From willemw@komp.ace.nl Thu Oct 5 12:07:44 1995 Received: from komp.ace.nl (komp.ace.nl [193.78.104.90]) by dkuug.dk (8.6.12/8.6.12) with SMTP id MAA27285 for ; Thu, 5 Oct 1995 12:07:44 +0100 Received: by komp.ace.nl with SMTP id AA31360 (1.10/2.17); Thu, 5 Oct 95 11:59:44 +0200 (MET) To: sc22wg11@dkuug.dk Subject: WG11/N419: report on LIA-2 discussion (part 1) Reply-To: Willem Wakker Date: Thu, 05 Oct 95 11:59:31 N Message-Id: <31290.812887171@komp> From: Willem Wakker ISO/IEC JTC1 SC22 WG11/N419 - part 1 Meeting at NIST, Gaithersburg, MD, USA, 25-Sep-1995 to 27-May-1995 Notes on the discussions of SC22/WG11 document N416, editor's draft of LIA-2, dated 6-Sep-1995. Notes taken by: Kevin Harris, Digital Equipment Corp additional comments added by the note-taker after the meeting are introduced by the notation: "(ed ....)" Additional documents forming the basis for the discussions: 1) "Cover Letter for SC22/WG11 N 416", by Mary Payne, dated 4-Sep-1995 2) Marked up copy of SC22/WG11 N 416, from Roger Scowen 3) Editor's written proposals carried to the meeting: 16-Sep-1995: Draft Specs for EXPM1 and LN1P 20-Sep-1995: "Two Proposals for Changes to LIA-2" 1) Removal of clause 14 and its subclause 2) Removal of clause 19.2 20-Sep-1995: "Another Proposal for Changes to LIA-2" Removal of the "max_arg_PROC and "argument_too_big" concepts 20-Sep-1995: "Yet Another Proposal for changes to LIA-2" Limitation of the accuracy requirements for 2 argument trigonometric functions. 20-Sep-1995: 2 messages on the specification style of the SQRT functions 4) Email message from Roger Scowen, 8-Sep-1995, "LIA-2 Cover Letter Response" 5) Email message from John Diamond, 15-Sep-1995, "LIA-2", opinions on topics raised in the cover letter. 6) Email messages from Kent Karlsson: SC22WG11.570, 21-Sep-1995: "Some (many) discussion points for the WG11 late September meeting" which references to 4 earlier messages: 4-Aug-1995: "Suggested section on LIA-2 integer operations, with "new" semantics for those integer operations." 8-Aug-1995: "Suggested section on LIA-2 basic floating point operation s, with "new" semantics for those floating point operations." 21-Aug-1995: "Part 1 of suggested LIA-2 section on elementary transcendental f.p. operations, with "new" semantics for those floating point operations." 8-Sep-1995: "Part 2 of suggested LIA-2 section on elementary transcendental f.p. operations, with "new" semantics for those floating point operations." 25-Sep-1995: "Re: LIA-2 at the next meeting" 25-Sep-1995: "Re: Kent's comments on minimum and maximum procedures" 25-Sep-1995: "Re: Corrected trig comments" Meeting minutes: Attendence: Willem Wakker, Netherlands Brian Meek, UK Kevin Harris, USA representing LIA-2 editors Mary Payne, Craig Schaffert Ed Barkmeyer, NIST, USA, host Hiko Kakehi, Japan Dan Lozier, NIST, USA Paul Rabin, OSF, USA Reports regarding LIA: From the SC22 Plenary at Annapolis, 18-22 Sep 1995: Resolution AH: Authorization for concurrent CD registration and ballot for WG11 documents ISO/IEC JTC1/SC22 authorizes the SC22 secretariat to conduct a concurrent CD registration and CD ballot for project 22.33 - Language Independent Arithmetic, Part 2: Mathematical Procedures. Resolution AI: Change of titles for WG11 projects: ISO/IEC JTC1/SC22 approves the change of titles for parts 2 and 3 of the Language Independent Arithmetic Standard (ISO/IEC 10967): - part 2 (project 22.33), new title: Elementary Numerical Functions - part 3 (project 22.34), new title: Complex Floating Point Arithmetic and Complex Elementary Numerical Functions Resolution ZB: Appreciation: general ISO/IEC JTC1/SC22 expresses its appreciation to all those who have assisted in the work of this meeting, and in particular . . . to Mr. Kevin Harris for his informative presentation on Language Independent Arithmetic; . . . Liaison Report from UK: Happy to see LIA go to CD registration without delay - no further meetings needed. Liaison Report from USA: No WG11 liaison specific activity. X3T2 happy with current progress of Language Independent work. Discussions regarding LIA-2: Monday, 25-Jan-1995 Harris: Presents summary of the changes to the LIA-2 draft since the last meeting and the current issues: Editor's issues: 1) Clause 2.1; Should we retain the concept of "Conformity to the whole"? 2) Clauses 4.1 and A.4.1; Duplication of definitions issues 3) Clause 7; Modify the Description for SQRT_F, use approximation style 4) Clauses 8 and 9; Inclusion of EXPM1 and LN1P functions 5) Clauses 10 and 11; Modify accuracy requirements for 2 argument trigonometric functions 6) Clauses 10 and 11; Removal of the "argument too big" concept 7) Clause 14; Removal of conversions between Integer Signed and Unsigned forms 8) Clauses 16.5-16.8; Removal of Max/Min with conversions 9) Clause 19.2; Remove POLY Scowen's general comments: 1) Congratulations on a well constructed draft 2) Rationale section still incomplete in many areas 3) Where text is taken verbatim from LIA-1, please indicate in (draft) forms as an Editor's note. This will allow reviewers to concentrate on new LIA-2 text, and limit debate on previously resolved issues. 4) Clauses 7.1 and later: Move all requirements concerning extensions to an annex. Meek: Clarification: There is a problem with conformance rules regarding IEC 559 conforming implementations. Need to make it clear that if IEC 559 conforming implementations support extension rules for any supplied functions, they must support the extension rules for all supplied functions. Extension support is an all-or-nothing support issue. Harris: Volunteers to make a detailed list of Scowen comments for the next day's discussion. Karlsson's general comments: Email message (SC22WG11.570) "Some (many) discussion points for the WG11 late september" from 21-Sep-1995 Barkmeyer: Distributes printed copies (SC22 plenary attendees were travelling and couldn't receive email) and WG11 attendees review and discuss: Harris: Summarizes editor's issues with the Document structure (section 2 of SC22WG11.570) and Semantic style and notational style (section 3 of SC22WG11.570) proposals: 1) The number of changes required is very large. 2) Even assuming WG11 agreed with all these proposals, the document changes implied would require extensive editing, requiring an estimated 1-2 years to reach the level of confidence that we now have in the current document. In particular, the fact that the specified semantics of each procedure are potentially changed as well as the format of their description, implies that a lengthy review and comparison will need to be made for each function. 3) Publication timeliness is important, for example, the WG14 meetings (next C standard) are considering arithmetic conformance and LIA bindings issues now. 4) Agreement is strongly in doubt on many of these proposals. Many have been raised previously and rejected by consensus of attendees at prevous meetings. Many make unwarranted assumptions and claims. 5) The SC22WG11.570 document points out several technical problems and weaknesses in the current draft. These are acknowledged and can be addressed without radically changing either the format of the specifications or the organizations of the document. 6) Almost all the proposals in (SC22WG11.570) are specified as "make or break", i.e. Karlsson will vote no on forwarding the document unless these changes are made in his suggested fashion. It is unlikely that the editors and WG11 generally will agree to his view on all these points. Thus, based on the current views expressed in SC22WG11.570 and its supporting documents, it is not likely that Karlsson would support the progression of any document that the editors and the balance of the WG11 members could agree on. 7) The intended audiences for the LIA family of standards are hardware and software vendors, and other standards writers. Strict adherence to existing practice in notation used by the refereed mathematical publishing world and their readers is not required. Instead, the virtues of clarity, internal consistency, and practicality of implementation are emphasized. Thus, the editors suggest that major specification style changes and document organization changes proposed in sections 2 and 3 of SC22WG11.570 and its supporting documents not be considered further by WG11. Wakker: Supports the position of the editors. There are a variety of time pressures that must be considered Wants to accept the major concerns of Kent that can be addressed without significantly delaying CD registration Has not received any negative comments from SC22 national representatives or WG convenors concerning LIA-2 Would prefer to basically continue in the current mode with current document structure and formatting Kakehi: Supports Wakker. Most important issues are: 1) Technical correctness 2) Timeliness - especially with imminent LID and LIPC publication Meek, for UK: Happy with overall document structure. Repeats request that text replicated from LIA-1 be marked, so as to avoid reviewers needing to raise problems with such text. Meek, personal: Don't like to dismiss restructuring just because of time pressure. Advocates careful consideration of structure for general benefits. The LIA-1 approach has benefits. But don't agree with many of Karlsson's assumptions and conclusions: Side by side reading of LIA-1 and LIA-2 The two documents will not be read by the same people and not at the same time. More concerned with notation than document structure Good benefits from current structure Seeing all aspects of a given function in one place. The case for a major change isn't strong enough Had it been done in the LIA-1 style originally, there would be requests to convert it to the current style There are both technical and practical reasons for using the current style. Lozier: Don't have a problem with the current structure, not a weakness Perhaps there are important technical comments to consider A standards document is not a mathematical document Editors should accommodate technical comments as necessary. General consensus: adopt the suggestion of the editors. Wakker: Presents "best case" schedule for LIA-2: Canonical voting requirements: Step 0) WG11 agrees LIA-2 draft is a "Committee Draft" (CD) ready for outside review. Step 1) Register working document as a CD (4 month ballot). Step 2) If CD registration ballot passes, hold a CD content ballot (4 months). Step 3) If the CD content ballot passes, and it was designated as a "final" CD content ballot, request permission to hold a DIS country ballot. If successful, 4 month ballot DIS ballot. Step 4) When step 3 is complete, with adequate votes, DIS is slightly modified to accommodate comments, and published as an IS. Under recent rule changes, it is now allowed to request that steps 1 and 2 be combined. WG11 received permission to combine steps 1 and 2 for LIA-2 (see resolution AH above) New DIS rules starting 1-Apr-1996: Only very minor editorial changes ( spaces, obvious misspellings) are allowed after a successful DIS ballot. ITTF (editorial style) comments are provided as part of a "final" CD content ballot. These comments must be accepted. So, at best, WG11 could achieve the following schedule for LIA-2: 1) Revised draft by end of October 1995, short review by WG11 by email (yes/no), send to SC22 Secretariat for 4 month CD Registration/Content ballot, complete by Feb '96. 2) WG11 meeting in April to consider ballot comments 3) DIS draft text by May for WG11 approval 4) Send to SC22 secretariat in June for DIS country ballot, complete by Oct 1996. 5) Resolve ballot comments, publish IS in Dec 1996. Wakker: Presents balloting rules CD content balloting rules are unclear. Must achieve "substantial support". DIS balloting rules are concrete: 1) 2/3 or more "P" members must vote positively. Abstaining "P" members are not included in this fraction. 2) No more than 1/4 of all members (including "O" members) may vote against. Abstaining "P" members are counted as members in this fraction.