Internet Explorer – Eugene Eric Kim

At last Tuesday’s HyperScope meeting, Jonathan Cheyer and I spent an inordinate amount of time debating the syntax for HyperScope URIs, much to the amusement and chagrin of our peers. Although the topic may seem insignificant, it is actually quite layered with no easy answers. I’m going to summarize the issues here. Most of you will probably not care about the intricacies of the argument itself, but at minimum, it should reveal a bit more about the project itself. (KI2)

HyperScope is meant to be a transitional tool that will enable people to play with Augment‘s more sophisticated capabilities within their existing environments. In the immediate future, that means the Firefox web browser (and probably Internet Explorer as well). In the not-to-distant future, that could extend to a range of applications, from Eclipse to OpenOffice and beyond. (KI3)

This intent strongly informs our requirements. In particular, we need to make sure we are bootstrapping the system on top of existing technologies (such as URIs) effectively and correctly, and we need to make sure the system is evolvable. Both of these requirements play a big role in our debate. (KI4)

So what’s all the fuss about? One of Augment‘s coolest (and most fundamental) features is its sophisticated addressability. The example most folks know about manifests itself as Purple Numbers, namely the ability to reference a specific node in a document in a standard way. But Augment can do much, much more. It can do path expressions, similar in spirit to XPath, which allows you to reference some subset of nodes in a document. (See my notes on transcluding a subset of nodes via Purple Numbers). (KI5)

You can also embed View Specs in an address. For example, suppose you decided that the best way to view a page was the first level of nodes only. You could specify that as a View Spec in the link itself, so that when someone followed that link, they would see only the first level of nodes rather than the entire document. (KI6)

With the HyperScope, we’re bringing these capabilities to the plain ol’ World Wide Web — that is, assuming your client knows how to interpret these addresses properly. With our initial release (due September 2006), this will require loading a JavaScript library. All document addressability will happen entirely on the client-side. This is a good thing for a lot of reasons, the most important being adoptability. It will be easy for people to play with the HyperScope. All they’ll have to do is click on a link in Firefox (and probably Internet Explorer). (KI7)

However, the fact that we’re doing a client-side only version of the HyperScope does not preclude the creation of a joint client/server version or even a mostly server-side version where the client is essentially a dumb web browser. In fact, we’d encourage the creation of both. We don’t care some much about implementation as we do capabilities and interoperability. (KI8)

Here’s the question: How should we include these extended addressing capabilities in real-life URIs? (KI9)

There are three possible solutions: (KIA)

Embed them as a fragment address (i.e. following a hash mark at the end of the URI). (KIB)
Embed them either as part of the path or query string parameters (i.e. following a question mark at the end of the URI). (KIC)
All of the above and more. (KID)

I side with the first and third solutions. Jonathan thinks it should be the second. (KIE)

Fragment Address (KIF)

Pros: (KIG)

These extended capabilities seem to belong here semantically. Purple Numbers are an obvious example of this. XPointer is another. (KIH)
The URIs will be bookmarkable as the user manipulates the document from the HyperScope. We can do this, because we can change the fragment identifier from within JavaScript. We can’t do the same with any other part of the URI. (KII)

Cons: (KIJ)

These URIs cannot be used for a server-side only solution, because HTTP does not pass the fragment identifier to web servers. (KIK)

Path or Query String (KIL)

Pros: (KIM)

Can be used for client-only or server-only solutions, or anywhere in between. (KIN)

Cons: (KIO)

You’re still left with the problem of a standard addressing syntax that doesn’t interfere with any other kind of addressing. For example, if you’re going to use a query parameter, what do you call it? Granted, if you namespace it correctly, the likelihood of namespace clash is tiny, but it’s still there. (KIP)
No one’s building a server-side only solution right now. (KIQ)

Why Limit Yourself? (KIR)

This is ultimately my argument: Go with the first syntax for now, because it best suits our current needs, and don’t worry about the fact that it won’t satisfy all potential future needs, because nothing will. What’s important is that we standardize the conceptual semantics, and then standardize the syntax to the extent possible. In all likelihood, most people will be passing these links around by copying and pasting them anyway, so the actual link syntax isn’t completely critical. (KIS)

For a while, it was looking like I was going to break another personal blogging record last month, then things got so busy I had zero time to blog whatsoever. That means I’m in catch up mode again, so as usual, I’ll post in reverse chronological order (which in the blogosphere is really reverse reverse chronological order). (JYN)

Yesterday, I spent the afternoon at Socialtext, where they were having an all-hands meeting. They graciously invited me to participate in the Open Space segment of their gathering, which meant quality time with Chris Dent and a rare opportunity to evangelize the Church Of Purple together. Of course, Chris has been spreading the Purple religion at Socialtext for a while now, so it wasn’t as much about evangelism as it was about next steps. (JYO)

As I’ve mentioned many times before, browser-based WYSIWYG editors are an exciting development because they allow us to make Purple Numbers transparent in the authoring process. Right now, when you edit a PurpleWiki page, you see the node ID tags (e.g. {nid 123}). This is impossible to get around with the default browser text-editing widget. However, with a WYSIWYG editor, you can hide the Purple Numbers while still maintaining their associations with a node behind the scenes. (JYP)

That’s the theory, anyway. In particular, I’ve been excited about WIKIWYG ever since Ross Mayfield showed me an early prototype last August. I had a personal bias, since Chris Dent and Matt Liggett helped write it, as did Casey West and the inimitable Brian Ingerson, whom I finally met last weekend at Tag Camp. (JYQ)

Yesterday afternoon, we discussed Purple Numbers and WIKIWYG, and it was good. Then in the evening, Ingy and I spent a few hours trying to get WIKIWYG integrated into PurpleWiki. (JYR)

We didn’t quite make it. Our biggest roadblock was a bug we discovered in Mozilla’s design mode that we can’t do much about. (My days of statically typed languages are well behind me.) But, we got something somewhat working, and I learned a heckuvalot. You can play with our semi-working demo. (JYS)

WIKIWYG seems well-architected and is easy to customize. For folks with relatively standard Wiki editing requirements, I highly encourage you to play with it. PurpleWiki has some special formatting funkiness (mainly due to the Purple Numbers), but we were able to get around this fairly easily. (This was also true thanks to PurpleWiki‘s model of parsing to an intermediate data structure, then using view drivers to serialize. I wish more Wiki engines did this. I know Magnus Manske is thinking about doing this for Mediawiki, and I think Janne Jalkanen is already doing it with JSPWiki.) (JYT)

The Mozilla bug annoyed me, because it’s a show-stopper in some ways, and there’s not much I can do about it. I didn’t realize it, but all of the JavaScript WYSIWYG widgets actually switch to the browser’s “design mode” in order to handle WYSIWYG editing. As with many HTML editors, design mode does not handle structure cleanly, and you end up getting weird artifacts such as spurious break tags. Our problem was that we serialize node ID information as id attributes in the HTML tags. However, Firefox does not maintain those attributes correctly when you move content around. (JYU)

I’ll report the bug (if folks have suggestions as to the best way to bring this to the right people’s attention, let me know), but it also puts the kibosh on my hopes for WIKIWYG and Purple Numbers. Even if the bug is fixed in the next version of Firefox, we’re still prey to all the folks using older versions as well as Internet Explorer or Safari, which have their own problems with design mode. (JYV)

Chris and I discussed one workaround that I’m still pondering: render the Purple Number and have users be responsible for maintaining the association with the nodes. That’s the status quo, except users are doing it in WikiText rather than in WYSIWYG. Doing it in WYSIWYG certainly lowers the bar, and it’s probably the next best thing for us to do. (JYW)