loge.hixie.ch

Hixie's Natural Log

2004-08-04 13:40 UTC Why document.write() doesn't work in XML

Several people recently asked me why their scripts, which worked fine when they were using HTML, stopped working when they started using XHTML (and actually sent their markup with an XHTML MIME type).

There are plenty of reasons why that might happen. Not using the namespaced DOM Core methods is one, but few people use the DOM Core methods so it's rarely the cause. It could also be that their markup wasn't well-formed, but usually that's pretty obvious since the browser complains about it, and when people ask me, it's usually because the cause is subtle, so this is rarely it either.

It usually turns out that the reason their code is failing is that they used document.write(), which doesn't work in XML. Why is that?

XML requires that an XML process abort on any well-formedness error. Say you have the following:

<foo>
 <scrible/>
 </bar>
</foo>

Since the </bar> tag is unmatched, the entire document is not well-formed, so the UA has to abort and not display it.

But if we allow document.write(), then you could have things like this:

<foo>
 <script type="text/javascript" xmlns="http://www.w3.org/xhtml/1999"/><[!CDATA[
  document.write('<bar>');
 ]]></script>
 </bar>
</foo>

Now the document is still not well-formed, so it must still abort when it gets to the </bar> tag. But the problem is that if we process that document, then when we get to the script, we write out a <bar> start tag after the <script> block, and now when we reach the </bar> tag, it matches something! So the UA couldn't tell if it was well-formed or not.

So in order to remain compliant with XML, document.write() doesn't work on documents that are parsed with an XML parser.

Pingbacks: 1 2 3 4 5 6 7

2004-07-12 12:24 UTC Extending HTML

So Apple decided to make up some new tags.

To be precise, they invented a new element, a new attribute, and a new value. The new element is canvas, an element that does nothing except provide a DOM interface for graphics drawing; the new attribute is composite, a presentational attribute that can be applied to images; and the new value is search on the type attribute of input elements, which provides a control for incremental search interfaces.

These new tags are to be used in several places. One of these is Dashboard, where applications written in an HTML-like interpreted language are executed in a runtime environment similar to a Web browser's. Another place where these tags are to be used, I am told, is the real Web.

The response to this announcement from a number of people in the industry seemed to be outrage... about the syntax.

Hello?

Here is Apple introducing their own proprietary markup to the Web, without going through any sort of standardisation first (not even unofficial standardisation like the WHATWG), and what people complain about is the syntax?

For instance, Eric Meyer (great guy) says:

I hope that what Dave is really saying is that Dashboard widgets are actually XML, albeit an XML that looks very much like HTML except they've added some nifty stuff to it. If so, great, fine, no problem. XML lets you do whatever you want, really.

So let me get this right. It's ok to send proprietary non-standard markup over the Web, so long as the angle brackets are XML angle brackets and not HTML angle brackets?

This makes no sense. Proprietary markup is proprietary markup, whether it is HTML-based, XML-based, or any other language such as PDF, Microsoft Word, XAML, or Flash. It's not the exact order of the angle brackets that matters, it's the lack of open, consensus-driven specifications, the lack of interoperability.

Eric wasn't the only person arguing that Safari's extensions are abominable because they can be used as if they were HTML elements, I just picked on him because I know he won't take it personally!

There have been several "solutions" proposed to the "problem" of proprietary tags being added to HTML.

The first is using namespaces in HTML. There is one fundamental problem with that idea. HTML doesn't have namespaces. At all. HTML doesn't allow namespaces. The Namespaces-in-XML spec doesn't apply to HTML. However you cut it, there simply isn't a way to allow namespaces in HTML. If you were to add namespaces to HTML, all you would be doing is adding yet another non-ratified extension on top of the original extension.

The second proposed solution is to use namespaces in XHTML, sending it as text/html. This also can't be done, this time because while you can use namespaces in XHTML (with some caveats; see below), you can't send such XHTML as text/html. The only things you can send as text/html (according to the relevant RFC) are HTML documents (up to version 4.01), and XHTML1 documents complying to the compatibility profile. (Although I would strongly recommend against doing that anyway.)

The third proposed solution is to use namespaces in XHTML, sending it using a legitimate XML MIME type. This could work, but it is limited to XML, and Apple wanted to allow authors to use HTML. Also, it requires that authors use two namespaces (one for the XHTML content, one for the Apple-proprietary content), and exposing authors to namespaces is a non-starter. The entire point of the Dashboard development environment is that it be trivially easy to write these applications, and namespaces are not trivial.

The fourth proposed solution is to simply replace the XHTML namespace with an Apple namespace, so there is only one namespace. This is even worse. Now not only would Apple be introducing three new non-standard tags, but they'd be introducing the whole of HTML as non-standard tags. No other browser would be able to render such pages. Accessibility tools would be unable to recognise the elements. Tools like Google would be unable to understand the semantics of the page for proper indexing.

The fifth proposed solution is to use a DOCTYPE that would enable the extensions. This doesn't scale. What if another UA wants to implement the extension as well? For example, marquee is required by half the Chinese Web. Originally, only IE supported it. Now take canvas, the Safari extension. If we had used different DOCTYPEs for each set of extensions, which DOCTYPE would you use if you wanted both? (And wanting both is quite likely; imagine a stock ticker with a graph above it).

There are other problems with DOCTYPEs. Eric unintentionally reminded me of one when he said:

IBM already does something like this, having created an "IBM XHTML 1.1 Transitional" DTD that's used throughout their Web presence.

The IBM DOCTYPE actually had to be hard-coded into Mozilla to trigger quirks mode, otherwise their site wouldn't render right. It's also non-standard, and violates the HTML spec in several ways.

Eric continues:

If you run an IBM-XHTML compliant document through a validator that loads and uses IBM’s DTD to check over the document’s markup, then the document will validate.

This rather misses the point. It still won't be conformant. This is a key point. Creating a new DTD does not magically make you standards compliant. For example, here is a document that "validates":

<!DOCTYPE LI PUBLIC "-//W3C//DTD HTML 4.01//EN">
<li>Validation is only a small part of conformance.

I'm sure we all agree it's not conformant, though.

I've probably missed some of the other proposals. But the reality is that it doesn't matter. If you want to make a proprietary non-standard extension to a standard that doesn't support explicit namespacing, you do the same as is recommended for CSS: put your vendor name into the tag. In the case of the aforementioned Safari extensions, I would have recommended calling the new names apple-canvas, apple-composite, and apple-search. The real solution is to bring these proposals to the table, get some consensus between the relevant vendors and other interested parties, and then use that. For specific features like these, it doesn't take long to get consensus; they are small features whose basic design can be agreed reasonably quickly. Doing this also usually means getting wider peer review, which improves the quality of the API.

For instance, it would have been pointed out that composite should never have been put into the markup. It's presentational. It belongs in the presentation layer, namely CSS.

Apple were probably concerned that by talking about this stuff, they would have tipped their hand about Dashboard. WHATWG is the perfect cover, though. Apple could have put forward their proposals, discussed them, got concensus, and everyone would have thought they were just planning these features for the Web Application space.

In fact, they did do this. Apple also introduced the new range value for the type attribute of input elements, which introduces a new control for imprecise numeric data entry (e.g. a slider). This, as I mentioned before, is not a proprietary value. This value was discussed, in detail, on the WHATWG open mailing list, and forms part of the Web Forms 2.0 draft specification. Here, Apple did the right thing: they created an experimental implementation of a proposal. Their implementation experience has already helped the draft (several changes to the original text were made to more closely align with what Apple found was actually needed).

So why did they not do this with the other extensions?

In other news, Brendan Eich gives us his take on the WHAT Working Group work and other issues facing the Mozilla Foundation in the latest edition of the Gillmor Gang. I highly recommend giving it a listen.

Pingbacks: 1 2 3 4 5 6 7 8 9

2004-06-29 16:26 UTC State of the WHAT

At the start of the month, Opera and Mozilla announced the Web Hypertext Application Technology Working Group. There followed some 600 messages and lots of great ideas, resulting, last Sunday, in our first "call for comments" for the Web Forms 2.0 specification.

When we announced WHATWG, several people noticed that the list of members also contained, in addition to Mozilla and Opera employees, a couple of Apple employees who work on Safari. A few weeks into the project we also added Dean Edwards of IE7 fame to the list. The members of the WHATWG are an oversight committee intended to make sure we don't get off-track; the specs are actually edited by me and based directly on the input we get on the mailing list.

There were also some people who commented that instead of calling it a Working Group we should have called it a Task Force, because the resulting acronym would have been much more appropriate. All I can say is that I wish I had thought of that because that would have been a really funny name.

I recently started on some Web Forms demos, to show what Web Forms can do and how it can do it without losing backwards compatibility. I haven't done much yet, just a demo of how to make the repetition model degrade gracefully in HTML4-compliant UAs, but it's a start. (Sadly the only UA I could find that was compliant enough to degrade in the right way was Mozilla, so we might well have to revisit exactly how those repetition model buttons work.)

The first public demo of a Web Forms 2.0 page was actually a very high profile demo, though I highly doubt that anyone present actually realised they were looking at it. It was during Steve Jobs' keynote speech at Apple's Worldwide Developers Conference, during the Safari RSS spot. If you look at the top right of the screenshots of Safari showing an RSS feed, you'll see a little slider control. That, my friends, is a Web Forms 2.0 <input type="range"> control.

Something I hadn't realised until recently is quite how many Web applications are hidden away inside intranet sites. I always knew that there were some there, but the sheer numbers of such applications is quite surprising. A few people have sent me confidential screenshots of their intranet applications (with the sensitive parts censored, of course), which has really helped get me an idea of the kinds of features that would be most helpful to people writing such sites. We'll be addressing a number of these requests in Web Apps 1.0 proposals once I've dealt with all the recent Web Forms 2.0 comments.

Joel Spolksy wrote an interesting article explaining why Microsoft stopped Internet Explorer development in its tracks a few years ago, despite it having ridiculously many bugs. Here's the executive summary: they realised that IE was competing with their OS to be the preferred application deployment platform. Since they make money from their OS, but don't make money from IE, the choice was clear.

I think the irony of this situation is quite amusing. HTML, a document format, is invented, eventually Netscape is formed and creates a dramatically successful product that makes Microsoft jump. Microsoft instinctively react to the creation of this new networked document viewer market by developing a product to compete with it. They (inevitably) succeed, by the simple mechanism of creating incrementally better products until one is "good enough", aided by the chronic mismanagement that characterised Netscape during its last eight or so years. However, in the middle of this process, the fight stops being over who can make the best document viewer, and slips into who can make the best dynamic content viewer. And that slowly turns into an application platform. Oops. I'm no business expert but I would hazard a guess that plunging billions of dollars into a loss-leader that competes with your primary business model is not the best idea.

I've known all the bits of this story for years, but they never really clicked until Joel explained it.

As Joel points out, though, Microsoft's moves after they realised their mistake with IE have been harder to understand. I would have thought the solution most likely to succeed would have been to extend IE in ways that made it into a better application deployment platform. Eventually, this could have turned IE into the OS, either natively (making the next version of Windows basically be IE), or by selling IE with versions for all operating systems. This would have had several advantages:

  1. It would settle once and for all their claim that IE needs to be part of the OS.
  2. It would solve the Linux threat: instead of competing with Linux, Linux would become just another platform upon which IE can run.
  3. It is what their customers wanted.
  4. It could create lock-in just like Windows has.

They actually did start down that road. IE6 has support for a technology that Microsoft stopped advertising at the same time as they stopped developing IE, namely HTAs, short for HTML Applications. Similarly, even without really trying, Microsoft managed to lock large numbers of people into the IE platform — it has taken years for Opera, Mozilla, et al to catch up to IE, and even now, there are many sites that simply won't work in any other UA. The problem, I am told, is that the visionaries at Microsoft want the lock-in platform to be an OS. That makes sense; I mean, that's what they know.

And so they took all the things that they thought made HTML a good deployment platform, and all the things that they thought made Win32 a good development platform, and all the ideas that Microsoft Research said would be cool, and started a huge project: making a completely new OS. (It isn't clear to me if this project spawned sub-projects like .NET, or if smaller projects like .NET became subsumed in the larger OS project.)

There are several problems with this approach, as I see it. First, the market is no longer growing that fast. In the past, when they released a new OS, Microsoft could grow its marketshare primarily by growing the market, and then once the new market hit critical mass, people on older platforms that couldn't run the newer software would be forced to upgrade to run the newer software. (They did the same with Microsoft Office: by selling enough new copies to new users, eventually old users were forced to upgrade simply to be able to read documents written by new users.)

But in the current market, Microsoft are having to depend mainly on upgrades. Windows XP didn't do as well as previous releases, as far as market penetration goes. And many people are questioning whether Longhorn has enough new features to warrant upgrading to it. So Microsoft will either have to quite radically change their approach, or they will have to settle into a slower revenue cycle.

But IE is only to play a supportive role in this big project.

Of course, Microsoft aren't alone in thinking that extending IE to be a new platform is a bad plan. As I previously discussed, the W3C held a conference on Web Applications just last month, and the majority of people there disagreed with the concept of the browser being the platform as well.

This actually led several key figures in the industry to question the W3C's legitimacy. For example, David Baron ponders whether the W3C is really relevant any more. Brendan Eich also wrote some scathing commentary on the W3C.

Then again, I work for a browser vendor, and have been in the browser industry (both voluntarily and as a job) for years now. So it's not surprise that I think the browser is important as an application platform. (Obviously, though, as Robert Scoble is eager to tell us, Web apps aren't the answer to everything. I wouldn't recommend to anyone that they try to write a graphics manipulation package in HTML.)

The problem with the browser today is that applications based in the browser are constrained to nightmarish UI idioms and a severe lack of polish stemming from the fact that the platform was not really developed as a platform, and that no real progress has been made on this path for several years.

John Gruber points out that users don't really seem to care about the poor UI, though. The other advantages — especially the true zero-install cost of Web-based applications — far outweigh the costs.

But that's why we started WHATWG: we want to make it easier to make nicer the kinds of applications that it makes sense to deploy over the Web. Mail and news clients. Cinema ticket sales. Book stores. Auction sites. Multiplayer stategy games.

In other news, apparently the Windows IE team is starting to wind down its work on the "let's make IE secure" Windows XP service pack 2 release, and is starting to look at what people want in the IE that will ship with Longhorn. They are even looking for feedback. Amusingly one of the most popular requests in the forums I have looked at (which is probably a very skewed subset of the real feedback that Microsoft gets) is "better standards compliance". And even more amusingly, one of the most often repeated messages from Microsoft people in those same forums is "we're not going to be working on standards compliance" (although not usually phrased that way).

I honestly do not expect to see any changes at all to the HTML, CSS, DOM, and JS components in the Longhorn version of IE compared to Windows IE6. I've had Microsoft employees tell me to my face that they will not have XHTML or SVG support in the next release. I've only seen excuses for why features won't be added, never any Microsoft employees saying they are interested in supporting a new spec. I have thousands of tests available on my site if they want to find some bugs, but I doubt they will even look.

On the other hand, I can assure you that here at Opera we are implementing all kinds of things and every release has more CSS/HTML/DOM/JS/XML/etc fixes. And over in Mozilla land Brendan has written several posts listing the kinds of things Mozilla will be implementing over the next few years.

Pingbacks: 1 2 3 4 5 6

2004-06-04 22:20 UTC Spring 2004 Travelog: Part 9 (Return to Europe)

Travelling east is painful. I woke up around 05:40 on Thursday, and went to bed around 15:00 on Friday, nine time zones away. My brain now thinks that 23:00 is a good waking time. Also, it meant I had three breakfasts yesterday, which successfully smashed my fast so that I wasn't hungry even though I should have been. Quite the biological clock confusion.

The second day of the workshop had more discussion. We had some straw polls; of the 40 or so people there, around 8 said they wanted to work on the work Opera and Mozilla have been proposing recently, and about 11 said that not only did they not think it would be worth working on this, but they actively thought that the W3C should not work on it.

In my opinion that's pretty short-sighted, but as Steven Pemberton pointed out, six year ago, the W3C decided that HTML was dead, and the way forward was a host of new languages (what is now XHTML2, XForms, MathML, SVG) that would lead the world's population to a clean new world. So at least they are consistent.

Of course I had to point out that six years ago, I was in school, which got a good laugh. My point, though, was that times change. In the last six years we have seen that authors simply didn't agree: Mozilla has supported MathML for years, but it is still very rare to see any MathML content on the Web. Mozilla, Opera and Safari all support XHTML1, in fact Mozilla has supported XHTML1 since before it had an assigned namespace and MIME type, but again the amount of application/xhtml+xml content on the Web is trivial.

The truth is that the real Web, the Web that authors write for, is the Windows IE6 Web. The only way to change that is to reduce the IE6 market share, and new technologies don't do this. Marketing does. Once users are primarily using a browser that is being regularly updated, then we can start introducing radically new technologies. Until then, such technologies simply aren't going to become popular.

There were a lot of rather confused statements during the meeting. For example, it is clear that a lot of people think that the browser is dead and that the way forward is transparent "runtimes" that execute remote applications securely. But then these same people demand to know why Mozilla, Opera and Safari don't support XForms and SVG, saying that their lack of support is crippling their standards' adoption.

Surely if the browser paradigm is dead, it doesn't matter what we implement?

What I think most of the people at the meeting actually want is a standard that combines XHTML, XForms, SVG, and SMIL (and CSS, DOM, and ECMAScript, although they rarely if ever actually mention those by name), and then adds enough APIs to make the host into a platform in its own right.

Java tried the "provide lots of APIs that are interoperable across lots of platforms" and failed. Some people thought this was because Java, as a language, is too complex for most applications. And Java doesn't have a detailed spec (not detailed enough to write an interoperable implementation accurate to the level that is needed for applications).

The detailed spec problem is the big issue. There has simply never been a Web specification written in enough detail for this kind of work. Even "DHTML", which does just a fraction of the number of APIs needed for the kinds of applications these people are imagining, is completely inadequately specified. For example, if you have an object element followed by a script block, will the script execute before or after the object has loaded? This is the kind of behaviour that scripts depend on. (Answer: In IE, the script will block waiting for the object, and if the object doesn't load, it will be removed from the DOM. The exact behaviour depends on the extension of the filename in the data attribute and the local computer's registry. Feel free to explore this yourself using these testcases...)

Making these specs more detailed is the work that Opera and Mozilla want to do. But to do this for a sophisticated application platform on par with, say, Longhorn, is simply unfeasible. Notice how WINE has to reverse engineer Windows to determine how it should work. Or how the various Java clones have to reverse engineer Sun's Java to get interoperability.

Of course, if they want to do this, I wish them the best of luck. I might even want to participate in the working group, since someone will have to look out for the Opera and Mozilla interests!

Sadly there does seem to be a growing opinion in certain circles that the W3C is becoming more and more out of touch with the Web. In many ways, this makes sense: the membership has many more server-side people than client-side people, and most of the client-side people are plug-in vendors, not browser vendors. (All the browser vendors present at the meeting were in favour of variants on the Opera/Mozilla ideas, but they were easily out-numbered by the non-browser members.) Since most people consider "the Web" to be what browsers show, it's only natural for an organisation of people who are largely not doing Web browser work to appear to be "out of touch".

Really it's not that the W3C is out of touch with the Web, but that the W3C membership is solving problems that every day Web users don't see. For example things like CC/PP and SOAP are very much back-end technologies.

I've also heard a lot of comments recently from people asking if the SVG working group realises that SVG 1.2 is becoming a dumping ground for anything and everything, instead of remaining just a graphics language. For example, SVG 1.2 drafts feature raw socket APIs and a Window interface. So I asked people at the conference whether the SVG working group shouldn't, instead of adding every feature under the sun to SVG, simply define how SVG should interact with other languages like XHTML. The answer I got was highly dismissive. Basically: "Well we want it in SVG".

You'll note that Robert O'Callahan (one of the core layout developers for Mozilla) sent an e-mail last month pointing out the many areas of overlap between SVG and CSS. He never got a reply to his last message. (Hmm... this is reminiscent of the way the SVG working group effectively ignored a message I sent back in 1999 pointing out a problem in SVG 1.0 that is still present in SVG 1.2 drafts — reply once, ignore further e-mails on the thread, and don't actually address any of the problems...)

Since many of SVG's features (like Window, or like the way CSS properties are allowed to have lengths with no units) directly conflict with existing standards (both ad hoc and de jure), it's quite clear that browsers that implement SVG will only ever be able to implement a subset of SVG.

Another point that was mentioned at the workshop was that DOM3 core was a long specification, checking in at some 216 pages. The SVG 1.0 spec is 719 pages long. (More than twice the length of CSS2.1, which I thought was already a ridiculously long specification.)

Still, it's a nice language if all you want to do is draw vector graphics.

Anyway. The issues have been discussed, the positions have been given, everyone knows where everyone else stands, now it's time to get down and actually start doing work.

What working group is going to work on extending HTML...

Pingbacks: 1 2 3

2004-06-02 06:48 UTC Spring 2004 Travelog: Part 8 (First Day of the Workshop)

Today I listened to a good dozen presentations from various groups on the subject of Web Applications and Compound Documents.

Some interesting things came out. First, the only sustained spontaneous clapping of the entire day came as someone suggested, in response to my brief statement of how backwards compatibility is critical, that it was about time to drop HTML and Windows IE6 from the roadmap.

So I can assume from that that most people don't agree with the whole backwards-compatibility thing!

Second: I was quite amused to see that, of all companies, Microsoft, Red Hat, and Sun Microsystems actually agreed on something. Namely that trying to standardise an API for sophisticated applications is simply a non-starter. The argument, which I agree with, is that such APIs are simply insanely complicated, and that making interoperable implementations is nigh on impossible. Just look at the trouble WINE has had trying to implement Win32 again — now imagine if you had to write a spec to actually describe the entire Win32 API in terms that could actually be implemented interoperably without reverse engineering the first implementation as the WINE people do.

What was funny was watching the other people then disagree with them. Hint: If three of the most bitter rivals in the marketplace — all of whom have extensive experience in the subject in question — agree on something, then it is probably true.

What we (Opera and Mozilla) want to do is simply extend HTML, DOM, and CSS a bit so that the most common things are easier to do. Things like my Web Forms 2 proposal or my server sent events proposal. These are simple extensions, not an attempt to provide comprehensive platform APIs.

Another point that came out of the discussions is that, in case there was any doubt, Internet Explorer in Longhorn will not support XHTML or SVG. (Microsoft suggested they would need some significantly more comprehensive test suites before they started working on standards compliance again.)

After the meeting, a bunch of us had dinner at La Fiesta. Our table had three Microsoft employees, two Red Hat employees, two Mozilla Foundation employees, and an Opera Software employee. I bet you won't see that very often.

Tomorrow we have another dozen presentations. I expect to see more of the same; mostly people expounding on the virtues of XForms, SVG, XHTML2, or their own radically new proprietary technologies, and explaining how Web Applications would all be much better off if the W3C would go down their chosen route.

We, of course, want the W3C to go down our chosen route. Since there doesn't seem to be much consensus on doing that, though, the question is what should we do now? Should we do our own thing (in public of course) and then submit it to the W3C (or IETF or ECMA) at some future point once we have initial implementations? Should we simply do our own thing (Opera, Mozilla, and a few interested parties) and forget standardisation altogether? Should we just take part in whatever Web Applications working group the W3C sets up and implement whatever comes out of that in several years' time, despite being fully aware that few people will ever use it? (Which is a foregone conclusion since it wouldn't work in Windows IE6.)

I'm learning towards the first of the three at the moment. I guess the Opera and Mozilla people will have to discuss this in more detail before we decide anything though.