On Mar 16, 2008, at 8:18 AM, Doug Schepers wrote:
Hi, Maciej-
I'm taking this discussion to www-archive, since it affects a lot of
groups and a lot of interests. I'm BCCing related groups (HTML,
SVG, MathML, CDF, XHTML2), but discussions should take place on www-
archive. (This is part of a couple of long threads that started on
blogs and IRC, and were continued on public-html; I'd suggest that
those interested in this review those threads. [1][2])
I disagree with moving this discussion to www-archive, since it is
directly related to the discussion of allowing SVG in text/html that
is happening on public-html.
To broaden the dialog to other audiences, so I'm going to try to
summarize the issue along the way. Please correct me if I fall
astray, or expand on points. I'm going to talk about SVG, but many
of the same issues may apply to MathML.
Maciej Stachowiak wrote (on 3/16/08 1:12 AM):
HTML has the feature of two serializations: a classic serialization
that is error-tolerant, and an XML-based serialization that has
draconian error handling. These have different costs and benefits,
ultimately it is a benefit to HTML authors that they have a choice.
I think SVG deserves to have this feature as well, there's no
reason it should fall short of HTML in this regard. Supporting SVG
inline in text/html seems like a good opportunity to add this
feature to SVG.
There are 2 scenarios here: SVG inline (Compound Documents by
Inclusion), and external SVG (Compound Documents by Reference).
Your proposal is that, in order to facilitate inline inclusion of
SVG into text/html, changes should be made to other uses of SVG,
including those intended for SVG-only UAs. I'll ask you to detail
what changes would be required, but the topic of hottest debate so
far is the ability to use unquoted attribute values in SVG, when
there are no spaces or other ambiguous characters. This would look
something like this:
<circle id=circle_1 class="category1 medium" cx=75 cy=25 r=20
fill=orange stroke=red stroke-dasharray="3 5" />
You characterize this as non-draconian error handling; for the
purpose of authoring conformance, would this fragment indeed be in
error (and therefor in need of error recovery behavior), or would
this be legal syntax?
I don't see a proposal, a characterization, or indeed any reference to
specific syntax issues in my remarks above. The rest of your
discussion is all about unquoted attributes, which is something I
didn't mention at all, and certainly not the same thing as tolerant
error handling. But if you would like me to expand on my remarks, I'd
be glad to do so on public-html.
- Maciej
A counter-proposal is that SVG retain its existing serialization
(meaning, among other things, that it would require quoted
attributes), for inline (in both XHTML and text/html) and for
external SVG documents (standalone or externally referenced). I'm
going to argue for this proposal, as a Devil's Advocate, but I'm
actually open to any proposals that can be demonstrated as workable.
Given past author practice of copy-paste authoring, SVG that is
written inline in HTML is unlikely to stay there exclusively; it
stands a very good chance of being extracted and reused, be it
referenced as an external image, or as standalone content in a
mobile SVG-only viewer, or pasted into a graphical SVG authoring
tool. Content that breaks from one UA to another is too brittle,
and would fracture the SVG market. So, I agree with you here, that
we shouldn't burden users, authors, and vendors with multiple non-
interoperable serializations.
As you mention, there are indeed known costs to any change in the
written form of SVG (the serialization), notably incompatibility
with all existing SVG viewers, authoring tools, automatic
generators, and XML processors up and down the entire toolchain in
general. There is an enormous infrastructure investment involved in
an enterprise like this, which affects not just SVG and MathML, but
XML in general, and which would require substantial costs in money,
developer hours, and time to market.
On mobile devices in specific, this would require implementors to
write a new parser (which has yet to be shown as practical on the
Web in general, or with SVG in specific); this may not be a burden
for Apple (or certain other desktop browser vendors), which has
already invested resources in an HTML UA, but it would certainly be
for multiple other mobile vendors (those not on the iPhone), who
have a very large (in the upper hundreds of millions) existing SVG
UA deployment, and this could have a substantial cost for them.
For desktop browsers, your proposed content would not work in the
most widely deployed SVG UA plugin for Internet Explorer (the Adobe
viewer, which though no longer maintained, is still the best and
fastest overall SVG UA for browsers); again, I can see this being a
benefit for Apple, but not necessarily for SVG. If Microsoft were
to pledge and deliver resources for adding high-quality native SVG
support (at least as good as Firefox, Opera, and WebKit), this would
certainly alleviate my concerns about desktop browsers.
With the XML serialization of SVG (the way it is now), authors are
already able to do inline SVG in the XML serialization of HTML
(XHTML); this works even in Internet Explorer using the Adobe
viewer, by use of "XML data islands"... and in fact works equally
well in text/html and XHTML (though it has to be served as text/
html, a known problem in IE in general).
Since it's not clear what changes would have to be made to SVG, your
proposal may or may not affect existing SVG content that is intended
for insertion into text/html; that is, presumably content can be
made that works in all SVG UAs, following the existing rules for
SVG. So, this is a neutral point... neither proposal would break
existing content from existing SVG authoring tools and generators.
Still, it would be best to get feedback from known tool vendors that
produce SVG, such as Inkscape, Illustrator, CorelDraw, GraphViz,
Visio, and others (I'm probably forgetting someone major, sorry).
As far as I know, there has been no implementation of the HTML5
parser by a major browser vendor (though there have been test
implementations by others). The algorithm is relatively untested on
the vast body of existing HTML content on the Web, and completely
untested for SVG content. I'm concerned that this is taking an
unnecessary risk with SVG's future, in spite of its already growing
appeal and popularity across a variety of platforms, since it is
competing with similar proprietary technologies. I argue that this
is a bad time to be taking risks for what I see as a largely
theoretical benefit to authors.
As I've stated before, in 8 years of active participation in the SVG
community, I've heard no serious request for unquoted attributes; by
contrast, I have heard many people extolling the benefits of the
cleaner XML model in contrast to the more complex HTML legacy. As
it happens, SVG makes liberal use of spaces in attributes, much more
so than HTML, so the analogies between the two are not quite
accurate; the majority of SVG content uses not the basic shapes like
circles, lines, rectangles and ellipses, but the more complex shapes
represented by paths, polygons, and polylines, and those elements
use attribute values most clearly represented by space-separated
coordinate pairs. It's not clear that there would be a substantive
gain by authors in real-world content by excluding attribute value
quotes. Since SVG doesn't have the legacy content of HTML content's
unquoted attributes, there is little need to extend the same rules
to SVG as are necessary for HTML.
You state that "SVG deserves to have this feature as well" and that
"there's no reason it should fall short of HTML in this regard" as
if unquoted attributes are innately a clear benefit, rather than a
burden that implementors of HTML vendors and authors have to put up
with, because vendors have to deal with legacy content, and authors
have to maintain content they didn't create; SVG authors have never
had to deal with this before, and it's really not clear they want
to. You represent this as an authoring benefit, but I have not seen
the evidence that this is something authors are requesting. I may
be wrong, but I'd like to be proven wrong, rather than take it on
faith.
In summary, as I see it, my proposal works today in a larger number
and wider variety of user agents, and imposes less burden on
vendors, and arguably on authors.
Could you summarize all the changes that you would require for SVG
UAs and authors to implement your proposal, and give the benefits
and rationales for each?
[1] http://lists.w3.org/Archives/Public/public-html/2008Mar/0039.html
[2] http://lists.w3.org/Archives/Public/public-html/2008Mar/0124.html
Regards-
-Doug Schepers
W3C Team Contact, SVG, CDF, and WebAPI
