protocol page 2 - Dave's Blog

Thought Experiments and Design Principles

Raymond Chen has some thought experiments useful for discovering various kinds of stupidity in software design:

• What if two programs did this?
• Who says there's only one?
• Imagine if this were possible. / Imagine if things actually worked that way.

Tim Berners-Lee's principles of Web design includes my favorite: Test of Independent Invention. This has a thought experiment containing the construction of the MMM (Multi-Media Mesh) with MRIs (Media Resource Identifiers) and MMTP (Muli-Media Transport Protocol).

The Internet design principles (RFC 1958) includes the Robustness Principle: be strict when sending and tolerant when receiving. A good one, but applied too liberally can lead to interop issues. For instance, consider web browsers. Imagine one browser becomes so popular that web devs create web pages and just test out their pages in this popular browser. They don't ensure their pages conform to standards and accidentally end up depending on the manner in which this popular browser tolerantly accepts non-standard input. This non-standard behavior ends up as de facto standard and future updates to the standard essentially has had decisions made for it.

WPAD Server Fiddler Extension

I've made a WPAD server Fiddler extension and in a fit of creativity I've named it: WPAD Server Fiddler Extension.

Of course you know about Fiddler, Eric's awesome HTTP debugger tool, the HTTP proxy that lets you inspect, visualize and modify the HTTP traffic that flows through it. And on the subject you've probably definitely heard of WPAD, the Web Proxy Auto Discovery protocol that allows web browsers like IE to use DHCP or DNS to automatically discover HTTP proxies on their network. While working on a particularly nasty WPAD bug towards the end of IE8 I really wished I had a way to see the WPAD requests and responses and modify PAC responses in Fiddler. Well the wishes of me of the past are now fulfilled by present day me as this Fiddler extension will respond to WPAD DHCP requests telling those clients (by default) that Fiddler is their proxy.

When I started working on this project I didn't really understand how DHCP worked especially with respect to WPAD. I won't bore you with my misconceptions: it works by having your one DHCP server on your network respond to regular DHCP requests as well as WPAD DHCP requests. And Windows I've found runs a DHCP client service (you can start/stop it via Start|Run|'services.msc', scroll to DHCP Client or via the command line with "net start/stop 'DHCP Client'") that caches DHCP server responses making it just slightly more difficult to test and debug my extension. If a Windows app uses the DHCP client APIs to ask for the WPAD option, this service will send out a DHCP request and take the first DHCP server response it gets. That means that if you're on a network with a DHCP server, my extension will be racing to respond to the client. If the DHCP server wins then the client ignores the WPAD response from my extension.

Various documents and tools I found useful while working on this:

• WPAD IETF draft document the never out of draft document describing WPAD
• DHCP RFC
• DHCP options RFC
• DHCP Wikipedia article which includes helpful byte layout tables for DHCP messages.
• Fiddler Wiki esp. the dev section but also the portion on the new preferences API.
• Fiddler Dev Docs and for those things not covered in the Fiddler docs, Visual Studio's name completion (or the .NET Reflector) usually served me well.
• Netmon a great network sniffing tool for all my non-HTTP network inspection needs.
• Tiny DHCP Server on CodeProject which implements a conventional DHCP server.

IE8 Search Providers, Accelerators, and Local Applications Hack

There's no easy way to use local applications on a PC as the result of an accelerator or a search provider in IE8 but there is a hack-y/obvious way, that I'll describe here. Both accelerators and search providers in IE8 fill in URL templates and navigate to the resulting URL when an accelerator or search provider is executed by the user. These URLs are limited in scheme to http and https but those pages may do anything any other webpage may do. If your local application has an ActiveX control you could use that, or (as I will provide examples for) if the local application has registered for an application protocol you can redirect to that URL. In any case, unfortunately this means that you must put a webpage on the Internet in order to get an accelerator or search provider to use a local application.

For examples of the app protocol case, I've created a callto accelerator that uses whatever application is registered for the callto scheme on your system, and a Windows Search search provider that opens Explorer's search with your search query. The callto accelerator navigates to my redirection page with 'callto:' followed by the selected text in the fragment and the redirection page redirects to that callto URL. In the Windows Search search provider case the same thing happens except the fragment contains 'search-ms:query=' followed by the selected text, which starts Windows Search on your system with the selected text as the query. I've looked into app protocols previously.

Thoughts on registerProtocolHandler in HTML 5

I'm a big fan of the concept of registerProtocolHandler in HTML 5 and in FireFox 3, but not quite the implementation. From a high level, it allows web apps to register themselves as handlers of an URL scheme so for (the canonical) example, GMail can register for the mailto URL scheme. I like the concept:

• Better integration of web apps with your system.
• Its easy for web apps to do.
• Links to URNs can now take the user to the sites the user prefers for the sort of thing identified by the URN. For example, if I have a physical address in HTML, instead of making that an http link to Yahoo Maps, I can make the link a geo scheme URI and those who follow the link will get their preferred mapping site that has registered for that scheme. Actually, looking at the geo scheme's RFC, maybe I'd rather use some other URN scheme to represent the physical location, but you get the point.

• There's no way to know if you are the handler for a particular URL scheme which is an important question for web app URL protocol handler authors.
• There's no way to fallback to an http URL in the case that a particular URL scheme isn't registered. A suggested solution to testing the registration of a scheme is for browsers to provide an additional script method to check if a scheme is registered. I don't like the idea of writing script that walks over all my page's links and rewrites them based on that method. I'd much rather see a declarative and backwards compatible fallback mechanism, although I don't know what that would look like.
• There's no way to register for a namespace within the urn scheme URI, the info scheme URI, or the tag scheme URI. I want to register info:lccn/... (Library of Congress Card Number identifiers) to LibraryThing or Amazon and I want to register urn:duri:... (dated URIs) to the Web Archive, among other things.


• Will this result in a proliferation of unregistered URL schemes with clashing namespaces? The ESW Wiki notes why this would be bad.
• And last, although this is nitpickier than the rest, I don't like the '%s' syntax used in the registration method. I'd much rather pass in an URL template, like the URL template used in OpenSearch. If an URL template is used for matching rather than registering against a particular URL scheme, this could also allow for registering a namespace within a URN. For example something along the lines of: registerProtocolHandler("info:lccn/{lccnID}", "htttp://www.librarything.com/search_works.php?q={lccnID}", "LibraryThing LCCN")