MathML-in-HTML5
r...@maths.uq.edu.au
MathML-in-XHTML that we already support). I have a patch that serves
the dual purpose of showing where things are going and the issues to
ponder about.
Here is a
[screenshot] https://bugzilla.mozilla.org/attachment.cgi?id=239771
which is a _live_ rendering of this testcase:
[mathml-in-html] https://bugzilla.mozilla.org/attachment.cgi?id=239769
Those interested in following this up can see bug 353926:
https://bugzilla.mozilla.org/show_bug.cgi?id=353926
Quick background:
=================
At the Firefox engineering meeting in Mountain Views (last December
2005), I pleaded that we enable MathML in HTML5 to advance the cause
of MathML, which is so far locked in a XHTML/XML world that does not
seem to be going anywhere in terms of display content as opposed to
data (witness the WHATWG effort -- http://www.whatwg.org). Those to
whom I spoke included dbaron, hixie and sicking, and they welcomed the
suggestion, asking for a broader discussion. Hixie raised the caveat
that MathML elements should still remain in the MathML namespace. He
e-mailed me a while ago about a discussion on this matter in the
WHATWG mailing list, which can be seen here
http://listserver.dreamhost.com/pipermail/whatwg-whatwg.org/2006-June/thread.html.
That discussion is however too broad and involves tangential issues such as
inventing another syntax, etc. My original take was simply to enable
MathML+HTML, in the same vein as we have MathML+XHTML. I think MathML
is suffering from having to fight the battle for adoption of XHTML as
well. As a niche technology, it does not have the means to be engaging
a fight. What it simply needs is MathML-in-HTML. W3C failed to
recognise that it could retrofit MathML in HTML -- see this archived
post for some insight:
http://groups.google.com/group/netscape.public.mozilla.mathml/msg/4d58c35217afcb54?dmode=source
But HTML5 being shepherded by WHATWG could provide the right framework
from this to happen now.
I have finally been able to code this up (while keeping MathML
elements in the MathML namespace). I attached the patch I had so far
in bug 353926.
Design & Technical issues:
==========================
How does MathML-in-HTML5 work?
We support MathML-in-HTML5 when these two conditions are met:
1. The DOCTYPE of the document says so. If yes, we enable
MathML entities (TODO) and flag mMayHaveMathML in the HTML content sink.
2. And either a) OR b) is met:
a) <html> has the MathML namespace as the value of an attribute with a
prefix, e.g., <html xmlns:m="http://www.w3.org/1998/Math/MathML">.
In this case, we cache the prefix "m" in mMathMLNameSpacePrefix,
and we intercept all <m:tag> in the document and create
MathML content nodes for them.
b) MathML fragments are in the document as
<math xmlns="http://www.w3.org/1998/Math/MathML">
...
</math>
In this case, we intercept all non-HTML elements inside the <math> tag
and create MathML content nodes for them.
Issues:
1. Tag soup: we understand that we are exposing ourselves to this.
2. a) What about CSS matching rules? From the Style System point of view,
the document is still HTML, but <m:math> is in the MathML namespace. We
might have to special case MathML-in-HTML5 in the Style System as well.
b) The second option raises an issue with HTML-in-MathML, e.g.,
<math xmlns="http://www.w3.org/1998/Math/MathML">
<b>bold</b>
</math>
We don't intercept the <b> in this case. Hence, even though it is
HTML-in-MathML without an explicit XHTML namespace for <b>,
the HTML sink
will give <b> a HTML content node. This is not really XHTML friendly.
On the other hand, we don't want to be an XML parser either... These
are conflicting objectives. We need to decide what to do. We may agree
to only support tags with prefixes as in a), or also keep b) knowing
that it has this XHTML unfriendly behavior.
---
RBS
----------------------------------------------------------------
This message was sent using IMP, the Internet Messaging Program.
Ian Hickson
>
> Hixie raised the caveat that MathML elements should still remain in the
> MathML namespace.
I meant in the DOM, I didn't mean in the markup. I don't think we should
have any namespace declarations or namespace prefixes in text/html; I
would just have the HTML parser always support the MathML elements, in
the same way that it supports any random unknown element today, except
that when it sees a MathML element it puts it into the MathML namespace in
the DOM rather than the XHTML namespace.
I really don't think we want to introduce namespace prefixes or namespace
declarations into tag soup. I think that would be a big mistake.
--
Ian Hickson U+1047E )\._.,--....,'``. fL
http://ln.hixie.ch/ U+263A /, _.. \ _\ ;`._ ,.
Things that are impossible just take longer. `._.-(,_..'--(,_..'`-.;.'
L. David Baron
> On Sat, 23 Sep 2006 r...@maths.uq.edu.au wrote:
> >
> > Hixie raised the caveat that MathML elements should still remain in the
> > MathML namespace.
>
> I meant in the DOM, I didn't mean in the markup. I don't think we should
> have any namespace declarations or namespace prefixes in text/html; I
> would just have the HTML parser always support the MathML elements, in
> the same way that it supports any random unknown element today, except
> that when it sees a MathML element it puts it into the MathML namespace in
> the DOM rather than the XHTML namespace.
>
> I really don't think we want to introduce namespace prefixes or namespace
> declarations into tag soup. I think that would be a big mistake.
Agreed.
-David
--
L. David Baron <URL: http://dbaron.org/ >
Technical Lead, Layout & CSS, Mozilla Corporation
Paul Topping
would be necessary. However, if MathML is going to work in HTML that
isn't declared as HTML5 (not clear to me from this thread), then the
document would be poorly specified without it, IMHO.
At the risk of enciting an anti-Microsoft backlash, I should remind some
on the list that IE has covered this territory before. They already have
a mechanism for declaring XML islands in HTML that seems to work just
fine. Of course, Mozilla won't be interested in duplicating IE's way of
associating a plugin as the renderer of the namespace in the document.
IMHO, it doesn't belong there anyway. It is better (ie, more secure) to
keep such associations out of the content.
Paul Topping
Design Science, Inc.
www.dessci.com/mathplayer
> -----Original Message-----
> From: dev-tech-ma...@lists.mozilla.org
> [mailto:dev-tech-ma...@lists.mozilla.org] On Behalf
> Of Ian Hickson
> Sent: Saturday, September 23, 2006 2:06 PM
> To: r...@maths.uq.edu.au
> Cc: dev-tec...@lists.mozilla.org;
> dev-tec...@lists.mozilla.org
> Subject: Re: MathML-in-HTML5
>
> On Sat, 23 Sep 2006 r...@maths.uq.edu.au wrote:
> >
> > Hixie raised the caveat that MathML elements should still remain in
> > the MathML namespace.
>
> I meant in the DOM, I didn't mean in the markup. I don't
> think we should have any namespace declarations or namespace
> prefixes in text/html; I would just have the HTML parser
> always support the MathML elements, in the same way that it
> supports any random unknown element today, except that when
> it sees a MathML element it puts it into the MathML namespace
> in the DOM rather than the XHTML namespace.
>
> I really don't think we want to introduce namespace prefixes
> or namespace declarations into tag soup. I think that would
> be a big mistake.
>
> --
> Ian Hickson U+1047E
> )\._.,--....,'``. fL
> http://ln.hixie.ch/ U+263A /, _.. \
> _\ ;`._ ,.
> Things that are impossible just take longer.
> `._.-(,_..'--(,_..'`-.;.'
> _______________________________________________
> dev-tech-mathml mailing list
> dev-tec...@lists.mozilla.org
> https://lists.mozilla.org/listinfo/dev-tech-mathml
>
Ian Hickson
>
> If MathML is considered a subset of HTML5, then no namespace declaration
> would be necessary. However, if MathML is going to work in HTML that
> isn't declared as HTML5 (not clear to me from this thread), then the
> document would be poorly specified without it, IMHO.
As far as HTML5 UAs are concerned, declaring HTML as HTML5 consists of
labelling it as text/html. It isn't clear to me what you would consider
HTML that isn't declared as HTML5. With the exception of quirks which are
required for compatibility with de facto standards that disagree with de
jure standards, HTML has no practical versioning story -- all features
work in all documents, regardless of the official "version" of HTML used.
> At the risk of enciting an anti-Microsoft backlash, I should remind some
> on the list that IE has covered this territory before. They already have
> a mechanism for declaring XML islands in HTML that seems to work just
> fine.
XML data islands don't form part of the parent DOM (they are "islands", as
opposed to part of the document). I'm not sure how wrapping <xml> tags
around the MathML content would help. :-)
> And, I should have added that without a namespace declaration there
> would be no way to differentiate different versions of MathML. While
> most MathML instances are now MathML 2.0, the MathML 3.0 effort is just
> now starting up.
Why would you need to distinguish them? MathML2 is a superset of MathML1,
and (for all intents and purposes) any compliant MathML2 UA can process
any compliant MathML1 content. I would assume that this would continue to
be the case; if not, then this is IMHO a problem with MathML3.
Note that the namespace declaration can't currently distinguish between
MathML1 and MathML2, I don't see any reason why MathML3 would change this.
Paul Topping
would be no way to differentiate different versions of MathML. While
most MathML instances are now MathML 2.0, the MathML 3.0 effort is just
now starting up.
Paul Topping
Design Science, Inc.
www.dessci.com/mathplayer
> -----Original Message-----
> From: dev-tech-ma...@lists.mozilla.org
> [mailto:dev-tech-ma...@lists.mozilla.org] On Behalf
> Of Paul Topping
> Sent: Saturday, September 23, 2006 3:39 PM
> To: Ian Hickson; r...@maths.uq.edu.au
> Cc: dev-tec...@lists.mozilla.org;
> dev-tec...@lists.mozilla.org
> Subject: RE: MathML-in-HTML5
>
> If MathML is considered a subset of HTML5, then no namespace
> declaration would be necessary. However, if MathML is going
> to work in HTML that isn't declared as HTML5 (not clear to me
> from this thread), then the document would be poorly
> specified without it, IMHO.
>
> At the risk of enciting an anti-Microsoft backlash, I should
> remind some on the list that IE has covered this territory
> before. They already have a mechanism for declaring XML
> islands in HTML that seems to work just fine. Of course,
> > --
> > Ian Hickson U+1047E
> > )\._.,--....,'``. fL
> > http://ln.hixie.ch/ U+263A /, _.. \
> > _\ ;`._ ,.
> > Things that are impossible just take longer.
> > `._.-(,_..'--(,_..'`-.;.'
Boris Zbarsky
> I meant in the DOM, I didn't mean in the markup. I don't think we should
> have any namespace declarations or namespace prefixes in text/html; I
> would just have the HTML parser always support the MathML elements
I assume we have data that shows there would be no collisions with random
user-defined tag names in random pages? Including intranets?
> I really don't think we want to introduce namespace prefixes or namespace
> declarations into tag soup. I think that would be a big mistake.
I agree with this, for what it's worth.... But perhaps we do want a way to
explicitly flag tag-soup documents as "this document uses MathML".
-Boris
Ian Hickson
>
> Ian Hickson wrote:
> > I meant in the DOM, I didn't mean in the markup. I don't think we should
> > have any namespace declarations or namespace prefixes in text/html; I would
> > just have the HTML parser always support the MathML elements
>
> I assume we have data that shows there would be no collisions with random
> user-defined tag names in random pages? Including intranets?
Nope. Just blind faith. (Well, we have some evidence for the Web at large,
but nothing substantial, only a billion pages or so. I'm working on a more
substantial survey but that still won't cover the intranets.)
We didn't check that <canvas> wouldn't cause clashes, either.
> > I really don't think we want to introduce namespace prefixes or
> > namespace declarations into tag soup. I think that would be a big
> > mistake.
>
> I agree with this, for what it's worth.... But perhaps we do want a way
> to explicitly flag tag-soup documents as "this document uses MathML".
I don't see why. We don't want a flag for when people can use the storage
APIs. Or when they can use <img> elements. Or whatever.
Boris Zbarsky
> We didn't check that <canvas> wouldn't cause clashes, either.
I see. I had assumed that we in fact had.
> I don't see why. We don't want a flag for when people can use the storage
> APIs. Or when they can use <img> elements. Or whatever.
True, because those are very unlikely to collide with random stuff the pages are
doing (e.g. the storage APIs are using fairly long names that are unlikely to
collide with page-defined functions and variables).
If we think MathML has a similarly low risk of collision, great.
-Boris
David Carlisle
Ian
> XML data islands don't form part of the parent DOM (they are "islands", as
> opposed to part of the document). I'm not sure how wrapping <xml> tags
> around the MathML content would help. :-)
The syntax Paul was referring to here wasn't the <xml> convention, but
the ability in IE to have (explicitly prefixed) XML elements within an
HTML document with rendering controlled by an external component,
but _without_ any other flag at that point in the in the markup, such as
<xml> or <object> etc.
In the IE implementation you need to have an <object> in the head
pointing at the particular rendering component, which is fairly horrible
and also, you need to declare the namespace using (a variant of) an
early working draft namespace syntax using a PI, but as Paul said, those
parts needn't be copied. an example of a document using this syntax is
shown here:
http://www.dessci.com/en/products/mathplayer/author/creatingpages.htm#AnatomyMathPlayerWebPage
By using a different classid you can do the same thing to include
(explicitly prefixed) svg into an htm document and have it rendered by
Adobe's svg viewer, and in principle any other vocabularies (although I
don't personally know of any other implementations of this, except
techexplorer, which is again for MathML).
I'm not sure, having math more or less added directly to html would be
nice in many ways but I'm not sure how well it scales, if you think
people might want to have html+svg+chemml+... then perhaps having an api
that allows processing to be attached to namespaced elements would be
more general. On the other hand that was part of the reason for having
namespaces (and for that matter, xml itself) that people could serve all
sorts of different xml vocabularies and have clients do whatever is
necessary. I suspect part of the reason for "html5" is a feeling that
that never happened and isn't going to be mainstream any time soon, and
that a solution that directly addresses the fixed html vocabulary, with
perhaps two specific extensions such as svg and mathml will in practice
cover the vast majority of browser needs, and other vocabularies can be
transformed to html+.. before being served.
David
Ian Hickson
>
> The syntax Paul was referring to here wasn't the <xml> convention, but
> the ability in IE to have (explicitly prefixed) XML elements within an
> HTML document with rendering controlled by an external component, but
> _without_ any other flag at that point in the in the markup, such as
> <xml> or <object> etc.
Oh, well, as noted earlier, the idea of namespace prefixes in HTML isn't
one that I personally am particularly fond of.
> I suspect part of the reason for "html5" is a feeling that that never
> happened and isn't going to be mainstream any time soon, and that a
> solution that directly addresses the fixed html vocabulary, with perhaps
> two specific extensions such as svg and mathml will in practice cover
> the vast majority of browser needs, and other vocabularies can be
> transformed to html+.. before being served.
I think that's pretty much exactly correct, yes.
Ian Hickson
>
I don't know about "we".
What I would be proposing for HTML5 is just the following list of
elements:
math, mrow, mfrac, msqrt, mroot, mstyle, merror, mpadded, mphantom,
mfenced, menclose, msub, msup, msubsup, munder, mover, munderover,
mmultiscripts, mtable, mlabeledtr, mtr, mtd, maction
...and of those only <math> came up at in the top 1000 elements in my
search of elements on about one billion pages.
According to that same research, <math> is, on the Web, less frequent than
the following elements: <m>, <e>, <rem>, <tab>, <yr>, <prohibits>, <your>,
<lable>, <text-spez>, etc. It was present on less than 0.002% of the pages
the research covered. (To give an idea of scale, <h8> is used on more than
0.003%, so if we avoid <math> because of this, we should probably
introduce <h7> and <h8> into HTML, since we're saying that's an important
enough level to worry about.)
Now, of course, it could be that those 0.002% of pages are all hugely
important and that we'll break the Web in adding this feature. We can't
know until we've tried.
Boris Zbarsky
> According to that same research, <math> is, on the Web, less frequent than
> the following elements: <m>, <e>, <rem>, <tab>, <yr>, <prohibits>, <your>,
> <lable>, <text-spez>, etc. It was present on less than 0.002% of the pages
> the research covered. (To give an idea of scale, <h8> is used on more than
> 0.003%, so if we avoid <math> because of this, we should probably
> introduce <h7> and <h8> into HTML, since we're saying that's an important
> enough level to worry about.)
The last statement doesn't follow, for what it's worth. There's a difference
between "introduce support for tags that currently do nothing in all UAs" and
"don't introduce support for a tag that other UAs do nothing for because it will
make us behave differently from those UAs".
I do agree that it sounds like this won't be too big an issue.
-Boris
Roger B. Sidje
> elements:
>
> math, mrow, mfrac, msqrt, mroot, mstyle, merror, mpadded, mphantom,
> mfenced, menclose, msub, msup, msubsup, munder, mover, munderover,
> mmultiscripts, mtable, mlabeledtr, mtr, mtd, maction
I don't like mlabeledtr very much (I have already expressed my views
about it to folks of the MathML WG), and would hope that they will take
my suggestion for <mtr label="..."> in MathML3. The former is
unnecessarily bloated and doesn't degrade gracefully at all with
renderers that don't support it (not to mention that it is hard to fit
in Gecko's existing table code).
However, your list misses some key tags, in particular leaf tags such as
<mspace/> -- which is sometimes quite useful. Also, <mprescripts/> and
<none/> are needed in <mmultiscripts> (albeit it can be argued that
<none/> is the same as <mrow></mrow> or an empty <mspace/>, but the
differentiation is worthwhile).
In general, I would prefer the list to at least include all the tags
that we already support, and which existing webpages have come to depend
on. This effectively boils down to your list above, excluding
<mlabeledtr>, and including <mspace/>, <mprescripts/>, <none/> and
<mi>, <mn>, <ms>, <mtext>, <mo>. In particular, <mo> is a vital tag as
it is at the heart of those stretchy MathML characters.
Implementation-wise, as this inclusion of MathML-in-HTML5 marks the
beginning of tag soup, it may be that the HTML parser would have to have
some knowledge of leaf tags, so that for example, a stray <mspace>
doesn't become the root of an entire HTML tree... which is later fed to
the hapless MathML engine. (The patch I attached in bug 353926 ignored
the issue.)
---
RBS
On 26/09/2006 3:59 AM, Ian Hickson wrote:
> On Sun, 24 Sep 2006, Boris Zbarsky wrote:
>
> I don't know about "we".
>
> What I would be proposing for HTML5 is just the following list of
> elements:
>
> math, mrow, mfrac, msqrt, mroot, mstyle, merror, mpadded, mphantom,
> mfenced, menclose, msub, msup, msubsup, munder, mover, munderover,
> mmultiscripts, mtable, mlabeledtr, mtr, mtd, maction
>
> ...and of those only <math> came up at in the top 1000 elements in my
> search of elements on about one billion pages.
>
> According to that same research, <math> is, on the Web, less frequent than
> the following elements: <m>, <e>, <rem>, <tab>, <yr>, <prohibits>, <your>,
> <lable>, <text-spez>, etc. It was present on less than 0.002% of the pages
> the research covered. (To give an idea of scale, <h8> is used on more than
> 0.003%, so if we avoid <math> because of this, we should probably
> introduce <h7> and <h8> into HTML, since we're saying that's an important
> enough level to worry about.)
>
Ian Hickson
>
> I don't like mlabeledtr very much (I have already expressed my views
> about it to folks of the MathML WG), and would hope that they will take
> my suggestion for <mtr label="..."> in MathML3. The former is
> unnecessarily bloated and doesn't degrade gracefully at all with
> renderers that don't support it (not to mention that it is hard to fit
> in Gecko's existing table code).
I'm happy to drop/add any tag to this list. Just give me the list you
want.
> However, your list misses some key tags, in particular leaf tags such as
> <mspace/> -- which is sometimes quite useful. Also, <mprescripts/> and
> <none/> are needed in <mmultiscripts> (albeit it can be argued that
> <none/> is the same as <mrow></mrow> or an empty <mspace/>, but the
> differentiation is worthwhile).
I missed anything that wasn't in the table I happened upon in the spec. I
didn't look very closely for the exact table I wanted.
Tell me what tags you want to have and we'll make that the list. You're
the expert. :-)
> Implementation-wise, as this inclusion of MathML-in-HTML5 marks the
> beginning of tag soup, it may be that the HTML parser would have to have
> some knowledge of leaf tags, so that for example, a stray <mspace>
> doesn't become the root of an entire HTML tree... which is later fed to
> the hapless MathML engine. (The patch I attached in bug 353926 ignored
> the issue.)
Don't worry, these tags auto-close when a parent tag is closed.
<foo><bar><baz></foo><quux>
...results in this DOM:
<foo>
<bar>
<baz>
<quux>
For leaf nodes with following siblings, people will have to use end tags,
as in:
<foo><bar></bar><baz></baz></foo><quux></quux>
If we want to start adding actual leaf tags, I'd rather do this in a
second stage, after we have a proof of concept. (I've so far avoided
adding any new tags to the HTML5 parser spec, but eventually there will be
a bunch we have to add.)
We can go from non-empty to empty much more easily than from empty to
non-empty.
Roger B. Sidje
> I'm happy to drop/add any tag to this list. Just give me the list you
> want.
OK.
> For leaf nodes with following siblings, people will have to use end tags,
> as in:
>
> <foo><bar></bar><baz></baz></foo><quux></quux>
>
> If we want to start adding actual leaf tags, I'd rather do this in a
> second stage, after we have a proof of concept. (I've so far avoided
> adding any new tags to the HTML5 parser spec, but eventually there will be
> a bunch we have to add.)
OK, I see.
The other issue are those 2000 entities that MathML has. You said that
you are not a big fan of a namespace thingy on the root <html> element.
Is is okay to assume HTML5 (with its <!DOCTYPE html>) as supporting all
W3C entities _by default_? We have a proof-of-concept of that in View
Selection Source, BTW. It will display any entity it can.
http://lxr.mozilla.org/mozilla/source/content/base/public/nsIDocumentEncoder.idl#125
As VSS has underwent the test of time without major complaints, perhaps
<!DOCTYPE html> could assume that too? If that is agreed, we are all clear.
The other remaining issue might be with style matching because <math>
will then be internally in the MathML namespace whereas the HTML
document is in the none namespace (at present), but we will see how it
goes from there.
---
RBS
Ian Hickson
>
Yeah... Do we really need those? Some of them seem reasonable to add, but
2000 seems like too many for the mnemonic advantage to beat just using
Unicode codepoints...
The problem with adding entities is that a LOT of people do things like
href="/u?aa=foo&ab=foo&ac=foo&ad=foo"
...which today works, but would break if MathML entities were introduced
(since &ac is a MathML entity).
> Is is okay to assume HTML5 (with its <!DOCTYPE html>) as supporting all
> W3C entities _by default_?
Don't do anything based on the DOCTYPE. HTML5 is anything sent as
text/html.
> The other remaining issue might be with style matching because <math>
> will then be internally in the MathML namespace whereas the HTML
> document is in the none namespace (at present), but we will see how it
> goes from there.
I don't see why this would cause any problems.
Roger B. Sidje
>
> The problem with adding entities is that a LOT of people do things like
>
> href="/u?aa=foo&ab=foo&ac=foo&ad=foo"
>
> ...which today works, but would break if MathML entities were introduced
> (since &ac is a MathML entity).
>
That list is so big that trying to hand-pick some and leaving some out
would need another committee...
>>Is is okay to assume HTML5 (with its <!DOCTYPE html>) as supporting all
>>W3C entities _by default_?
>
>
> Don't do anything based on the DOCTYPE. HTML5 is anything sent as
> text/html.
I thought the DOCTYPE was trustworthy -- based on this excerpt from the
HTML5 spec:
"HTML documents that use the new features described in this
specification must start with the string <!DOCTYPE html> and, if they
are served over the wire (e.g. by HTTP) must be labelled with the
text/html MIME type."
If so, it would have meant less conflicts with agreed entities in HTML5.
BTW, for my own information, do you intent HTML5 to be transitional,
almost-standards, or strict? If it is HTML5 (or XHTML5) served as
text/html but put in the XHTML namespace at some later stage (as the
HTML5 implies), it better be strict, no? And that would be driven by the
DOCTYPE detection code. Catch my drift? Or is tag soup going to be in
the XHTML namespace?
If it is strict then maybe entities could be required to have a
semi-colon -- which will then avoid the ambiguities you mentioned above.
Not that I have a position on this (at least as yet). I am just bringing
in some food for thoughts, to accommodate the realistic issues of MathML.
---
RBS
Ian Hickson
> >
> > The problem with adding entities is that a LOT of people do things
> > like
> >
> > href="/u?aa=foo&ab=foo&ac=foo&ad=foo"
> >
> > ...which today works, but would break if MathML entities were
> > introduced (since &ac is a MathML entity).
>
> That list is so big that trying to hand-pick some and leaving some out
> would need another committee...
Not really... I say we just add ApplyFunction, InvisibleComma, and
InvisibleTimes (but not their short aliases).
> > > Is is okay to assume HTML5 (with its <!DOCTYPE html>) as supporting
> > > all W3C entities _by default_?
> >
> > Don't do anything based on the DOCTYPE. HTML5 is anything sent as
> > text/html.
>
> I thought the DOCTYPE was trustworthy -- based on this excerpt from the
> HTML5 spec:
>
> "HTML documents that use the new features described in this
> specification must start with the string <!DOCTYPE html> and, if they
> are served over the wire (e.g. by HTTP) must be labelled with the
> text/html MIME type."
That's an authoring conformance requirement, and has no bearing on
implementations.
> BTW, for my own information, do you intent HTML5 to be transitional,
> almost-standards, or strict?
HTML5 documents starting with <!DOCTYPE HTML> must be in standards mode.
Documents with other DOCTYPEs or no DOCTYPE at all may be in another mode,
as already described in the spec. In due course I may specify quirks mode
and then there'll just be the spec, and no other modes.
> If it is HTML5 (or XHTML5) served as text/html but put in the XHTML
> namespace at some later stage (as the HTML5 implies), it better be
> strict, no? And that would be driven by the DOCTYPE detection code.
> Catch my drift? Or is tag soup going to be in the XHTML namespace?
Not sure what you mean my that. All HTML DOM nodes are (per HTML5) in the
XHTML namespace, irrespective of the standards/quirks thing.
> If it is strict then maybe entities could be required to have a
> semi-colon -- which will then avoid the ambiguities you mentioned above.
That would break back-compat.
Matt Sicker
> On Wed, 27 Sep 2006, Roger B. Sidje wrote:
>
> Yeah... Do we really need those? Some of them seem reasonable to add,
> but 2000 seems like too many for the mnemonic advantage to beat just
> using Unicode codepoints...
>
> like
>
> href="/u?aa=foo&ab=foo&ac=foo&ad=foo"
>
> ...which today works, but would break if MathML entities were
> introduced (since &ac is a MathML entity).
>
Shack, part of the URL contained "&image=foo". Of course, it looked
odd seeing the "&image" part become the imaginary part character (kinda
looks like a dragon), but the URL still worked. This is why I
encourage usage of the semicolon instead...
--
Matt Sicker
William F Hammond
> On Wed, 27 Sep 2006, Roger B. Sidje wrote:
> . . .
>> Implementation-wise, as this inclusion of MathML-in-HTML5 marks the
>> beginning of tag soup, ...
>
> Don't worry, these tags auto-close when a parent tag is closed.
Two points for clarification:
1. There's the old issue, related to dual parsers, of trying to get
Mozilla family user agents to give proper handling of XHTML+MathML
when served through text/html -- following early Amaya practice. (In
the end the W3C HTML WG refused to support this idea and spawned the
mimetype application/xhtml+xml.) It seems that formally correct
XHTML+MathML would now gain coverage as text/html under current WhatWG
thinking, at least when XML namespaces are evident only through use of
the xmlns attribute (which would be ignored in tag soup), i.e., no use
of xml namespace prefixing. Is this correct?
2. Is WhatWG entertaining the idea that off-the-cuff tag soup writers
will generate MathML content that's good enough for Mozilla rendering?
---
In case you don't know:
The W3C Math group has announced that it is beginning to think seriously
about author-level markup for math.
Long term -- say ten years in the future (we've already been at this
for ten years) -- I think author level math additions to the tag soup
vocabulary would work out much better, especially with enhanced CSS
support.
Cheers.
-- Bill
----------------------------------------------------------------------
William F. Hammond Dept. of Mathematics & Statistics
518-442-4625 The University at Albany
hammond At math.albany.edu Albany, NY 12222 (U.S.A.)
http://www.albany.edu/~hammond/ Dept. FAX: 518-442-4731
----------------------------------------------------------------------
David Carlisle
> What I would be proposing for HTML5 is just the following list of
> elements:
>
> math, mrow, mfrac, msqrt, mroot, mstyle, merror, mpadded, mphantom,
> mfenced, menclose, msub, msup, msubsup, munder, mover, munderover,
> mmultiscripts, mtable, mlabeledtr, mtr, mtd, maction
You would beed to include the leaf elements (mi mn mo mtext) otherwise
there'll be no characters in the mathml!, also mspace is pretty
important.
But a more general point I think it's dangerous for a spec to be
profiled by _implementations_. The Math WG activity has just been
restarted at W3C and if there is a need to profile MathMl to
presentation MathML (or a subset thereof) please can it be done _there_
so that there is some chance that mathml authoring tools can be
customised to have options to generate code to match any profiled spec.
> I don't like mlabeledtr very much (I have already expressed my views
> about it to folks of the MathML WG)
Roger, I don't see anything searching for
http://www.w3.org/Search/Mail/Public/search?type-index=www-math&index-type=t&keywords=mlabeledtr&search=Search
I know you've talked to us at conferences etc, but we're all getting old
and if comments aren't on the comment list, then they are likely to get
forgotten over time.
_Now_ would be a really good time to make such comments as we are in the
process of finalising the requirements for what extar features should
be in MathML3, and what if necessary, features should be deprecated.
I don't remember specific discussions about an <mtr label="..."> I
would guess there woul dbe some convern about the label being an
attribute rather than an element restricting the possibilities, but
implementation advice on difficulties on teh current schem woul dbe
taken seriously....
Ian wrote about entities
> Yeah... Do we really need those? Some of them seem reasonable to add, but
> 2000 seems like too many for the mnemonic advantage to beat just using
> Unicode codepoints...
I'd say that it's probably not worth including only a few, it would just
lead to confusion. The problem is that much mathml is generated using
tools and those tools may use entities, and if they do that the user
hasn't much control over which are used, and how to fix things to remove
entities that are not supported in the browser. It would be better to
just get the MathML authoring tools to use characters or character refs
directly and tell the user mathml entities are not supported (but html
ones are)
David
Ian Hickson
>
> Oh, I've seen this problem before; when people would link to Image
> Shack, part of the URL contained "&image=foo". Of course, it looked odd
> seeing the "&image" part become the imaginary part character (kinda
> looks like a dragon), but the URL still worked.
Really? If it still worked, that implies that argument was ignored, no?
Ian Hickson
>
> 1. There's the old issue, related to dual parsers, of trying to get
> Mozilla family user agents to give proper handling of XHTML+MathML when
> served through text/html -- following early Amaya practice. (In the end
> the W3C HTML WG refused to support this idea and spawned the mimetype
> application/xhtml+xml.) It seems that formally correct XHTML+MathML
> would now gain coverage as text/html under current WhatWG thinking, at
> least when XML namespaces are evident only through use of the xmlns
> attribute (which would be ignored in tag soup), i.e., no use of xml
> namespace prefixing. Is this correct?
I'm confused by your terminology.
MathML using namespaces and XML syntax would not, under the WHATWG
proposals here, be formally correct. XML sent as text/html is never
correct per the "WHATWG thinking".
What is being proposed here is a non-XML syntax, to be formally described
in the HTML5 specification, which, went processed by an HTML5 UA, would
generate a DOM that can then be processed per the MathML2 specification.
Per the WHATWG specifications, the presence of an "xmlns" attribute is
always a conformance error in any content sent as text/html.
> 2. Is WhatWG entertaining the idea that off-the-cuff tag soup writers
> will generate MathML content that's good enough for Mozilla rendering?
The idea being entertained is that off-the-cuff HTML5 authors, and HTML5
editors, would create content which, when processed by an HTML5 UA (such
as Mozilla, in due course), would render as MathML markup would.
> The W3C Math group has announced that it is beginning to think seriously
> about author-level markup for math.
>
> Long term -- say ten years in the future (we've already been at this for
> ten years) -- I think author level math additions to the tag soup
> vocabulary would work out much better, especially with enhanced CSS
> support.
On the very short term, the proposal here is just a proof of concept. On
the medium term (12 months) I was considering specifying more complex
parsing rules for MathML such that the same MathML2-compatible DOM could
be obtained from much smaller markup, e.g. by implying <mo> tags around
operators and <mn> tags around numbers.
HTH,
Roger B. Sidje
> I don't remember specific discussions about an <mtr label="..."> I
> would guess there woul dbe some convern about the label being an
> attribute rather than an element restricting the possibilities, but
> implementation advice on difficulties on teh current schem woul dbe
> taken seriously....
Here is an informative thread about it:
http://groups.google.com/group/netscape.public.mozilla.mathml/browse_thread/thread/d77d015a1fffc6fb/5b0eb0cc9724ce72
(not on www-math, though. Maybe I should forward it there?)
It appeared that attributes (like those in <mfenced>) aren't unanimous
either. But having a bloated tag that won't be implemented in the next
several years isn't really helpful.
> Ian wrote about entities
>
>>Yeah... Do we really need those? Some of them seem reasonable to add, but
>>2000 seems like too many for the mnemonic advantage to beat just using
>>Unicode codepoints...
>
> I'd say that it's probably not worth including only a few, it would just
> lead to confusion.
I am actually a fan of entities because they improve readability a fair
bit. I hope Ian won't give up thinking on this issue so quickly...
especially in the context of MathML where strange characters are quite
common.
As to my suggestion that "if [a document] is strict then maybe entities
could be required to have a semi-colon -- which will then avoid the
ambiguities", to which Ian responded that, "That would break back-compat."
We have other cases of broken back-compat. -- where users were told to
use a non-strict DOCTYPE or some other workaround, e.g, line-height of
images.
---
RBS
David Carlisle
Roger,
Thanks for the link on <mtr label="mylabel">,
> It appeared that attributes (like those in <mfenced>) aren't unanimous
> either.
yes mfenced also "suffers" from requiring attributes, but probably one
is more likely to need markup in an equation label than in a stretchy
operator. It's not so uncommon to want superscript * or daggers etc to
highlight special versions of formulae, and mfenced is explictly a
shorthand form so you can always use the mwrow/mo form if you need an
operator that is "decorated" in some way. That would not be the case
here if mlabeledtr were deprecated and an attribute form was the
only version. (Actually it would if the attribute could then be
css-styled using css generated content. Allowing css (or other
mechanism) auto numbering is I think a highly requested feature for
mathml3.
> (not on www-math, though. Maybe I should forward it there?)
Yes please do. When we are doing a pass for errata or pulling in feature
requests for a new version we can do a more or less exhaustive check of
the official comment list but (even with google's help) doing an
exhaustive check of the entire web's a bit hard:-)
The charter for the current working group
http://www.w3.org/Math/Documents/Charter2006.html
has as one of its headline work items
Extension of MathML with enhanced support for equation labeling,
including automatic numbering, general label placement and style, and
resolution of references.
so getting that specified out in a way that ensures that implementations
can implement it sounds like a good idea, and the timiming is good now
to get new features in this area if that is needed. If WhatWG members
are interested in mathml most of them are w3c members and could join the
WG of course (currently only Opera is represented out of the main
browser vendors) But WG membership isn't really needed we can do the
technical discussion on the public www-math list if that is appropriate.
> I am actually a fan of entities because they improve readability a fair
> bit.
Well as you know I've invested a frightening number of houres maintaining
that entity set (and the draft iso set at www.w3.org/2003/entities,
which is the same thing, really) so I'm also think they are valuable,
although it's a kind of love-hate relationship most of the time:-)
> I hope Ian won't give up thinking on this issue so quickly...
> especially in the context of MathML where strange characters are quite
> common.
Yes I think the ideal situation is that they all be allowed. My comment
was that subsetting them is likely to be more confusing than helpful.
> As to my suggestion that "if [a document] is strict then maybe entities
> could be required to have a semi-colon -- which will then avoid the
> ambiguities", to which Ian responded that, "That would break back-compat."
Requiring a ; would seem reasonable to me (ie make the lack of a ; make
the & into an implict & rather than be an error as in xml).
That does have a theoretical backward compatibility problem in that
→ would be an arrow instead of &rightarrow; but I would
have thought that the occurrences of any such construction outside of
test suites was rather rare.
David
Ian Hickson
> >
> > Ian wrote about entities
> >
> > > Yeah... Do we really need those? Some of them seem reasonable to add, but
> > > 2000 seems like too many for the mnemonic advantage to beat just using
> > > Unicode codepoints...
> >
> > I'd say that it's probably not worth including only a few, it would just
> > lead to confusion.
>
> common.
I really don't want to start introducing weird rules for parsing entities
(I'm trying to simplify the entity parsing rules, not make them worse). At
least not at this stage. Maybe once we have a proof-of-concept working, it
would make more sense to revisit the issue, but I'd want to do a thorough
scan of the Web to see how common these entities actually are today.
> As to my suggestion that "if [a document] is strict then maybe entities
> could be required to have a semi-colon -- which will then avoid the
> ambiguities", to which Ian responded that, "That would break
> back-compat."
>
> We have other cases of broken back-compat. -- where users were told to
> use a non-strict DOCTYPE or some other workaround, e.g, line-height of
> images.
Yeah. And we can see how well _that_ went. QA nightmare, multiple
overlapping codepaths, obscure bugs, confused authors, contradicting
documentation, etc. Let's not go there again. The whole point of
MathML-in-HTML is to have back-compat work -- if we didn't care about
back-compat, we would just have people use MathML-in-XHTML.
Roger B. Sidje
> Roger,
> Thanks for the link on <mtr label="mylabel">,
>
>
>>It appeared that attributes (like those in <mfenced>) aren't unanimous
>>either.
>
>
> yes mfenced also "suffers" from requiring attributes, but probably one
> is more likely to need markup in an equation label than in a stretchy
> operator. It's not so uncommon to want superscript * or daggers etc to
> highlight special versions of formulae, and mfenced is explictly a
> shorthand form so you can always use the mwrow/mo form if you need an
> operator that is "decorated" in some way. That would not be the case
> here if mlabeledtr were deprecated and an attribute form was the
> only version. (Actually it would if the attribute could then be
> css-styled using css generated content. Allowing css (or other
> mechanism) auto numbering is I think a highly requested feature for
> mathml3.
The danger (and problem) with that tag is that it is over-designed to
accommodate the tiny set of special-cases you alluded to, while holding
the 99.99% majority of cases hostage. One could put up with CDATA all
the way, e.g., (6') or (7*), (8†), (9a), etc -- if a subequation
is really needed. I would think we can put with this and reap the
benefits. A <mtr label="mylabel"> tag that stands a chance, degrades
gracefully, *free* cross-referencing (with href#mylabel -- by just
invoking what the browser already does with <a name="...">), the
counters that you mentioned (which work in Gecko today, BTW), etc.
(Also conceivable, optimistically, is a pseudo-class :label to style the
label text, but we might going ahead of ourselves...)
Seems to me that the concrete benefits that might result outweigh the
feeling against an attribute.
>
>>(not on www-math, though. Maybe I should forward it there?)
>
> Yes please do.
OK.
> Well as you know I've invested a frightening number of houres maintaining
> that entity set (and the draft iso set at www.w3.org/2003/entities,
> which is the same thing, really) so I'm also think they are valuable,
> although it's a kind of love-hate relationship most of the time:-)
Yeah. Let's hope Ian is listening and keeps these entities on his radar...
---
RBS
David Carlisle
> Seems to me that the concrete benefits that might result outweigh the
> feeling against an attribute.
Which is why it's good to get real implementation experience into the
language design (or update). Either by implementors joining the WG or
by doing the technical design on the public www-math list so you and
others can join in (or both).
David
White Lynx
to move in this way.
My main concerns are twofolds, first is the fact that converting
currently XML based MathML to text/html tagsoup is not the step in
right direction as MathML community has no text/html legacy content and
there are simply no MathML implementations that support text/html, but
does not support X(HT)ML, while there definitely are XML only MathML
implementations and different XML tools are used by MathML community
and the fact that all MathML content is wellformed be design is
something that did not come for granted, people had to pay price for
this for seven years until finally MSIE/MathPlayer started to recognize
application/xhtml+xml and thus allowed people to deliver the same
XHTML+MathML to MSIE/MathPlayer and Mozilla. So today is the most
inappropriate time to make experiments that may convert XML math
content to tag soup.
Second concern is related to MathML/CSS integration issues, I think I
should not be explaining that for mathematical markup language it is
important to be integrated with the environment in which math formulae
are embedded. In case of web this environment is XML/CSS/DOM (well
there are people that think that universe = text/html, but math markup
simply MUST be highly extesible as requirements of math community are
diverse and it is not uncommon to combine presentational MathML with
other XML applications, including OpenMath, content MathML, sometimes
SVG and other XML applications).
Currently there is subgroup in W3C Math WG chartered to address
MathML/CSS integration issues so one could use appropriately designed
profile of MathML in XML/CSS framework, that is we are moving towards
markup language that admits default CSS style sheet. This process
requires coordination between Math WG (some changes are required on
MathML side), CSS WG (at least some CSS extensions are needed to ensure
minimal requiremnts of math markup are addressed) and browser
developers that are most exposed to issues that arise from lack of
MathML/CSS integration. Opera is involved in process in hope that in
long term perspective MathML will work smoothly in XML/CSS environment
and many artificial incompatibility issues and unnecessary doubling of
functionality at the price of tons of bugs and underspecified issues
that emerge in componud documents due to incompatibility between
different specs, Microsoft intends to join Math WG that hopefully will
help to coordinate efforts between math markup related activities in
ECMA backuped by Microsoft and at least partly avoid problems with
double standards W3C vs. ECMA in the area, I am sure we would be glad
to see Mozilla onboard either by joining WG or being involved in the
process otherwise in any convenient for you form, as well as developers
of other layout engies including KHTML/WebKit and Prince. However I am
seriously concerned about spontaneous changes that may take MathML
further from XML+CSS framework and I am surprised that initiative for
such a changes is coming from Mozilla foundation. Today is the most
inapproprite time for turning things upside down, but definetely is a
good time to improve coordination between different working groups and
browser developers to resolve actual problems.
Note also that Math WG has subgroups like liason with WhatWG and liason
with CDF chartered to impove coordination between different units and
ensure that MathML works smoothly in compound documents. So there is a
lot of space for moving the process through this channels instead of
making unilateral steps that may damage MathML in long term perspective.