I think that you are missing the obvious: it worked.

This is absolutely correct. And, if you check, this is what any system that wants to be able to manage tens of thousands of threads does (not necessarily with the exact same techniques). Including all major databases, all large scale virtual machines, etc. Recent versions of MacOS X even provide system libraries to simplify the implementation of such user-level threading (under the name of Grand Central Dispatch).

Actually, there is one. It is just not exported to userland yet.

Ah, I think I finally understand your point. It is about compatibility, not about paradigm, is it?

If so, you may be happy to learn that the low-level layers of Opa can already access other databases. The feature is not accessible at high-level yet because we are going to make a number of breaking changes before long, but eventually, we do intend to give access to most mainstream databases. Would this solve your problem?

Self-promotion on Slashdot? Nah, this never happens.

Actually, it's a graph database and one of the most powerful database engines I have seen on the net. With this engine, you store complete data structures, thinking in terms of references instead of keys – except when you want to think of keys, of course. What is the problem with that?

Good point. Unfortunately, as the CSO, I have strictly no say in the license. If you wish to discuss it with someone who does, you should ping phy_si_cal, our CEO and the OP of this thread, or send an e-mail to Mathieu Baudet, our COO.

Good point. We had really hard times tweaking the compiler until it produced JavaScript that worked on all browsers (note: in our books, IE7 is not a browser), or html that worked in all browsers, and with linking with all mainstream JS libraries (note: thanks Facebook for the nice code provided, that explicitly depends on server misconfigurations and/or browser bugs and suddenly stops working when confronted to a standards-compliant web browser and a well-configured server), and with getting our Comet to work with all the different possible interpretations of "asynchronous".

But that is not the point. The point is that this is a problem that *every* non-trivial web application needs to face. The point is that we faced it and we beat it, so that Opa developers do not need to. Should a new browser version arise and cause new incompatibilities, *we* will be the ones doing the heavy debugging until we can produce a new version of the compiler and runtime that lets developers forget about the incompatibilities.

So yes, these brittle bases made our work difficult (and interesting :) ). But they also constitute an important part of the interest of Opa.

Caveat I'm part of the Opa team. Well, worse than that, I am the architect-in-chief.

Well, in my experience, most of Scheme (i.e. not call/cc) ports quite well to JavaScript and performs quite well once compiled to JS. Performing a CPS transform on JS cuts performances by roughly 50%, which is not *that* bad.

I can understand the issue. Would this calm your concerns if I told you that

• on the Opa git, there is an Opa version for Windows – it's not quite ready for prime-time, but it's there;
• Opa is designed to be able to link with other server-side technologies – for the moment, OCaml and C, other technologies will arrive in time;
• at low-level, the Opa database can access other, mainstream, database – the feature is not quite exported at high-level yet, because we are not quite satisfied with it yet, but in time, it will;
• and of course, there is always the option of modularizing your code through the use of web services, which is a nice way of allowing [hot]swapping of components.

Of course, what do I know about compilation? I only have a PhD on the topic and 9 years of professional experience.

Yeah, well, nobody in the team speaks Lithuanian :)

Interesting question. Let's forget about the framework for the moment and concentrate on what the compiler should do:

• analyze statically the whole code and determine what goes on the client and what goes on the server (I tend to believe that this is not feasible with a dynamic language, but of course, feel free to prove me wrong);
• compile client-side code to JavaScript or to code that can be interpreted by a JS-based VM (lots of work, but no theoretical barrier, emscripten can even make your life somewhat easier at least for C++);
• inject Ajax entry points on the server (no real problem there);
• inject Comet re-entry points on the server (requires a server-side language that supports either CPS or tens of thousands of lightweight non-blocking threads – theoretically possible with custom implementation of the languages).

That's without security or distribution, without database, without a syntax for css or html, but it looks somewhat possible for some languages, with considerable effort. Say, if you attempt to do this, don't hesitate to ping me, it might be interesting.

Just for clarity: I'm David, the Chief Scientific Officer. Mathieu is the Chief Of Operations.

One word: sandbox.

Frameworks definitely would not suffice. Reaching the level of features of Opa, even without security, requires quite complex static analysis that strictly eliminates Python, Perl or Ruby, and in practice, even C++. Now, it might just be possible with Java or C#, with non-trivial changes to the language and compiler.

Unsurprisingly, we agree.

Caveat I am part of the Opa team. Well, worse than that, I am the architect-in-chief.

That's an interesting claim. Do you have any example to back it?

Er... the opposite of what? Still, I think I understand your point – see a past discussion on my blog on the topic of syntax.

Not my call, unfortunately. You may wish to convince phy_si_cal (our CEO) or Mathieu Baudet (or COO), though.