CLI on the Web

In the last few days Joe Hewitt has been lamenting the state of client side web development technologies on twitter. TechCrunch covered the progress in their The State Of Web Development Ripped Apart In 25 Tweets By One Man.

Today Joe followed up with a brilliant point:

joehewitt: If CLI was the ECMA standard baked into browsers instead of ECMAScript we'd have a much more flexible web: http://bit.ly/sLILI

ECMA CLI would have given the web both strongly typed and loosely typed programming languages. It would have given developers a choice between performance and scriptability. A programming language choice (use the right tool for the right job) and would have in general made web pages faster just by moving performance sensitive code to strongly typed languages.

A wide variety of languages would have become first-class citizens on the web client. Today those languages can run, but they can run in plugin islands. They can run inside Flash or they can run inside Silverlight, but they are second class citizens: they run on separate VMs, and they are constrained on how they talk to the browser with very limited APIs (only some 20 or so entry points exist to integrate the browser with a plugin, and most advance interoperability scenarios require extensive hacks and knowing about a browser internal).

Perhaps the time has come to experiment embedding the ECMA CLI inside the browser. Not as a separate plugin, and not as a plugin island, but as a first-class VM inside the browser. Running side-by-side to the Javascript engine.

Such an effort would have two goals:

• Access to new languages inside the browser, optionally statically typed for major performance boosts.
• A new platform for innovating on the browser by providing a safe mechanism to extend the browser APIs.

We could do this by linking Mono directly into the browser. This would allow developers to write code like this:

<script language="csharp">
from email in documents.ElementsByTag ("email")
	email.Style.Font = "bold";
</script>
	

Or pulling the source from the server:

<script language="csharp" source="ImageGallery.cs"></script>

We could replace `csharp' with any of the existing open sourced compilers (C#, IronPython, IronRuby and others).

Or alternatively, if users did not want to distribute their source and instead ship a compact binary to the end users, they could embed the binary CIL directly:

<script language="cil" source="MailApp.dll"></script>
	

Pre-emptive answer to the "view-source" crowd: you can use .NET Reflector to look at the source code of a compiled binary. If it is obfuscated, you will enjoy reading that as much as you would enjoy reading any Javascript in the real web.

Embedding the CIL on the browser would create isolated execution environments for each page (AppDomain in ECMA parlance) and sandbox the execution system.

The security model exposed by Silverlight exposes exactly what is needed to go beyond a language-neutral runtime. The security model distinguishes between untrusted code which is subject to verification, strict requirements on what it can do and limits what they can do from "platform" code that is trusted.

Trusted platform code is granted special permissions that untrusted code is not given. The runtime enforces that no untrusted code can call into any security sensitive and protected areas.

This would allow browser vendor to expose new APIs that get full access to the underlying operating system (for example getting direct access to special hardware on the system like microphones and camera) while enforcing that the user code accesses them only through safe gateways.

This is very important to allow developers to try out new trusted APIs: new UI models, rendering systems and APIs built entirely on the same core.

I am absolutely fascinated by the idea and I only regret not having come up with it. We have been too focused on the Moonlight-as-a-plugin to take a step back and think in more general terms: how can we use the ECMA CIL engine for *all* applications without a browser plugin.

Joe like many of us is conflicted between the power of reach that the web has versus the polish and speed that native GUI toolkits deliver.

Althought Silverlight provides a nice UI system inside the plugin, Joe's point is that we need a platform in which we can more quickly innovate new UI ideas, and probably completely different new ideas quickly. And with the ECMA sandbox model we can start testing new ideas without waiting for browser vendors to add the features themselves and we can make the integration between these plugins and the browser stronger than they have ever been.

The plugin model does not provide the necessary tools to drive more innovation in the web. We need a new model, and I am ready to start prototyping Joe's idea.

The only question is what browser to target first Firefox or Chrome.

Gestalt

Gestalt allows developers to use the CLI inside the browser and it shows what can be done with other languages in the browsers. It requires Silverlight to run and the interaction between the code and the browser is limited in the level of integration that can be achieved via the browser-plugin interface.

This solves only the half the problem: multi-languages and does it in a limited way.

Resources

You can learn more about the security model in my previous blog post.

Our open source Silverlight implementation.