Hi Dave We should also bear the problem below in mind when thinking about DFDL Infoset & XDM. XDM assumes that an element with a concrete type-name has a typed-value conforming to the type-name, ie, it has been 'validated'. If this is not the case then the type-name is set to xs:untyped or xs:untypedAtomic (extra types added to XDM for this purpose). In DFDL Infoset we had been assuming that the [dataType] would be set to that implied by the DFDL xsd, regardless of whether validation succeeded or not - though there are issues with this as explained below. Regards Steve Hanson Programming Model Architect WebSphere Message Brokers Hursley, UK Internet: smh@uk.ibm.com Phone (+44)/(0) 1962-815848 ----- Forwarded by Steve Hanson/UK/IBM on 13/05/2009 12:51 ----- Steve Hanson/UK/IBM 29/04/2009 15:54 To Alan Powell, Dave Glick, Mike Beckerle (Work) cc Suman Kalia/Toronto/IBM@IBMCA, Tim Kimber/UK/IBM Subject Fw: Action 020 completion Hi Dave We discussed this on the call and agreed that the unsigned types are just range restrictions so treating negative numbers in the data as parse errors instead of validation errors seemed inconsistent. Various options discussed: 1) Remove the unsigned types altogether. - Means we'd need an extra property to describe binary integer representation, as we could no longer infer the rep from the logical type - Loses type information for applications where the fact that data is unsigned is important. - Means DFDL modelers would have to create their own duplicate restrictions for common C etc data types. 2) Change [dataType] to point to the XML Schema primitive type instead of the XML Schema built-in type. - Means that the value and the type would be xs:decimal which is too general 3) Change [dataValue] to say "The value in the value space of the underlying XML Schema primitive type forthe [datatype] member or special value nil" - Allows the infoset to carry integer data that is invalid due to range regardless of value. - Means that the value would be a decimal even though the data type was (say) xs:unsignedLong, ie, the datatype and datavalue are no longer in step unless validated 4) Option 2) with the modification that the primitive type for all integer types was xs:integer and not xs:decimal. 5) Option 3) with the modification that the primitive type for all integer types was xs:integer and not xs:decimal. We agreed not to close on this until you had reported back on your action 032. Regards Steve Hanson Programming Model Architect WebSphere Message Brokers Hursley, UK Internet: smh@uk.ibm.com Phone (+44)/(0) 1962-815848 ----- Forwarded by Steve Hanson/UK/IBM on 29/04/2009 15:32 ----- Steve Hanson/UK/IBM 27/04/2009 13:42 To Alan Powell/UK/IBM cc dglick@dracorp.com, mbeckerle@oco-inc.com Subject Re: Action 020 completion Thanks Alan I've added correct property names, see below. But I've omitted floats deliberately for clarity, the logical type is always signed and physical type is always signed, so there's no issue. However, there is a problem with what I have stated, as you pointed out on Friday. On parsing I am effectively validating the input data, but on unparsing I am not assuming the data has been validated. This is not consistent and needs correcting. But as I looked into this, I realised we have a problem with how we have described the DFDL infoset. The spec says "There is no requirement for DFDL-described data to be valid in order to have a DFDL information set.", which is in accordance with our agreed position on validation being optional. But further on it also says: [datatype] String. The name of the XML Schema 1.0 built-in simple type to which the value corresponds. DFDL supports a subset of these types listed in the specification at section 4.1. [dataValue] The value in the value space of the [datatype] member or special value nil. This says to me that the DFDL parser must have done enough validation to ascertain that the value matched the underlying built-in type. For example, I have a user-defined simple type that adds a max/min range of +100-+200 to an xs:unsignedInt. If the input data has value 99, the value will be accepted into the infoset, but will not validate if validation is switched on. If the input data is a packed decimal with value -1, the value will not be accepted into the infoset. Given that xs:unsignedInt is itself just a range restriction of xs:integer (via xs:nonNegativeInteger), this seems a bit arbitrary. Dave - given your action item looking at DFDL Infoset versus XDM, I'd be interested in your opinion here. Regards Steve Hanson Programming Model Architect WebSphere Message Brokers Hursley, UK Internet: smh@uk.ibm.com Phone (+44)/(0) 1962-815848 Alan Powell/UK/IBM 24/04/2009 15:13 To Steve Hanson/UK/IBM@IBMGB cc dglick@dracorp.com, mbeckerle@oco-inc.com Subject Re: Action 020 completion Steve Looks OK But can you use the correct property name eg binaryNumberRepresentation and for completeness add binaryFloatRepresentation (even though it may be obvious) Alan Powell MP 211, IBM UK Labs, Hursley, Winchester, SO21 2JN, England Notes Id: Alan Powell/UK/IBM email: alan_powell@uk.ibm.com Tel: +44 (0)1962 815073 Fax: +44 (0)1962 816898 From: Steve Hanson/UK/IBM To: mbeckerle@oco-inc.com, Alan Powell/UK/IBM, dglick@dracorp.com Date: 23/04/2009 12:50 Subject: Action 020 completion Here's my proposal for the behaviour when a logical type is signed and the physical data has no sign (either because it is not capable of carrying a sign, or it carries an unsigned indicator), and when the logical type is unsigned and the physical data has a sign. The principle is to be flexible and only to give errors when things are clearly mis-matched Logical type textNumberRepresentation= text (4) textNumberRepresentation= zoned (2) (6) binaryNumberRepresentation= packed (5) binaryNumberRepresentation= bcd (1) binaryNumberRepresentation= binary Signed (decimal, integer) Parse: OK Unparse: OK Parse: Unsigned data => +ve Unparse: Data always punched with sign Parse: Unsigned data => +ve Unparse: Data signed as per +ve/-ve nibble specifiers, unsigned nibble specifier never used Parse: Data always +ve Unparse: -ve data is error N/A Signed (long, int, short, byte) Ditto Ditto Ditto Ditto Parse: Data assumed 2's complement binary Unparse: Data output as 2's complement binary Unsigned (unsigned long, unsigned int, unsigned short, unsigned byte) (3) Parse: -ve data is error Unparse: -ve data is error Parse: +ve data => OK, -ve data is error Unparse: Sign never punched, -ve data is error Parse: +ve data => OK, -ve data is error Unparse: Unsigned nibble specifier always used, -ve data is error Parse: OK Unparse: -ve data is error Parse: Data assumed unsigned binary Unparse: Data output as unsigned binary (1) Can not physically carry a sign (2) Some systems output unsigned for +ve, but accept +ve on input (eg, IBM iSeries) (3) Assumes that on unparsing, the infoset could still present a -ve value (4) The -ve sign is indicated by numberPattern property (5) The exact sign nibbles are given by the packedDecimalSignCodes property (6) The punching style to use is given by the numberZonedSignStyle property Mail back any comments before next week's call. Regards Steve Hanson Programming Model Architect WebSphere Message Brokers Hursley, UK Internet: smh@uk.ibm.com Phone (+44)/(0) 1962-815848 ----- Forwarded by Steve Hanson/UK/IBM on 23/04/2009 12:04 ----- Steve Hanson/UK/IBM 18/02/2009 16:55 To Mike Beckerle (Work) cc Alan Powell/UK/IBM, dglick@dracorp.com Subject DFDL: Packed & zoned decimals - more thoughts (was Action 020) Hi Mike While we are on the subject of how to handle signs, the spec does not fully define what happens for a number if the logical type is unsigned. We need to say what is expected in the physical data and what happens if the data contains a sign. For example, we say that for an unsigned integer, if the rep is binary then we treat the data as 'unsigned binary' and not twos complement. And we say that BCD is only allowed for unsigned logical types. That is good. But we don't do the same for packed, text, zoned. I think we need to say that no explicit sign is expected in the data (eg, packed should have only F or 0, no A,B,C,D) and if it does: Alternatives: i) Error ii) Positive sign discarded, negative sign gives error iii) Sign discarded iv) As per i) if 'strict' set, as per ii) if 'lax' set v) As per i) if 'strict' set, as per iii) if 'lax' set Personally I vote for i) Regards Steve Hanson Programming Model Architect WebSphere Message Brokers Hursley, UK Internet: smh@uk.ibm.com Phone (+44)/(0) 1962-815848 ----- Forwarded by Steve Hanson/UK/IBM on 18/02/2009 16:41 ----- Steve Hanson/UK/IBM 12/02/2009 13:18 To <mbeckerle.dfdl@gmail.com> cc dfdl-wg@ogf.org Subject RE: [DFDL-WG] Fw: DFDL OGF WG call - Action 020 Hi Mike I think it's a simplification too far. Many people especially those with a mainframe or COBOL background know what a zoned decimal is. The wikipedia entry for binary coded decimal explicitly covers the BCD, packed & zoned 'variants'. MRM and WTX both explicitly support zoned too. And it's easier to say that the 'decimalSignStyle' property applies to zoned decimals than to say it applies to any patterns that happen to have a P in them. On balance I would keep zoned as a representation. So we need to decide whether zoned is only allowed for a signed decimal. There's no harm in allowing it for unsigned, just some redundancy, and it makes validation of the pattern against the rep easier (if something is zoned it can only have a subset of pattern chars). Btw we don't need leading overpunched sign, only trailing - see my case for this below. Regards Steve Hanson Programming Model Architect WebSphere Message Brokers Hursley, UK Internet: smh@uk.ibm.com Phone (+44)/(0) 1962-815848 "Mike Beckerle" <mbeckerle.dfdl@gmail.com> 12/02/2009 12:51 Please respond to <mbeckerle.dfdl@gmail.com> To Steve Hanson/UK/IBM@IBMGB, <dfdl-wg@ogf.org> cc Subject RE: [DFDL-WG] Fw: DFDL OGF WG call - Action 020 This does suggest another simplification. Zoned is so close to text....Suppose we scrap the concept of "zoned" altogether, and just add a character to our number pattern language to allow one to specify a overpunched sign digit. E.g., "+00000" is text "P0000" same with overpunched leading sign. "00000+" text "0000P" same with overpunched trailing sign The decimal point would normally be implied in these, (I still like having a cobol-style "V" to position this instead of separate properties stating the position - one of the few good features about cobol is the number patterns. I still think we could quite easily pre-process the "V" out of these strings and then hand the rest through to an ICU library as an implementation - however the "P" probably does need to be a change in that library.) Mike Beckerle | OGF DFDL WG Co-Chair | CTO | Oco, Inc. Tel: 781-810-2100 | 504 Totten Pond Road, Waltham MA 02451 | mbeckerle.dfdl@gmail.com From: dfdl-wg-bounces@ogf.org [mailto:dfdl-wg-bounces@ogf.org] On Behalf Of Steve Hanson Sent: Wednesday, February 11, 2009 1:34 PM To: dfdl-wg@ogf.org Subject: [DFDL-WG] Fw: DFDL OGF WG call - Action 020 It was noted on the call this week that there is an alternative to my zoned decimal overpunching proposal i) below. I said: - If it is an unsigned type then DFDL expects the rightmost byte to have a zone nibble when parsing, and outputs a zone nibble when unparsing. - If it is a signed type then DFDL expects it to have a sign nibble when parsing, and outputs a sign nibble when unparsing. But my unsigned type behaviour could be achieved by specifying a rep of text instead of zoned. If that is the case, the alternative is to only allow zoned rep for signed decimal logical types. Thoughts? Regards Steve Hanson Programming Model Architect WebSphere Message Brokers Hursley, UK Internet: smh@uk.ibm.com Phone (+44)/(0) 1962-815848 ----- Forwarded by Steve Hanson/UK/IBM on 05/02/2009 10:17 ----- Steve Hanson/UK/IBM 28/01/2009 13:54 To DFDL Working Group cc Subject DFDL OGF WG call - Action 020Link Action 020: 020 SH: Resolve packedDecimalSignCodes behaviour depends on NumberCheckPolicy 22/10: No progress 10/12: added how to decide to overpunch and sign position a) Resolve packedDecimalSignCodes behaviour depends on NumberCheckPolicy Add new property to section 15.4 Properties Specific to Number with Binary representation. binaryNumberCheckPolicy Enum Values are ?strict? and ?lax?. Indicates how lenient to be when parsing binary numbers. If ?lax? then the parser tolerates all valid alternatives where such alternatives exist. Specifically, for binaryNumberRepresentation = 'packed' the sign nibble for positive, negative, unsigned and zero is allowed to be any of the valid respective values. On unparsing, the specified value is always used. Also suggest changing some of the other property names in 15.4: "decimalVirtualPoint" -> "binaryDecimalVirtualPoint" "packedDecimalSignCodes" -> "binaryPackedSignCodes" And changing binaryNumberRepresentation enumeration: "BCD" -> "bcd" b) Zoned decimals: How to decide to overpunch and sign position Spec assumes that overpunching of the rightmost character always takes place. IBM architecture allows no overpunching (ie, Fx instead of Cx/Dx) - this is supported by IBM MRM & WTX parsers. Additionally IBM MRM parser allows separate sign byte, and sign byte on left. Let's deal with these separately: i) No overpunching. The IBM architecture allows the rightmost byte to have a zone (Fx) or a sign (Cx/Dx) as the left nibble. I don't see why we can't base what to expect when parsing, and output when unparsing, on the logical xsd type. - If it is an unsigned type then DFDL expects the rightmost byte to have a zone nibble when parsing, and outputs a zone nibble when unparsing. - If it is a signed type then DFDL expects it to have a sign nibble when parsing, and outputs a sign nibble when unparsing. For analogy with DFDL packed decimals, it seems at first glance that we should also extend the numberCheckPolicy 'lax' setting to treat a zone nibble as a +ve sign nibble for a signed type. However, IBM iSeries always outputs Fx to mean +ve but accepts both Fx & Cx on input. It is perhaps better therefore that DFDL always tolerates Fx when parsing a signed zoned decimal, otherwise iSeries users would always have to set numberCheckPolicy to 'lax', which might have other implications in the future. ii) Separate sign byte. I don't believe the IBM architecture allows this. I don't think DFDL needs to support it. MRM has this, but I think it's because early on MRM did not explicitly support text decimals as such, just COBOL variations, and it was easier just to call them all zoned. iii) Sign byte on left. I don't believe the IBM architecture allows this. I don't think DFDL needs to support it. MRM has this, but for the same reason as ii) Conclusion: No new DFDL properties needed, but words need adding to explain zoned parse/unparse behaviour better. Also suggest changing property names: "zonedDecimalSignStyle" -> "numberZonedSignStyle" "zeroNumberRep" -> "numberZeroRep" Should also make clear that any explicit negative pattern in numberPattern will be ignored if the xsd type is unsigned. (We could make this an error but it precludes creation of a textNumberFormat that works with both signed and unsigned types, plus pattern "##0.0" implictly is equivalent to "##0.0;(##0.0)" ). Regards Steve Hanson Programming Model Architect WebSphere Message Brokers Hursley, UK Internet: smh@uk.ibm.com Phone (+44)/(0) 1962-815848 Alan Powell/UK/IBM@IBMGB Sent by: dfdl-wg-bounces@ogf.org 23/01/2009 13:36 To dfdl-wg@ogf.org cc Subject [DFDL-WG] DFDL: Minutes from OGF WG call, 21 January 2009 Open Grid Forum: Data Format Description Language Working Group Weekly Working Group Conference Call 14:00 GMT, 21 January 2009 Attendees Alan Powell (IBM) Mike Beckerle(Oco) Apologies Steve Hanson (IBM) 1. XSD 1.1 Deferred to next call 2. Calendar formats Discussed updated (v4) supplement emailed by AP Agreed millisec/secSinceEpoc cannot be implied by length of logical data so need seperate enumerations. Observed that these options were really combination of 3 properties binary, length and sec/millisec. Suggested renaming to binarySeconds and binaryMilliseconds Packed calendars: decided that need to be able to specify at least the packedDecimalSignCodes property rather than assuming a default so reference will be added to calendar description Locale needs to be specified for numberformats and calendarFormats (didn't identify any other areas) as it modifies the behaviour of ICU. Decided to add locale to numberFormat and CalendarFormat 3. Escape Schemes Agreed need for multiple escape delimiter pairs but not nested. Need an escape for escape character even though in most cases this will be the same character, eg /n //, There are some formats that have a different escape, eg /n &/. Only need single escape characters and one level of escape characters. Discussed how to deal with comments of the form /* comment */ where the escape delimiters are also the initiator and terminator of the field. Semantic needed is 'only look for field terminator not any parent terminator or any other syntax elements'. May fall out naturally from the speculative parsing rules. Need further discussion. 4. AOB Next call 28 January 14:00 Meeting closed, 15:00 GMT Actions raised at this meeting No Action 031 Current Actions: No Action 012 AP/SH: Update decimalCalendarScheme 10/9: Not allocated yet 17/9: No update 24/9: Add calendar binary formats to actions 22/10: No progress 16/1: proposal distributed and discussed. Will be redistributed 21/1: add locale, 020 SH: Resolve packedDecimalSignCodes behaviour depends on NumberCheckPolicy 22/10: No progress 10/12: added how to decide to overpunch and sign position 023 MB: Review Schema 1.1 024 String XML type 025 Escape schemes 21/1: discussed requirements 026 SH: Envelopes and Payloads 027 Property precedence tables 028 Variable markup 029 valueCalc (output length calculation) 030 AP: confirm with WTX that can drop duration 21/6: WTX confirm that they do not have a duration type so do not need it in dfdl. Will drop from spec. Closed Closed actions: 030 AP: confirm with WTX that can drop duration 21/6: WTX confirm that they do not have a duration type so do not need it in dfdl. Will drop from spec. Closed 034 Work items: No Item 001 String XML type (Ian P) - Apr 30, 2008 002 Escape schemes (Ian P) - Apr 30, 2008 003 Variables - ??, 2008 (Mike) 005 Improvements on property descriptions - ??, 2008 (All - split TBD) 006 Envelopes and Payloads (Steve) - Apr 30, 2008 007 (from draft 32) valueCalc (Mike) - ??, 2008 mostly complete 008 (from draft 32) Property precedence for writing (Steve) - under review 009 (from draft 32) Variable markup (Steve) - Mar 31, 2008 proposal needs writing up 010 (from draft 32) Assertions, discriminators and choice, including discussion of timing option (Suman) - Mar 31, 2008 * in progress * 011 (from draft 32) How speculative parsing works (combining choice and variable-occurence - currently these are separate) ??, 2008 (IBM) in progress 012 (from draft 32) Reordering the properties discussion: move representation earlier, improve flow of topics ??, 2008 (Alan) * not started * 025 Augmented infoset and unparsing (Alan) added but needs work complete - specification updated Alan Powell MP 211, IBM UK Labs, Hursley, Winchester, SO21 2JN, England Notes Id: Alan Powell/UK/IBM email: alan_powell@uk.ibm.com Tel: +44 (0)1962 815073 Fax: +44 (0)1962 816898 Unless stated otherwise above: IBM United Kingdom Limited - Registered in England and Wales with number 741598. Registered office: PO Box 41, North Harbour, Portsmouth, Hampshire PO6 3AU -- dfdl-wg mailing list dfdl-wg@ogf.org http://www.ogf.org/mailman/listinfo/dfdl-wg Unless stated otherwise above: IBM United Kingdom Limited - Registered in England and Wales with number 741598. Registered office: PO Box 41, North Harbour, Portsmouth, Hampshire PO6 3AU Unless stated otherwise above: IBM United Kingdom Limited - Registered in England and Wales with number 741598. Registered office: PO Box 41, North Harbour, Portsmouth, Hampshire PO6 3AU Unless stated otherwise above: IBM United Kingdom Limited - Registered in England and Wales with number 741598. Registered office: PO Box 41, North Harbour, Portsmouth, Hampshire PO6 3AU Unless stated otherwise above: IBM United Kingdom Limited - Registered in England and Wales with number 741598. Registered office: PO Box 41, North Harbour, Portsmouth, Hampshire PO6 3AU Unless stated otherwise above: IBM United Kingdom Limited - Registered in England and Wales with number 741598. Registered office: PO Box 41, North Harbour, Portsmouth, Hampshire PO6 3AU Unless stated otherwise above: IBM United Kingdom Limited - Registered in England and Wales with number 741598. Registered office: PO Box 41, North Harbour, Portsmouth, Hampshire PO6 3AU Unless stated otherwise above: IBM United Kingdom Limited - Registered in England and Wales with number 741598. Registered office: PO Box 41, North Harbour, Portsmouth, Hampshire PO6 3AU