[bzbarsky]

Actually, writing a fast and compatible browser engine is a much larger project (at least an order of magnitude larger) than writing a fast JS JIT.

Just for scale, the latter takes about 2-3 years as recent history has shown, with a team that numbers a few dozen people at most. The former takes hundreds of developers, and several more years...

[Arelius]

While it may be true, we haven't actually seen a new layout engine written, so it's pretty difficult to estimate how long it'd take. Webkit was mostly functional as KHTML many years and tens to hundreds of developers ago.

[bzbarsky]

We've seen a new layout engine written. Gecko. It took about 15-16 years to get to where it is now, with on average probably ~100 developers. Granted, you can argue it wasn't the most efficient path to get there, but I doubt that you could rebuild Gecko (or WebKit) from scratch today in fewer than 5 years with a team that's about that size...

For that matter, it's not like KHTML sprang out of the void fully-formed; I distinctly recall it being written. Again, the "the path wasn't optimal" arguments may apply.

[Arelius]

I understand that a new layout engine could be an order of magnitude more work, but I'd like to be optimistic.

The problem I see, is when we start talking about projects that require 100+ men at 5+ years of work, on a project that in particular has no real viable intermediate stages, history seems to suggest to me that it just doesn't get done. The difference with gecko, is it was being advanced with the state of the web, rather than spending 10 years catching up.

A very large part of the problem is that you can't just render part of HTML5 and expect to be in a shipping, and major browser. What do you do with a half done layout engine?

Does a project like Servo really have any hope of completion if that's the case?

[bzbarsky]

That's a really good question. I wish I knew the answer.

Such projects have in fact happened. Windows Vista development started in 2001 and did not finish until end of 2006, with a much larger team than we're talking about here. Of course it's not clear to me that this was _planned_ to take 5 years...

[tmzt]

Ah yes, NGLayout :)

And of course moving from XPCOM to a more direct smart pointer implementation, and various other stages along the way.

[johansch]

More like half a dozen people and something like 12-18 months (jgraham will correct me) to make the world's fastest JS JIT. But your bigger point stands.

[gsnedders]

Carakan was 16.5 months from first commit to shipping (and a month or so more to not being notably buggy), with a team of five for the vast majority of that time. The only code carried over verbatim from Futhark was the regexp engine (though that had machine-code generation added to it), and the parser was also pre-existing (though not shared with Futhark!).

For comparison: MS were the last to do a major rewrite of their layout engine, which started, AIUI, before IE7 shipped (Oct 2006) and shipped in IE8 (Mar 2009), and more-or-less reached feature parity with others with IE9 (Mar 2011). To be fair, they were working from a far more archaic codebase than anyone else, so probably had more work, but they also had a far larger team than anyone else working on this.

[bzbarsky]

When discussing timeframes, it's important to keep the moving goalposts in mind.

Back in 2009 it took order of 16 months to write a competitive JIT. But today's JITs are a good bit faster than they were in 2009, which means more special cases that need to be considered and made fast, from what I've seen of JIT development.

I should also note that I included QA resources of various sorts in my team size estimate....

[gsnedders]

Right: I think if Carakan were done now it would likely take two years to being competitive, given the same team size (and I included QA in team size — varying between 1.5 and 2).