See also: IRC log
<trackbot> Date: 22 August 2013
<scribe> scribenick: nigel
glenn: volunteers to chair for next 2 weeks
mike: submission to W3C was not
what we had drafted in appendix C, so we got repeat
comments.
... one other comment: lose hyperlinks and references
... taken care of by submitting what we drafted in appendix C
and linking to IRC instead of bracketed symbol.
<scribe> ACTION: mike to work with phillipe to deal with this [recorded in http://www.w3.org/2013/08/22-tt-minutes.html#action01]
<trackbot> Created ACTION-192 - Work with phillipe to deal with this [on Mike Dolan - due 2013-08-29].
sean: there's time pressure to get this done.
mike: it's a gating factor for publication
sean: this is the only thing in the way of 2nd edition publish
glenn: the second item where they
say "encoding considerations ... 8 bit" is not consistent with
our appendix C
... we've updated it so they need to be using the right version
of the appendix first.
mike: that's not how it works. Someone has filed with the old 1.0 text rather than the 2nd edition draft.
glenn: agrees.
sean: that can go ahead whilst I'm away please.
glenn: andreas_irt and nigel have both submitted new messages, at least andreas is.
andreas: the whole area needs
some clarification or guidance. Good email discussions but not
sure how transparent and clear it is for everybody.
... want to make clear how much this relates to what is already
specified in XSL-FO.
glenn: want to review the new
comments received today. I have not seen yet anything that I
think is a substantive or technical issue in the spec.
... most of it has to do with getting on the same page
regarding existing assumptions and assumptions that are not
specifically documented.
... for example we've assumed a common understanding of the CSS
understanding of font size which is mostly the same as
XSL-FO.
sean: might we need better explanatory text in v2?
glenn: we have general language
in the font size section where it refers to the fontSize
property express scaling. It's more of an indirect expression
with hidden steps.
... it would certainly be possible to clarify that in some
informative material. I'm reluctant to try to document all of
the details.
... to have a full understanding of this would require its own
appendix probably.
sean: has thought for a while
that some of the other W3C specs have an informal 'document
zero', and we could probably benefit from that.
... some sort of more motivational stuff that explains the more
arcane parts. The spec is opaque if you have to dig through
XSL-FO and SMIL.
... it could be worth us thinking about that and maybe getting
some volunteers, perhaps on the wiki. If tehre's enough
material we could publish it as a report or something
similar.
... So there's currently nothing that's generated an issue that
we need to respond to?
glenn: I need to review today's emails.
mike: glenn I sent you a private email this morning with archealogy regarding lineHeight.
glenn: yes, lineHeight is an even more dense subject.
mike: I provided lots of quotes from XSL for your pleasure.
glenn: the new CSS3 font spec has an elaboration of a variety of issues that weren't elaborated in CSS2.
sean: is that a model we could use for future reference?
<glenn> http://www.w3.org/TR/css-fonts-3/
glenn: I haven't reviewed it for further discussion of the proccess of translating the fontSize property into scaling in the font rasteriser so I don't know if it helps.
sean: it would be useful to know
if we could construct something more useful than just XSL-FO
refs
... for the rest of the time today agenda is profiles.
glenn: I had added some text into
the PER for 2nd edition regarding resolving percentages and
related to the case where there's no root element.
... I now think that text is wrong. There's always a root
element for every element that has to reference a computed
value for fontSize.
... in this case inheritance works on the intermediate
synchronic document, and the root is a region element. So if
there's a percentage fontSize on the region element for
... fontSize then it would get resolved via the initial value
mechanism. I plan to revise the change in SE to address
that.
nigel: apologises for mistaken interpretation and response on behalf of the group.
glenn: it did point out the difference in interpretation so it was useful.
pierre: maybe just to make sure, Mike put together a zero point status of where we are. Do we all agree that this captures where we are?
sean: it was a fair assessment as far as I could tell.
glenn: mike, sean and I discussed
those matters in detail in private as it turned out there were
some differences in understanding, so
... it does represent a consensus between the three of us at
this point.
mike: yes, the three of us are in
agreement.
... some of it is obvious, some of it is not.
nigel: has contributed an EBU view.
sean: any disagreements with the mike's summary?
all: silence.
andreas: one question to add. One
of the main missing features is a good mechanism to specify
document conformance.
... is it taken as a starting point that the current mechanism
is a base for that or is a different approach possible?
sean: anything is possible at the
moment. What we had didn't do what people wanted it to do. We
may need a completely different mechanism. Anything is on the
table.
... we first need to understand what we want and if the tools
we have today are adequate.
... If we don't know what we want the mechanism to do then
we'll never get anywhere.
... the goal today is to understand what we need to do.
nigel: outlines EBU perspective
sean: parameterising existing features
glenn: like some of the CSS query work
sean: concept of 'supports' to
add more rich boolean logic around features.
... which begs the question: should we take on the entire model
of CSS media queries.
glenn: certainly a question to consider. especially when mapping to CSS/HTML. Do you pass through to the HTML/CSS engine?
sean: does anybody think about
subsetting or parameterising anything other than rendering and
layout aspects of TTML?
... if we were to adopt CSS media queries we'd have to extend
the vocabulary of what you can express inside a query.
glenn: there are a number of
types of parameters e.g. colour, opacity, resolution, scaling,
clipping capabilities etc.
... also for example inter-synchronic scrolling capabilities.
we have this vague notion of having the implementation sort of
auto-scroll between synchronic documents.
... I can see a variety of parameters being defined. the top
level bit is to take note that a boolean definition of a
feature is not adequate.
sean: another thing that has come up is the ability to adapt the content to different presentation aspect ratios. We can't subset the styles for different environments.
nigel: there was discussion of locally-supplied display characteristics.
glenn: had specced out the idea
of a composition method to express when making reference to
multiple profiles.
... right now we had a particular algorithm that involves
overwriting or augmenting what was previously expressed in
other profiles.
... it depends on order whether features are required or
optional. What we don't have is ANDs and ORs instead of
overrides.
... this is suggesting that we generalise the way that we
combine profiles. The second part of this which is different is
a possible formal system for comparing profiles.
... is the intent for use in profile designer or used during
runtime for example if the processor needs to determine if two
profiles have a subset relationship.
sean: the practical thing is that
designers of equipment want a way to advertise what their
processor will accept or not accept.
... in broad perspective its a 2 part contract. Doc says "i
need this", Processor says "if doc looks like this then it can
work"
... we don't have a way to express the combination of features
to create a pass into all relevant bits of equipment.
... we also need to know 'does this document meet this
profile'.
glenn: we need to develop the
idea of document or content profiles, and processor profiles.
What we have today is neither of those exactly.
... it's closer to a processor profile. There is some language
in TTML today regarding claims of conformance.
... when we talk about profiles we need to contextualise our
use of that word so that we're on the same page.
nigel: agrees.
sean: that's what 'profile' means
today.
... any other use of the word profile is an extrapolation
relative to the current profile.
... what else are we going to add to our toolkit?
glenn: we defined something in
3.3 called 'Claims' that refer to an implementation compliance
statement, relating to documents and processors
independently.
... what we talked about is a bit similar to a processor
compliance statement, so I'd invite people to review that
language.
sean: that's informal.
glenn: that's right. The suggestion is to formalise that process.
sean: do we need new mechanics to do this?
glenn: it's a reasonable question to ask. We should re-use the mechanism to the extent that that makes sense rather than chuck it out wholesale.
andreas: For subsets the current
mechanism of TTML in defining a document type couldn't be used.
TTML is defined in terms of the XML reduced infoset
... so that restricts how a TTML document could be authored.
This mechanism could be used also for subsets.
sean: so you're thinking of something like a schema?
andreas: yes. In XMLSchema 1.0
not all of the constraints can be expressed. Infosets go beyond
the schema. It might not necessarily be processable.
... Sugggests using XML 1.1 to express the profile.
sean: we need a way of formally
evaluating a document instance against a profile.
... it may not always be possible, if its NP-incomplete in
terms of processing.
nigel: points 8 and 9 in the doc.
glenn: there are some features in
the current TTML spec that are not schema-validatable. We've
been adding requirements into the various profiles.
... e.g. SDP-US specifies requirements on how many regions,
lines, captions, characters can appear at a given time.
... they're not related to performance and are not
schema-validatable. It's unlikely that all requirements can be
schema-validated.
sean: some of the codec
specifications have a formal reference decoder model for
testing.
... html is going that way, and some of the css
nigel: this is the only way I know of to do it.
sean: one of the ideas is a
service to provide a TTML document and a query 'what does it
look like at time X' and get an image back.
... this is like the CSS test mechanism. We could beef up our
test suite to be a lot more like that.
<glenn> https://github.com/skynav/ttv
glenn: re the validator, there
are about 50 additional semantic tests enumerated in the
documentation for that validator as of today.
... that gives an example of some of the semantic tests that
can't be tested with a schema.
sean: that's one thing that we could consider, and are heading in that direction based on the EBU input.
glenn: that's one of the possible ways that we could proceed. We haven't had a plethora of public implementations of TTML e.g. open source.
sean: there are >=3 open
source ones I'm aware of. Adobe has one in their open media
platform profile.
... it is an issue that there aren't that many.
unlike HTML where you know which are the big browsers, we don't have that kind of model today in TTML.
scribe: I don't see any major
browser with a TTML implementation at the moment.
... we need to at least pursue that kind of model to answer the
questions that have been raised.
sean: a registration model for expressing formal semantics of extension features is needed to make extension features useful.
nigel: also the ability to have a fallback 'preferred behaviour' feature that shouldn't block processing.
glenn: this is related to the more complex logic, e.g. an extension or an existing superset feature.
sean: the current values are
'required' and 'optional' and there are combining rules. We
only want one algebra not two, but a better one.
... we have a bit of functionality in the value space and a bit
in the combinator space but not a complete solution in either
space.
glenn: that was driven by the
desire to do something simple.
... I'm all for expanding the algebra of combination and making
it explicit.
sean: I'd agree - I'd prefer to reduce the value space and increase the combination space.
glenn: we have a fixed algorithm
for combining multiple profiles. We also need to combine
specifications within a given profile.
... the third thing is 'disposition'. We only have a
disposition of 'abort' or 'abort unless override'. We probably
want to introduce some others.
sean: something like a Supports mechanism to avoid aborting the entire document in a binary way would work.
glenn: in some sense those are independent. The issue of processors selecting parts of the document based on parameters is independent but related.
sean: agree they're independent
and orthogonal.
... we have the ability to combine profiles, a disposition
model, a smaller value space, the ability to apply them to
processors as well as documents, the ability to validate
... documents and processors.
<glenn> my list: (1) profile combination; (2) feature/extension combination (within a profile); (3) disposition of constraint violation (lack of constraint satisfaction); (4) feature/extension parameterization; (5) runtime selection of constructs, e.g., styles, timing, animation, based on processor capabilities;
<glenn> (6) ability to specify alternative initial values;
mike: agree that overriding
initial values leads to non-interoperable documents.
... unless applications can define initial values.
pal: it's even worse than that because the same document can not conform simultaneously to two profiles with different initial values.
mike: you can't override an initial value and have a TTML-compliant document.
glenn: there's a use case here
that we don't necessarily address: in some of our initial
values they are implementation dependent e.g. font colour
... another is that what we need to make clear is that the
inheritance mechanism should be used rather than changing an
initial value.
... we do have some cases of non-inheritable style
properties.
sean: we also have the problem that we can't place inheritable properties on the root because of the synchronic document structures.
glenn: if we use anonymous regions there's no way for it to inherit from the root.
pal: to Mike's point, we may have seen it in the past but going forward we should do everything we can to discourage changes to initial values or leave them floating.
andreas: agrees. where it's
dependent on implementation, for TTML2 there should be an
initial value.
... if this isn't wanted it must be overridden explicitly in
the document.
sean: to make that more convenient we should allow some styling on the root and have it inherited.
glenn: we still have some non-inheritable properties.
sean: we would need to state them where they need to be used.
glenn: another thing we did not do is, in CSS, for every style property the keyword "initial" can be used. We may want to think about that.
sean: we might also want to think
about is: we only did effectively inline style in CSS, but
didn't use applicative style.
... we could have rules like 'all regions are blue'.
... choice to adopt CSS wholesale or adopt our own syntax. We
decided to cut complexity by not including applicative
styling.
... not prejudging the way that we do it, e.g. using CSS, but
applicative styling has some advantages
<glenn> glenn: will be extremely reluctant to entertain adoption of CSS syntax in any form
sean: anything that hasn't been covered so far?
mike: having a document profile that says 'I'm compatible with profile X, Y and Z' is a useful tool.
sean: some kind of branding mechanism attached to a profile?
mike: that's okay, but at a high level, to express the compatibility with more than one profile is important and powerful.
glenn: that goes back to the combination method for expressing multiple profiles.
mike: yes, or some other implied semantics. to provide a list of profiles that are compatible. Could be just a list or URIs.
pal: +1
sean: next question is how we
achieve that, and also what the outstanding issues are.
... how to make progress in TTML2, or is it too big a piece of
work?
glenn: assuming that we'll tackle pretty much all of the things on my list in TTML2. Not anticipating pushing it into a v3.
all: no alternative views.
andreas: for the solution, make
it clear and useful in operation, using current technologies as
much as possible. And think why the current profile mechanism
wasn't adopted.
... the complexity of the model and how to apply and use it
should be kept as minimal as possible.
sean: agree we need to prevent misuse
glenn: We need some user guidance or tutorials to go along with the specification, without burdening the spec with being a user guide.
sean: that's what I meant re 'document zero'.
glenn: we don't want to turn the
profile mechanism into an alternative schema mechanism. We
should reuse those tools.
... we didn't adopt schemas because we wanted to support
multiple schema systems, which made the abstraction better but
the practicalities harder.
sean: also schemas weren't rich enough to express everything we wanted. We may still face that issue.
andreas: there are some semantic
constraints that can't be expressed in XML1.1.
... the document structure and normal rules could be expressed
in these technologies and its worth thinking about to make the
schema document normative instead of informative,
... even if they give some false positive messages. So if
there's an error with a schema then the document definitely
doesn't conform.
... A lot of constraints and features of TTML really sit well.
Some others like timing don't fit the model so well, when
dependent on the intermediate synchronic document.
sean: can there be a profile
sub-team to advance this aspect of the spec? i.e. to specify
content profiles.
... the validation of the profile model is the thing we need it
to say.
pal: sounds like a reasonable plan. I can try to capture using the current TTML2 draft what the profile definition would be for those two proposed profiles.
sean: and where there are things that you can't express like performance, profile intersection, start with the profile, and if you document how you'd like to express that then it can inform the architect of the technical features.
pal: happy with that.
... One thing i'd like to make progress on is the observation
that in TTML1 today that if a specification subsets a standard
feature that specification has to define a brand new
features
... I'm not unhappy with that - it's straightforward but
unambiguous.
... The proposal is to be able to define that a feature is a
subset of another feature.
glenn: suggests using the OR
construct.
... in a thread with nigel two possible things, one is an OR
construct, the other an attribute 'subsetOf' declaring the
subset relationship.
... the OR is more general but I could see use of subset too.
It's worth thinking about the usage needs for each of
those.
... you can't define a feature in a profile document, only an
extension. The term 'extension feature' may be closer to the
mark but the term 'extension' is more precise.
pal: encourage people to look at
the existing issues and file new issues if they aren't covered,
and relate them to the main profile issue.
... returning to the discussion about a profile document: is it
possible in the latest TTML2 draft to create a profile document
to import another profile document and specify only exceptions
or differences?
glenn: right now we use an
override logic for combining multiple profile elements. See
section 5.1 (?) where superset and subset profiles are
discussed.
... So we don't need to repeat every feature.
<scribe> ACTION: pal to come up with a sample profile definition document for at least the proposed text profile [recorded in http://www.w3.org/2013/08/22-tt-minutes.html#action02]
<trackbot> Created ACTION-193 - Come up with a sample profile definition document for at least the proposed text profile [on Pierre-Anthony Lemieux - due 2013-08-29].
glenn: we don't have a way to
specify that a feature must be used in a document. Required
implies support in a processor only.
... I have mixed thinking about that at this point. When people
define profiles should they just define schemas that impose
these requirements.
pal: at some point you cross the line into redefining XML schemas.
glenn: the boundary point is where constraints like 'must' and 'may' appear.
sean: if parameterised features point to schemas that could be the way to go.
glenn: interesting idea.
pal: there are some aspects of
the prose that will never be captured completely mechanically.
The objective shouldn't be for the profile document to allow
full validation by a processor.
... in my mind it's more a convenient way to understand the
relationship between profiles.
glenn: agree.
... if we define content profiles, what are the uses - in the
authoring processes as metadata to help label the document, or
at runtime in a content processor to perform validation.
nigel: the former is more likely in broadcast environments where a whole ecosystem needs to be 'trusted'.
<glenn> i must leave the call early... ttyl
pal: I guess the vast majority of applications will simply want to look at a label in the head of the document to see which profiles are complied with.
andreas: agree.
nigel: agree.
<glenn> ok
sean: good discussion.
sean: encourage people to scribble on wiki page, and I'll gather when I return.
This is scribe.perl Revision: 1.138 of Date: 2013-04-25 13:59:11 Check for newer version at http://dev.w3.org/cvsweb/~checkout~/2002/scribe/ Guessing input format: RRSAgent_Text_Format (score 1.00) Found ScribeNick: nigel Inferring Scribes: nigel Default Present: Sean, glenn, pal, mike, nigel, +49.893.aaaa, Thomas_Bause_Mason Present: Sean glenn pal mike nigel +49.893.aaaa Thomas_Bause_Mason WARNING: No meeting chair found! You should specify the meeting chair like this: <dbooth> Chair: dbooth Found Date: 22 Aug 2013 Guessing minutes URL: http://www.w3.org/2013/08/22-tt-minutes.html People with action items: mike pal[End of scribe.perl diagnostic output]