From LJM@SLACVM.BITNET Wed Nov 17 06:07:00 1993 Received: from vm.uni-c.dk by dkuug.dk with SMTP id AA13175 (5.65c8/IDA-1.4.4j for ); Wed, 17 Nov 1993 23:07:24 +0100 Message-Id: <199311172207.AA13175@dkuug.dk> Received: from vm.uni-c.dk by vm.uni-c.dk (IBM VM SMTP V2R2) with BSMTP id 2117; Wed, 17 Nov 93 23:08:23 DNT Received: from SLACVM.SLAC.STANFORD.EDU by vm.uni-c.dk (Mailer R2.07) with BSMTP id 5305; Wed, 17 Nov 93 23:08:22 DNT Received: by SLACVM (Mailer R2.08 R208004) id 9523; Wed, 17 Nov 93 14:07:48 PST Date: Wed, 17 Nov 1993 14:07 -0800 (PST) From: "Len Moss" To: "SC22/WG5 Mailing List" Subject: Trip Report on X3J3 meeting, 8 - 12 Nov, 1993 X-Charset: ASCII X-Char-Esc: 29 S L A C M E M O R A N D U M November 17, 1993 ____________________________ To: Interested FORTRAN users From: L. Moss Subject: Trip Report on X3J3 meeting, 8 - 12 Nov, 1993 _________________________________________________________________ Note: This is a personal report of these meetings and in no sense does it constitute an official record. X3J3 met in Albuquerque, New Mexico from 8 through 12 November 1993. The US TAG (Technical Advisory Group) for Fortran (which consists of US members of X3J3) held a brief session on Thursday afternoon, 11 November, which I will report on separately. The principal goals for this meeting were to: * begin work on the preliminary list of Fortran 95 requirements contained in Resolution B9 from last summer's WG5 meeting; * continue processing Fortran 90 defect reports; * review additional items in X3J3's Journal of Requirements (JOR) for possible submission to WG5; and, * review the editorial process for producing Fortran 95. The first two goals were accorded higher priority than the last two. X3J3 also heard presentations from three visitors on various topics. B9 PROPOSALS ____________ ISO/IEC JTC1/SC22/WG5 met in Berchtesgaden, Germany last July and began the process of selecting requirements for Fortran 95. Their preliminary selection is contained in resolution B9, which consists of 3 separate lists: A Items WG5 definitely intends to be in Fortran 95: the corrigenda and edits generated from the defect reports for Fortran 90, plus all or most of the features needed to make any HPF-compliant program also Fortran 95 compliant. B Desirable features for Fortran 95: initialization of pointers and, possibly, derived type components; removal of conflicts with IEEE arithmetic; allocatable components in derived types; deletion of some obsolescent features and some additions to the obsolescent list; and some minor additions to formatted I/O. C Areas requiring further investigation by X3J3 so that a decision on final requirements can be made by the summer 1994 WG5 meeting: 1 a CPU-time intrinsic function; 2 KIND parameters for derived types; 3 allowing some classes of user-defined functions in declarations; 4 full support for IEEE arithmetic; 5 exception handling; 6 object oriented programming; and, 7 derived type I/O. Tutorials were presented on all of these topics except the corrigenda and edits from Fortran 90 and C.7 (at meeting 126, immediately following the WG5 meeting, X3J3 determined that it did not have the resources to pursue this last item, and recommended to WG5 that it be deferred to Fortran 2000). The A and B items will all be followed up with draft proposals at the February meeting. Straw votes at this meeting indicated that X3J3 felt it did not have the resources to address items C.3, C.4, and C.5 for Fortran 95. Recommendations to defer these items until Fortran 2000 will be voted on at the February meeting. However, the /parallel subgroup volunteered to research C.3 further, with a view to providing at least the minimum functionality needed to support HPF. X3J3 was also doubtful about C.2, but will continue work on this item for now and would like to see another tutorial and a draft proposal at the next meeting. It was noted that a multi-thread definition of a CPU-time system call is due out from the POSIX community by the time of our next meeting, so action on C.1 will be deferred until then. Finally, another tutorial on C.6 was requested for the February meeting. FORTRAN 90 MAINTENANCE ______________________ Beginning at this meeting, processing of defect reports was consolidated under a single /interpretations subgroup disjoint (more or less) from the subgroups working on future developments. At the end of the meeting, responses to about 18 defect reports had been approved. These will be voted on in a letter ballot following the meeting. A backlog of about 35 open defect reports remains. REVIEW OF JOR ITEMS ___________________ The following items in X3J3's JOR have been approved by X3J3 and are awaiting action by the US TAG before being forwarded to WG5 (requirements newly approved at this meeting are flagged with an asterisk, and the X3J3-recommended target revision cycled, if any, has been indicated in parentheses): 004 (F95) FORALL 010* Nesting of Internal Procedures 012 (F2K) Condition Handling 015 (F95) Conditional Compilation 016 (F95) Command Line Arguments and Environmental Variables 017 (F2K) Bit Data Type, String 018 (F2K) Controlling Pointer Bounds 031* (F95) Varying Length Character with Declared Maximum X3J3 also voted to archive JOR items 026 (Directives) and 029 (Interline Optimization). Finally, the /JOR subgroup noted that a number of items in the JOR are related in one way or another to the operating system, and suggested that the charter of the /POSIX subgroup be broadened somewhat to include such requirements. The chair agreed, and the yet-to-be-renamed subgroup was assigned responsibility for the following items: 016 (Command Line Arguments and Environmental Variables) 032 (POSIX Binding to Fortran) 039 (SYSTEM Command) 040 (Obtain the Command Line) 041 (Obtain Program Startup Command) 047 (File Sharing Modes) EDITORIAL PROCESS _________________ A working document, including all the WG5 corrigenda plus a few additional minor edits, has been prepared and was proofread by the /editorial subgroup. This document will be available to members in FrameMaker, PostScript, and straight ASCII, but has not yet been approved as a new base document by the committee (so any text proposals should continue to reference Fortran 90, taking note of the edits in the approved corrigenda). A new procedural rule was approved at this meeting calling for every proposal to include a rationale section suitable for inclusion as a note in the text of the standard. A number of global edits proposed by the /edit subgroup were approved by the full committee. In additional, several significant reorganizations of the document were also approved in principle (pending proposals with text), including the following: * Move sections 12.1.2.2.1 and 12.1.2.2.2 on host association to 14.6.1.2. * Gather together the text on actual and dummy arguments. * Correct the definition of tokens. * Move Annex C to footnotes. ADDITIONAL PRESENTATIONS ________________________ Jim Giles of Los Alamos gave a presentation on what he liked and disliked about Fortran 90. Among the features he felt were still missing from Fortran were REALLOCATE; variant records; some sort of standard macro (i.e., preprocessor) language, preferably one that worked well (better than cpp) across multiple programming languages, enumerated data types, exception handling, and conformance requirements in the presence of extensions. I found one of Mr. Giles' observations both new and interesting (though it's probably well known among numerical analysts): he mentioned that it is the speed of processors which is driving the need for greater precision. In other words, as it becomes possible to do bigger problems, the effect of cumulative round-off errors becomes significant. He also felt that this problem could be addressed either by giving the users more control over arithmetic (e.g., rounding modes in IEEE arithmetic) or by building hardware with ever wider floating point formats. David Epstein from IBM/Santa Teresa gave a tutorial on a simple conditional compilation facility for Fortran, called CCF. CCF does not support macro, or even simple text, substitution (though it is careful not to block future extensions in those directions). It also can be used in a mode where it simply transforms the original file, thus avoiding the need to maintain separate input and output files (and also preserving line numbers for debuggers). John Sharp of Sandia Labs, the new chair of X3T2, gave a presentation on a Draft International Standard (DIS) which is currently out for international voting: "ISO/IEC 10967-1 Information Technology - Language Independent Arithmetic - Part 1: Integer and Floating Point Arithmetic" (LIA-1). X3T2 is the US TAG on this DIS and is currently conducting a US public review which will end on 11 January 1994. X3J3 is a coordinating liaison with X3T2, and is therefore responsible for reviewing LIA-1's implications for Fortran and communicating its findings to X3T2. Unlike IEEE 754/854, LIA-1 does not push for any one particular type of floating point arithmetic, but mainly requires implementations to rigorously define how their arithmetic behaves, i.e., it is primarily a descriptive rather than prescriptive standard. Nevertheless, it does set some minimum requirements on arithmetic behavior. These requirements permit a very broad spectrum of implementations covering nearly all current hardware, but they do exclude one popular supercomputer (Cray). Opinions on LIA-1 within X3J3 appear to be somewhat divided. A number of members feel that, given the growing acceptance of IEEE arithmetic within the industry, the value of a descriptive standard is very low. In addition, the exclusion of Crays bothers some (it should be noted that, though the latest version of LIA-1 still does not cover Crays, it is possible to write programs that are portable between Cray platforms and strict LIA-1 platforms). Others within the committee do not believe that IEEE floating point will ever completely supplant proprietary forms of arithmetic, and feel that LIA-1 at least provides a common basis for developing programs with predictable numeric properties. They also point to LIA-1's requirements on integer arithmetic, which is not covered at all by the IEEE standards. In addition to the technical issues, several members felt that there was a procedural problem in that X3T2 has been somewhat late in returning to X3J3 the results of the previous public review (which ended last January) and thus has left us insufficient time to respond before next January's deadline. Action on LIA-1 was deferred to the (Fortran) US TAG meeting, since this is part of forming the US position on an international issue. ADMINISTRATIVE BUSINESS _______________________ MEMBERSHIP At the beginning of this meeting X3J3 had 28 members, giving a quorum of 8 (=1+INT(Members/3)), and a majority of the membership of 15 (=1+INT(Members/2)). At the end of the meeting, no members had been lost (though one had announced informally that he would probably have to resign) and three visitors expressed interest in joining the committee. FUTURE MEETINGS: 128th 7-11 February 1994, Palmdale, CA (host: Richard Maine). 129th 2-6 May 1994, Lake Tahoe, CA (host: Tom Lahey). 1994 WG5 Meeting 8-12 August 1994, Edinburgh, Scotland (host: David Muxworthy). 130th 15-19 August 1994, Edinburgh, Scotland (host: Geoff Millard). NEXT DISTRIBUTION The closing date for the next pre-meeting distribution is 3 January 1994. To get an item into the distribution it should be received before this date by: David Mattoon Assoc. of American Railroads 3140 South Federal Street Chicago, IL 60616 Phone: 312-808-5868 FAX: 312-808-5425 Email: David_Mattoon_at_CTC@relay.aar.com -- Leonard J. Moss o/ My views don't necessarily Stanford Linear Accelerator Center, MS 97 o/ reflect those of SLAC, P.O. Box 4349; Stanford, CA 94309 o/ Stanford or the DOE