Re: Support for Canonical EXI interoperability test in TTFMS

Daniel, all,

I’ve been following the discussion regarding Canonical EXI’s treatment of empty elements and would like to offer a suggestion to simplify the wording and improve the efficiency of the proposed solution.

Here is what I would propose:

“When strict is false or the current element grammar contains a production of the form LeftHandSide : EE with event code of length 1, EXI can represent the content of an empty element explicitly as an empty CH event or implicitly as a SE event immediately followed by an EE event. In these circumstances, Canonical EXI MUST represent an empty element by a SE event followed by an EE event.”

I think this description states the issue and the alternate solution simply and clearly. The alternate solution improves compactness by prescribing the most efficient representation of an empty character event when it is available (i.e., by omitting the CH event). It improves processing efficiency by requiring only 1-2 checks (strict & available EE) and does not require knowledge or checking against DTR types. These checks occur in a relatively hot code path, so minimizing overhead is important for efficiency. Because the alternate approach does not depend on DTR knowledge, it also avoids the need to describe how to handle user defined DTRs that can also encode empty strings (which the current proposal does not address). 

I hope this is helpful. Please let me know if you have questions or if I’ve missed anything important.

 Best wishes!,

 John

> On Nov 18, 2015, at 8:39 AM, Peintner, Daniel (ext) <daniel.peintner.ext@siemens.com> wrote:
> 
> Hi Taki, all,
> 
> Thank you for your reply and your valuable comments.
> 
> I updated the proposal to incorporate your feedback. Also, the description now states the intent and lists again the rules.
> 
> 
> --->
> 
> In general, Canonical EXI MUST NOT change the sequence of XML information items. However, the XML Infoset in some rare cases (e.g., due to API characteristics) may miss "Character Information items" such as strings with the number of characters equal to 0 (zero). EXI encoding may also fail without such an "empty" character information item (e.g., strict schema-informed streams that state the requirement of an expected character string - even if empty).
> 
> Hence, Canonical EXI aims for adding an "empty" character information item if the intent requires to do so (e.g., expected character string) and not for any other use case (e.g., mixed content).
> 
> That said, a canonical EXI processor MUST add a CH event with a String of length 0 (zero)
> * if processing the current XML Information item fails by means of existing event codes
>  of length 1 (i.e., no EE or SE event exists), and
> * when processing a schema-informed grammar where a CH event code of length 1 exists with
>  Built-in EXI Datatype Representation "Binary" (exi:base64Binary and exi:hexBinary),
>  "String", "List" or an Enumeration with an empty item.
> 
> In all other cases no further events MUST be added.
> <---
> 
> What do you think?
> Do you have any updates/proposals?
> 
> Thanks,
> 
> -- Daniel
> 
> 
> 
> ________________________________
> Von: Takuki Kamiya [tkamiya@us.fujitsu.com]
> Gesendet: Mittwoch, 11. November 2015 22:40
> An: Peintner, Daniel (ext); public-exi@w3.org
> Betreff: RE: Support for Canonical EXI interoperability test in TTFMS
> 
> Hi Daniel,
> 
> In schema-informed context, CH event-type with event-code length 1 comes from
> two different schema constructs. One is from simple type content, the other is
> from mixed-content.
> 
> For CH event types that came from mixed-content, there is no need for inserting
> empty CH event. Therefore, I would suggest to exclude mixed-content CH event
> types from the rule you described below.
> 
> You listed three EXI datatype representations (i.e. Binary, String and List) as
> applicable to the described empty CH event insertion rule. I would like to point
> out that enumerated values where one of the values is an empty string (i.e. "")
> also should also apply. In other words, in all context where the EXI datatype
> representation associated with the current CH event type allows for an empty CH,
> empty CH event should be inserted.
> 
> Thanks,
> 
> taki
> 
> 
> -----Original Message-----
> From: Peintner, Daniel (ext) [mailto:daniel.peintner.ext@siemens.com]
> Sent: Wednesday, November 11, 2015 5:08 AM
> To: Takuki Kamiya; public-exi@w3.org
> Subject: AW: Support for Canonical EXI interoperability test in TTFMS
> 
> All,
> 
> According to yesterday's telecon I explored the empty CH("") event a bit further.
> 
> There are various situations when an empty CH could be added. One rather obvious case is a schema-informed stream that states the requirements of an expected character string (even if the string is empty). However,  also in schema-less mode one could assume that a previously "learned" CH event could mean that a CH is expected even if it is not there...
> 
> Summarizing I would like to propose the following requirement/addition to the Canonical EXI document.
> 
> --->
> The XML Infoset in some rare cases (e.g., due to API characteristics) may miss "Character Information items" such as strings with the number of characters equal to 0 (zero). That said, EXI encoding may also fail without such an "empty" character information item. Hence, a canonical EXI processor MUST add a CH event with a String of length 0 (zero), if not already there, when beeing in a schema-informed grammar where a CH event code of length 1 exists with Built-in EXI Datatype Representation "Binary" (exi:base64Binary and exi:hexBinary), "String" or "List". The availability of such a CH event in the grammar clearly states the intent, in this case the requirement of empty characters. In all other cases no further events MUST be added.
> <---
> 
> What do people think?
> 
> Thanks,
> 
> -- Daniel
> 
> 
> 
> 
> 
> ________________________________
> Von: Peintner, Daniel (ext) [daniel.peintner.ext@siemens.com]
> Gesendet: Montag, 9. November 2015 17:06
> An: Takuki Kamiya; public-exi@w3.org
> Betreff: AW: Support for Canonical EXI interoperability test in TTFMS
> 
> Taki, all,
> 
> we looked into the issue more closely and found the following issues.
> 
> 1. How to deal with conflicting framework options
> 
> The framework (or the associated test cases) may define conflicting parameters (e.g, preserve processing instructions and strict). In such a situation an EXI processor may decide whether to use non-strict encoding to support processing instructions or to eliminate PI support.
> 
> As it turns out the EXI processors (OpenEXI and EXIficient) tend to use different strategies. That said, both strategies are OK. Hence, I think we need to make the framework aware of such a situation so that the framework decides what is the desired result.
> 
> 
> 2. Empty CH("") events
> 
> An XML schema may define an element as follows
> <xs:element name="foo" type="xs:string"/>
> 
> A valid instance may look as follows.
> 
> <foo></foo>
> 
> Depending on the EXI options and the mode (strict vs. non-strict) the following two EXI streams are possible
> 
> SE(foo) EE(foo)                --> applicable in non-strict only
> SE(foo) CH("") EE(foo)      --> applicable in strict and non-strict
> 
> Again, we need to ensure all Canonical EXI processors behave the same.
> Hence, I would argue for the latter case given that it is usable in both (strict and non-strict) scenario but I am open for other ideas/thoughts.
> 
> 3. Whitespace handling
> 
> I wonder whether we need to define whitespace preservation rules in Canonical EXI similar to the TTFMS framework rules.
> 
> 
> Thanks,
> 
> -- Daniel
> 
> 
> 
> 
> 
> ________________________________
> Von: Takuki Kamiya [tkamiya@us.fujitsu.com]
> Gesendet: Mittwoch, 21. Oktober 2015 00:05
> An: Peintner, Daniel (ext); public-exi@w3.org
> Betreff: RE: Support for Canonical EXI interoperability test in TTFMS
> 
> Hi Daniel,
> 
> I fixed a bug in the TTFMS framework.
> 
> Next time you compile the framework and run the test,
> you will be able to see schema-informed EXI files generated
> when the test case provides one and schema use is enabled.
> 
> Thank you,
> 
> Takuki Kamiya
> Fujitsu Laboratories of America
> 
> 
> -----Original Message-----
> From: Peintner, Daniel (ext) [mailto:daniel.peintner.ext@siemens.com]
> Sent: Wednesday, October 14, 2015 6:50 AM
> To: Takuki Kamiya; public-exi@w3.org
> Subject: AW: Support for Canonical EXI interoperability test in TTFMS
> 
> Hi Taki,
> 
> I uploaded a revised EXIficient library but I agree, I do still see some issues.
> (in my test run 20 files out of 115 are still different)
> 
> Maybe this has to do with whitespace handling (will send separate email...)
> 
> Moreover, I am currently able to run schema-less test runs only by calling
> ant run-iot-c14n-classes -DtestCases=config/testCases-restricted/all-v1.xml
> 
> Maybe someone can point me to the configuration how to call schema-informed test runs or byteAligned test runs to facilitate debugging.
> 
> Thanks,
> 
> -- Daniel
> 
> 
> 
> ________________________________
> Von: Takuki Kamiya [tkamiya@us.fujitsu.com]
> Gesendet: Dienstag, 13. Oktober 2015 03:23
> An: Peintner, Daniel (ext); public-exi@w3.org
> Betreff: RE: Support for Canonical EXI interoperability test in TTFMS
> 
> Hi Daniel,
> 
> I also modified openexi driver so that it always output header options.
> 
> However, I still see many differences between exificient and openexi
> outputs. We will need to further investigate this.
> 
> Thank you,
> 
> Takuki Kamiya
> Fujitsu Laboratories of America
> 
> 
> -----Original Message-----
> From: Peintner, Daniel (ext) [mailto:daniel.peintner.ext@siemens.com]
> Sent: Thursday, October 01, 2015 5:50 AM
> To: Takuki Kamiya; public-exi@w3.org
> Subject: AW: Support for Canonical EXI interoperability test in TTFMS
> 
> Hi Taki,
> 
> Thank you for pointing me to the parameter "measure" which indicates the type of the test run.
> 
> I also uploaded a first snapshot of the EXIficient library supporting Canonical EXI. Additional updates may be necessary.
> When comparing the encoded files with OpenEXI I do see mostly diffs. I think it is because OpenEXI at the moment does not always include the EXI Options.
> 
> Please let me know if you encounter other issues.
> 
> Thanks,
> 
> -- Daniel
> 
> ________________________________
> Von: Takuki Kamiya [tkamiya@us.fujitsu.com]
> Gesendet: Donnerstag, 1. Oktober 2015 01:45
> An: Peintner, Daniel (ext); public-exi@w3.org
> Betreff: RE: Support for Canonical EXI interoperability test in TTFMS
> 
> Hi Daniel,
> 
> You should be able to get the test mode by accessing:
> measure field (of class MeasureParam) that is in _driverParams (of class DriverParameters)
> 
> When it is iot_c14n_encode, you should change the behavior of the
> processor to comply with c14n rules.
> 
> Do you plan to check-in new EXIficient jar to TTFMS soon?
> 
> Thank you,
> 
> Takuki Kamiya
> Fujitsu Laboratories of America
> 
> 
> -----Original Message-----
> From: Peintner, Daniel (ext) [mailto:daniel.peintner.ext@siemens.com]
> Sent: Wednesday, September 30, 2015 8:28 AM
> To: Takuki Kamiya; public-exi@w3.org
> Subject: AW: Support for Canonical EXI interoperability test in TTFMS
> 
> Hi Taki,
> 
> I did check out the new code and it worked as expected.
> Thank you for your work!
> 
> The only thing I miss is a testCase option that informs about whether the EXI processor is required to produce canonical EXI.
> 
> Did I miss anything with that regard?
> 
> Thanks,
> 
> -- Daniel
> 
> 
> 
> P.S. EXIficient does not sort attributes in schema-less mode
> 
> 
> ________________________________
> Von: Takuki Kamiya [tkamiya@us.fujitsu.com]
> Gesendet: Dienstag, 29. September 2015 02:20
> An: public-exi@w3.org
> Betreff: Support for Canonical EXI interoperability test in TTFMS
> 
> Hi,
> 
> I added support for Canonical EXI interoperability test in TTFMS.
> 
> You need to invoke target " run-iot-c14n-classes" in order to run the
> encoding process.
> 
> After that, diff tools such as WinMerge (on windows) can be used to
> compare the encoded files output by various implementations.
> 
> Initial experimental run showed quite a lot of differences in encodings
> between EXIficient and OpenEXI.
> 
> I found at least some of the diffs are due to the attribute orders in
> schema-less setting. Is it true that EXIficient sorts attributes whether
> it is schema-less or schema-informed?
> 
> Thank you,
> 
> Takuki Kamiya
> Fujitsu Laboratories of America
> 
> 
> 
> 

AgileDelta, Inc.
john.schneider@agiledelta.com
http://www.agiledelta.com
w: 425-644-7122
m: 425-503-3403
f: 425-644-7126