RE: [w3c sml][4665] Clarify URI equivalence in reference to RFC 3986

I do not have a strong opinion on the case issue. Since the WG seemed to prefer case sensitive comparison, that is how I have defined the comparison in the checked in spec.

As for normalization, we do not preclude it. We put that burden on the producer rather than the consumer. Once the producer guarantees it, the consumer need not perform any further normalization since a simple string comparison is guaranteed to achieve interop.


From: public-sml-request@w3.org [mailto:public-sml-request@w3.org] On Behalf Of John Arwe
Sent: Friday, September 21, 2007 11:07 AM
To: public-sml@w3.org
Subject: RE: [w3c sml][4665] Clarify URI equivalence in reference to RFC 3986


As I interpret "case insensitive", it means "a"=="A".  Is that how you mean it?

If so then I fail to see how this is simpler than "case sensitive" comparison.  Case-insensitive requires normalization (to one case), which is in direct opposition to your own first argument.  Thus, to me, it is confusing unless there is some other factor not yet expressed that makes this a tradeoff worth doing.  Requiring case normalization _might_ also cause burdens on implementations that choose to support encodings other than utf-8/16.  I am _far_ from an expert on encodings, but I do remember hearing about cases involving Katakana I believe where case-folding was quite expensive.  If this becomes a critical factor in the decision I will have to talk to some globalization folks.

This seems to me like it might be another floor/ceiling discussion.  3986 clearly discusses the tradeoff between computational cost and the risk of concluding that two URIs identify distinct resources when they do not, because of incomplete normalization (I think it refers to this case as false negative, but the language makes my head hurt so I wrote it out).  It seems as if we can allow simple string comparison as a floor without precluding more sophisticated implementations from climbing further up the comparison ladder.  Remember that the risk in our context of believing two URIs refer to different things is the risk of getting the model boundary wrong in the interchange case.  Interop would only be guaranteed at the floor of course, as usual, but it is still sufficient for interop.

Best Regards, John

Street address: 2455 South Road, Poughkeepsie, NY USA 12601
Voice: 1+845-435-9470      Fax: 1+845-432-9787
Kumar Pandit <kumarp@windows.microsoft.com>
Sent by: public-sml-request@w3.org

09/19/2007 06:15 PM

To

Sandy Gao <sandygao@ca.ibm.com>, "public-sml@w3.org" <public-sml@w3.org>

cc

Kumar Pandit <kumarp@windows.microsoft.com>

Subject

RE: [w3c sml][4665] Clarify URI equivalence in reference to RFC 3986







I added the ‘case insensitive’ clause because, if an implementation is based on a file-system that uses case insensitive paths then it fits nicely with the current proposal. That said, I do not have a strong bias towards that option. I am ok with defining the comparison as case sensitive while keeping the rest of the definition as is.

It is not clear from your reply if you agree with the proposal (sans the case-insensitive part) or if you want to base your decision on whether the URI/IRI gurus agree with it first. Can you please clarify?

Since no one has disagreed with the proposal (except the concern about case-insensitive part), if you agree with the amended wording, we may actually be able to get this into the second draft today.


From: public-sml-request@w3.org [mailto:public-sml-request@w3.org] On Behalf Of Sandy Gao
Sent: Wednesday, September 19, 2007 11:11 AM
To: public-sml@w3.org
Subject: Re: [w3c sml][4665] Clarify URI equivalence in reference to RFC 3986


This is a simple proposal, and being simple is normally good, but I'll leave this to the URI/IRI gurus to determine whether the simple solution is good enough to cover real-life scenarios.

One thing that worries me is the "case insensitive" part. Why? As far as I can tell, this doesn't match any of the steps in "6.2. Comparison Ladder" of RFC 3986. If we want the simplest possible solution, then we should use what's defined in 6.2.1 and compare strings character-by-character case-sensitivly.

Thanks,
Sandy Gao
XML Technologies, IBM Canada
Editor, W3C XML Schema WG<http://www.w3.org/XML/Schema/>
Member, W3C SML WG<http://www.w3.org/XML/SML/>
(1-905) 413-3255 T/L 969-3255
Kumar Pandit <kumarp@windows.microsoft.com>
Sent by: public-sml-request@w3.org

2007-09-12 11:02 PM


To

"public-sml@w3.org" <public-sml@w3.org>

cc

Kumar Pandit <kumarp@windows.microsoft.com>

Subject

[w3c sml][4665] Clarify URI equivalence in reference to RFC 3986











Here is my proposal to resolve this issue.

Proposal:
Uri equivalence in SML-IF should be defined as case insensitive simple string comparison based on codepoint-by-codepoint comparison of the corresponding characters in the uri.

Justification:
1.        Performance: Simple string comparison provides highest performance. Although it is true that two aliases of the same uri may not compare as equal without normalization, the problem does not exist in the specific context of an SML-IF producer. This is because, when a producer is writing out an SML-IF document, it can apply normalizations (if necessary) such that a given uri always appears in the same way. This allows consumers to perform fast string comparison without needing to perform any type of normalization.

RFC 3986 section 2 (Comparison Ladder) describes many different forms of normalizations (syntax-based/case/percent-encoding/path-segment/scheme-based/protocol-based). If we want a consumer to perform normalizations, we not only make a consumer less efficient but also need to add very specific normalization step definitions in the SML-IF spec. On the other hand, if we leave the burden of normalization to the producer, we can keep the SML-IF spec much simpler and allow consumers to be more efficient. This way the spec does not need to talk about any specific comparison ladder step(s) to be performed by a producer. The producer is free to apply any (or none) normalization steps as long as it knows it will write a given uri in the same format.
2.        Precise definition: RFC 3986 section 6.2.1 (Simple String Comparison) discusses issues involved in performing a string comparison but does not provide a precise definition of how the comparison must be performed. In other words, it leaves some room for interpretation. We should avoid this by presenting an unambiguous definition based on that discussion.