EPR Note
Kirk: will outline Note before taking
comments
... EPR reference scheme is different from other reference schemes in
that EPR does not give consumer enough information to resolve the
reference from service
... e.g., need to know protocol
... section 2 defines framework for EPR-based reference schemes not a
scheme itself
... section 3 shows how to take framework and build a scheme with info
about the protocol
Kirk is reviewing section 2
Ginny: is it possible to create an EPR
scheme that is a target-complete identifier?
Kirk: No, there will be some knowledge
needed out of band
John: strictly, technically
speaking you could but it might not be practical
Kirk is reviewing section 3 - a concrete reference
scheme definition that uses the framework of section 2
John: current text says SML extends
Schema; should drop "Schema" or use both "Schema" and "XML"
Discussing John's comments; John will enter bugs later.
Section 2 - separate bugs required for:
... content of service response
message
... use of model consumer
... different bindings may be required - which bindings are we talking
about
Sandy: section 2, bullet 2c - questions
"should"; how normative should we be?
John suggests using "is unresolved"
Sandy: need more detail on example used
in last paragraph regarding GED example
Ginny: section 4, bullet 1, first
sentence is not true ("will not be able to"); 2nd sentence is ok.
John suggests breaking these 2 sentences apart
Discussion of section 4
Ginny: premise of 4.1 is that an SML
reference that in an instance of EPR scheme cannot be changed to an instance of an SML URI scheme. This is
incorrect. Section 4.2 needs to be changed also since it is based on
4.1.
Test Cases
<pratul> Call for Implementations:
See http://www.w3.org/2005/10/Process-20051014/tr.html#cfi
Kumar: first step would be to have a list
of test cases to discuss
... each test case will be an SML-IF document
Kumar has an initial list; will email this to the
group tomorrow
Test cases should be in CVS; need to ask MSM about this
Bug 5558
<pratul> XML spec defines a document
as
Pratul: propose that the definition of
document changed to "an XML document" only, remove "well-formed"
<pratul> [Definition: A data object
is an XML document if it is well-formed, as defined in this
specification. In addition, the XML document is valid if it meets
certain further constraints.]
Pratul: should we repeat "well-formed" in
our conformance criteria when an XML document is defined as
"well-formed"
... how about "a model is a conforming SML model ..."
RESOLUTION: Change
conformance criteria to read "a model is a conforming SML model ..."
... change bug to editorial
<zeckert> scribe: Zulah
<scribe> scribenick: zeckert
continuation of XHTML discussion
xlink: 5561: http://www.w3.org/Bugs/Public/show_bug.cgi?id=5561
references in XHTML: 5562: http://www.w3.org/Bugs/Public/show_bug.cgi?id=5562
<johnarwe> Q from Ginny on the
bifocal solution to the <img > tag in xhtml. What if the 2 ref
schemes were not (1) src= (2) longdesc= ... per yesterday's discussion
... but instead were (1) src= alone (2) src= and longdesc= . Any
better/worse? Appears to be at least as good as yesterday's take.
MSM: response could be that we have convinced ourselves that a
reference scheme can be created for XLink but for XMHTML would would
get a partial solution. Agrees that we could use XLink as a reference
scheme.
. .. bifocal solution refers to (for XHTML) two
different reference schemes. One where only src is considered and the
other where only longdesc is considered.
pratul has sent notes for the group to consider
pratul: suggests
a formal note on the reference scheme
johnarwe: proposal that in response to 5561 and 5562, we will
publish a working group note on the XLink reference scheme only and
that we will not publish a working group note about XHTML
<MSM> Pratul's notes:
http://lists.w3.org/Archives/Member/member-sml/2008Apr/0000.html
<johnarwe> we need to SPLIT that into
2 proposals
johnarwe: proposal is that, in response to
bug 5561 for the XLink reference scheme, the group will respond by
publishing a working group note on the reference scheme that we
discussed yesterday.
kumar: will a content dependency in scheduling be introduced by
this?
johnarwe: no dependency is created by the W3C process. In
negotiating the resolution of the bug, one could be introduced.
RESOLUTION: the proposal passes and the group will create a
working group note.
<johnarwe> in particular, no
objections noted
johnarwe: should the bug be a won't
fix?
pratul: or something that we leave for a future version?
discussion ensues where the group agrees that the
issue with the XHTML reference scheme does not come from the simple and
extended links
MSM clarifies what could happen in the W3C process
if there was disagreement over resolving the bug as to be done in a
future version.
johnarwe: proposal is that, in response
http://www.w3.org/Bugs/Public/show_bug.cgi?id=5562
for the XHTML reference scheme, the group will respond by marking the
bug later (keyword decided), and we will create in email the text of
the resolution.
johnarwe: we will update the keyword to decided (now) and in
the text on the comment of the bug state our intent to respond once we have
drafted appropriate response text, we will make the bug resolved later.
<johnarwe> revised proposal for 5562:
(1) we commence drafting the eventual response text (2) once the
response text is available, update the bug with the agreed-to response
text plus the boilerplate and add the 'decided' kw (3) once the
submitter either acks our update from step 2 or the 2-week
submitter-response period expires, we update the bug to resolved+later
There are no objections.
RESOLUTION: the proposal (see above) is adopted
ACTION: MSM to draft
initial response text for 5562 and send it in email to the group for
discussion [recorded in
http://www.w3.org/2008/04/02-sml-minutes.html#action01]
<trackbot-ng> Created ACTION-178 -
Draft initial response text for 5562 and send it in email to the group
for discussion [on Michael Sperberg-McQueen - due 2008-04-09].
<johnarwe> Tomorrow's regularly
scheduled telecon is CANCELLED
Call for implementations discussion cont.
johnarwe: one
question from previous discussion that we have not answered is what do
we do to maintain the test cases and description document (referring to
change control)
MSM:
put the test cases in CVS on dev.w3.org
MSM: expects
that we would create a test suite directory parallel to existing build
directory
kumar: wishes
to discuss the directory structure now
johnarwe:
believes that this will require more thought than we have time for here
MSM:
recommends that a test suite plan document be created which includes
this information
MSM:
one question is how do we expect test cases to come about? Some
implementers will have tests. Do we expect that these test cases will
be contributed?
discussion of directory structure ensures
XInclude has directory for each contributor and
contribution is simply unzipped into that dir.
Appears that this is not always the process.
<MSM> For example: XQuery test suite is at http://dev.w3.org/cvsweb/2006/xquery-test-suite/
MSM: in the case of XML Schema there is a similar lack of
interoperability because processors can do whatever they like in
acquiring components. The schema test suite specifies the instance and
schema docs for a test. How the processor is told about these, is
outside of the test. He is nervous to stress that SML and SML-IF are
two different things and the the only way that we test SML is via...
kumar: wanted to identify a place where the test cases could be
added. Single directory where we put test cases.
MSM: Would work up to a certain size. XQuery has upwards of 40K
test suites.
johnarwe: COSMOS has about 130, expects it to go to 150.
kumar: doesn't mean unit tests. Means a complete SML-IF
document. Does not see us writing thousands of these.
ginny: suggests some structure within dir, e.g., testsource
MSM: we will have multiple tests using the same input model
documents
johnarwe: as soon as you use locators you are outside of the
scope of SML-IF. Thinks he heard here's a set of schemas, instances,
and schematrons and then we will bind these for differnet tests. Each
test case will have to be a complete SML-IF document.
MSM: planning not to test locator support?
johnarwe: testing would be meaningless, it is not required to
be processed. You have to treat located documents as if they are not in
the interchange set.
kumar: implementation ignores locator
...SML-IF docs.
pratul:
if you don't understand SML-IF, you can take the documents out
(of the IF doc)
MSM: thinks that we will want a test catalog
johnarwe: one could do the following: the default way to render
the catalog is via SML-IF, if you want to run a test, you can XPath, or
XSLT over SML-IF that uses locators only (or a mix of locators and
embedded).
MSM: what are the results of a test?
kumar: the only thing will be valid or invalid. No portable way
to test the error codes.
johnarwe: notes that the reasons for failing a test (invalid)
can be different
ginny: can map errors between implementations
pratul: concerned that it may be difficult to get interop. This
has been done in the past with small tests.
MSM: believes that it was suggested at one point that we needed
to be more clear about what the result of SML validation was.
johnarwe: yes, this desire was expressed.
MSM: has known specs that have tried to be very explicit about
raising specific errors, and others that left completely implementation
dependent exactly what errors were reported. Thinks that there is a
different view between the two implementers. Provides an example from
schema.
MSM: thinks that in a case where there are multiple errors, you
have no expectation that two implementations will report the same
error. Difficult to construct a test case that has only one way to view
its error.
kumar: SML does not specify the order for validation. This does
depend on how focused the test cases are.
ginny: required to test all valid schematron constraints? or
for schema? What is the expectation? are we required to test this.
kumar:
no. too many combinations.
johnarwe: just testing the additions from SML
<pratul> The interop tests used by
the private WG (that defined SML before submitting to W3C) are
available at http://serviceml.org/200701-Interop-b.zip
MSM:
is an SML implementation required to support schematron and ditto for
XML schema 1.0
johnarwe: required to support but a particular input might not
exercise requirements
MSM: process document says that we must show that we have two
independent interoperable implementations of each feature. Not sure
where we should draw the line. For example, don't want to have to do
full schema conformance checking. Yet nervous about saying that we
won't do any of this.
MSM:
Feels that he needs to ask around about this. How did XQuery and XSLT
do about schema awareness?
johnarwe: what did Schema do about its XML awareness?
MSM: schema doesn't check well formedness because input is an
infoset. All sorts of assumptions in schema about well formedness.
Possible that the test suite is inadequate.
MSM: Wants to know more about what different WGs have done
(that have done a good job with tests).
MSM: does not like the idea that SML produces one bit of
information
johnarwe: that's just the limit of interoperability
johnarwe: would feel comfortable for interop testing if the
reporting could be validated by a human as valid or invalid for the
same reasons.
MSM: given standard format for test cases and a catalog for
results it will be possible to produce results that can be eyeballed by
humans. And we have a small enough number of tests that this is
reasonable.
ACTION: kumar to draft a test plan
and send to the group for discussion at the 4/17/08 meeting [recorded
in
http://www.w3.org/2008/04/02-sml-minutes.html#action02]
<trackbot-ng> Created ACTION-179 -
Draft a test plan and send to the group for discussion at the 4/17/08
meeting [on Kumar Pandit - due 2008-04-09].
group discussed (1) directory structure, (2) how to
package tests, (3) error message compatibility issues, and (4) to what
degree should we test schematron and schema
MSM: would like a catalog and would like a standardized format
for reporting test results
The group had a clarifying question on acyclic and
made no decision as a result of the discussion
MSM: would like (3) to be and what format do we use to allow
tests report test results
(revised) group discussed (1) directory structure,
(2) how to package tests, and (3) how are test results characterized,
(4) in what format do we allow test reports, and (5) to what degree
should we test schematron and schema