/coco.ps We are still making changes, but the paper already looks much better. I am also putting a (less readable) text version in the same directory in the file coco.text. We are responding to the pre-LV ballot comments. The main changes that flow from them are: 1. The description is now independent of part 1. This has only lengthened the document slightly and is much more satisfactory since in most cases coco requires only a simpified version of the equivalent Fortran feature. 2. The model is now of an independent process that yields a source program for a Fortran processor. Of course, it is expected that implementations will usually integrate the two processes and only optionally provide the coco output code (Fortran source code). 3. SET is now standardized as a separate SET file that is coco program that consists entirely of coco type declaration directives and coco comment lines. Beyond this, we have made many editorial changes. I attach what we now say in the Rationale. It sums up my reasons for working on this. Best wishes, John. ........................... Programmers often need to maintain several versions of code to allow for different hardware and different applications. Keeping several copies of the source code is error prone. It is far better to maintain a master code from which any of the versions may be selected. This facility has deliberately been kept very simple. The additional lines inserted to control the process and all the lines that are not selected are omitted from the output or are replaced by comments. Those that are selected are copied to the output completely unchanged. Examples of the need for such a facility are: (1) Parametrized types do not solve all the problems associated with different precisions. For example, if the Basic Linear Algebra Subroutines (BLAS) are in use, the names are different for the single and double precision versions. (2) A version of a code for complex arithmetic may differ little from the version for real arithmetic. (3) The relative efficiency of different constructions may vary from processor to processor. While array syntax may be needed on a massively parallel machines, do loops may be more efficient on a scalar processor. Some may benefit from special comments. (4) Versions may be required for different message-passing libraries. (5) Additional print statements may be inserted into a program when under development. It may be very helpful have these readily available in case some unexpected results are found in production use. (6) Versions may be required for character constants being in different languages (internationalization). (7) If INCLUDE is in use, the file naming convention varies between systems. Some of these cases may be addressed within the Fortran code itself by run-time tests, but this will result in a large object code and some run-time overhead. Most of them, however, can only be solved without conditional compilation by keeping separate versions.