[ 
https://issues.apache.org/jira/browse/DAFFODIL-2019?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16714770#comment-16714770
 ] 

Steve Lawrence commented on DAFFODIL-2019:
------------------------------------------

See this thread for reent DFDL-WG discussions on this issue:

[https://www.ogf.org/pipermail/dfdl-wg/2018-December/003941.html]

The result of these discussions is that our implementation of hexBInary with 
non-byte lenghths is incorret. We are correctly taking into account bitOrder, 
but byteOrder should effectively be ignored. We need to change our 
implementation to use "The Byte String Way" as described in the email above, 
which is backwars comatabile with the DFDL specification. Note that this change 
will result in a change in how hexBinary is output, but this fixes Daffodils 
incorrect output.

> daf extension property to turn on/off "hexBinary with bits" lengthUnits 
> behavior
> --------------------------------------------------------------------------------
>
>                 Key: DAFFODIL-2019
>                 URL: https://issues.apache.org/jira/browse/DAFFODIL-2019
>             Project: Daffodil
>          Issue Type: Bug
>          Components: Compatibility, Front End
>    Affects Versions: 2.2.0
>            Reporter: Michael Beckerle
>            Priority: Major
>
> We need a property to turn on/off the daffodil extension where xs:hexBinary 
> elements can have length expressed in bits.
> This functionality, as implemented today (Nov 2018) is backwards incompatible 
> with the DFDL spec, so is unlikely to be accepted as part of the DFDL 
> standard; hence, a way to turn this on and off, with the default setting to 
> be off, is needed.
> Specifically, this feature of Daffodil makes the hexBinary infoset contents 
> of a hexbinary element dependent on the byteOrder and bitOrder properties. 
> Today, in the DFDL spec, xs:hexBinary is independent of these properties, 
> making it suddenly dependent on them will change behavior of existing 
> schemas. That just won't fly. 
> Probably we need to deprecate this functionality and remind users to model 
> such data as xs:nonNegativeInteger instead. 
> The xs:hexBinary type is really heavily constrained by its role in XSD. E.g., 
> it has length facets. This really makes it much more like a text string than 
> any sort of universal blob that can represent binary data. It's behavior 
> really should be like text strings. My preferred concept for hexBinary is 
> that you parse a text string as iso-8859-1 then convert each character to 2 
> hex digits one by one for the infoset. That keeps you out of trouble with 
> length facets, etc. 
> Users will still want to be able to express that some number of unaligned 
> bits, not a multiple of 4 or 8 long, and not starting, nor ending, on a byte 
> boundary, is in their data and is not numeric data, so base 10 numeric 
> presentation in the infoset is not natural.
> Such data can be best modeled as a xs:nonNegativeInteger. Changing the 
> infoset presentation of this data so that it looks like hex, not base 10, is 
> not really DFDL's primary purpose. It is always possible to convert to 
> xs:hexBinary via dfdl:inputValueCalc, and for unparsing, converted back by 
> dfdl:outputValueCalc. Alas this is always going to run into the "value 
> element problem", i.e., the data would be like:
> <myDataBits><hexvalue>F34A37</hexvalue></myDataBits>
> where the extra <hexvalue>...</hexvalue> is needed to support the 
> dfdl:inputValueCalc/dfdl:outputValueCalc pair, and a hidden group that hides 
> the base 10 nonNegativeInteger representation.



--
This message was sent by Atlassian JIRA
(v7.6.3#76005)

Reply via email to