Hacker News Reader

> So to me, a putting Dart source file right next to Dart2JS output on the open web is no different than asm.js, and I think both are equally valid approaches to moving the web forward.

I think this is potentially interesting comparison. Yes, both Dart and asm.js are tools that are involved with both generating content and executing that content (unlike your first example with servo, which is apples-to-oranges for the reason you mentioned).

But there are big differences here. With asm.js, it was a specific goal to make optimizing the code in other browsers as simple as possible. That was one of the main reasons for actually defining asm.js, so there is a clear "contract" - "this is what compiled C++ code looks like, these patterns are easy to optimize in obvious ways." That avoids the risk of there being anything special about generating code using emscripten and executing it in Firefox: the generated code is asm.js with the "contract" just mentioned.

For Dart there isn't anything comparable. dart2js emits JavaScript code, which runs in all browsers, but there is no reason to think it's performance is or will stay similar to the Dart VM. Instead, the Dart strategy there seems to be more along the lines of "the Dart VM is open source, if you want comparable performance to Chrome you can adopt that codebase." There is merit to that approach - it's much better than if Dart were closed source - but it is quite different than the one in the previous paragraph.

The second major difference is that while both Dart and asm.js relate to both content creation and execution, one is made by Google, which owns several of the most important sites on the web, whereas the other is made by Mozilla which obviously is nowhere near anything of the sort and not even trying to be.

Why this last point matters is that if the VM eventually becomes much faster than the dart2js path, Google can essentially force other browsers' hands by adopting Dart on Google sites. Imagine if google.com, gmail, google docs etc. all run much faster on the VM than otherwise - other browsers would be forced to adopt the VM to stay relevant. This is huge leverage that Google has. You can't be a browser that people use if you are not competitive on Google-owned properties.

Going back to the previous point, should that happen, adopting the VM might be the only option - there is no way, unlike with asm.js, to get similar performance otherwise (in the asm.js case, the high performance is achieved with a small modification to an existing JS engine, which could be done on other engines as well - no one needs to adopt SpiderMonkey). So this is the potential bad scenario that could happen due to both of these issues.

Obviously there are similarities between Dart and asm.js too, as you mentioned. But I believe the two factors I described explain much of the concern some people have about what Google's plans with Dart are. It might all end up ok, but Google might also leverage its huge influence in multiple areas for what is best for Google, and not for the web as a whole. There is no such risk with asm.js.

I see where you are coming from, and I think this was a very well-reasoned argument.

In my opinion though, your argument prioritizes browser vendors over users, and over the competitiveness of the web vs. competing technologies like native apps.

Let's take your scenario, where somehow the Dart VM become becomes noticeably faster than Dart2JS on Google properties, to the point that users notice and get a better experience by using a browser with the Dart VM embedded.

First of all I'd mention that there is a natural competitive pressure that keeps Gmail and other Google properties "honest" so-to-speak. Chrome is far from being the only web browser, and if Gmail is perceived as being slow compared to other webmail websites, Gmail will suffer accordingly. Gmail has no incentive to cripple the speed of JS-only browsers; it has the incentive to make its app as fast as it possibly can on all major browsers.

So if hypothetical future Dart-Gmail is noticeably faster than JS Gmail, then it is likely for some fundamental reason of Dart being more optimizable, and not because of any lack of effort to make the JS version fast.

At this point, where the Dart version is faster for purely technical reasons, you say that "adopting the VM might be the only option" for other browsers. I would say first of all that it seems that there are other options, like compiling Dart to asm.js in a very peripheral part of the browser so that you wouldn't have to change your core execution VM. Given that the Dart VM is open source, this would be adding a new backend to an existing compiler (sort of like you did with Emscripten and LLVM), and not an effort comparable to developing a new compiler from scratch. It's true that this would require you to adopt some third-party code, but it would keep the Dart VM out of your browser, if not the Dart compiler.

But suppose that even that wasn't acceptable or feasible. Now the question is, is it better for (some) users to get the advantage of technology that has demonstrated that it is inherently faster, at the cost of other vendors being pressured into adopting this (standardized) technology, or is it better for all users to be equally disadvantaged, to keep the browser playing field more level?

As you surely know, browsers aren't just competing against each other, they are competing against other app platforms like native apps. If the web stays a little slower but all browser vendors are more equal, then the whole web is falling behind compared to the other, far less open and level, app platforms.

Now all of this discussion is quite premature until/unless Dart is actually distributed as part of Chrome, and the Dart VM is actually found to be noticeably faster than regular JavaScript, and Google actually uses Dart on its big properties.

Well, the Dart VM is not like adding a backend to a compiler or VM - it is a complete VM in itself.

For example, it has its own concepts of references etc., which means that if you add the Dart VM alongside your JS VM, you need to be able to collect cycles spanning those two.

That is not a trivial thing to do, and in fact Google has been working for a while on the Oilpan project to make Chromium capable of collecting such cycles, that span C++, v8 and the Dart VM. It's a huge engineering effort. Gecko can currently collect cycles between C++ and JS, but extending that is not trivial, and I believe Safari cannot collect such cycles and instead by design avoids creating such cycles - but that would not be possible with the Dart VM.

It might not be practical for other browsers to adopt such a technology. So if Google sites force the Dart VM to be necessary to compete, that would basically mean that the other browsers cannot compete. Then there would be just Chrome, a single implementation of a web browser, dominating the market like IE used to.

It might be good for users to have a single browser for a while, single implementations can move faster without regard for standards or openness. But lack of standardization and openness also means it is a proprietary platform, of which we already have many.

> Well, the Dart VM is not like adding a backend to a compiler or VM - it is a complete VM in itself.

My straw-man idea was not to add the Dart VM to the browser, it was to add a Dart-to-asm.js compiler to the browser, as an AOT compilation step. This would not involve integrating the VM at all.

> It might not be practical for other browsers to adopt such a technology.

Given that there are at least three completely independent and highly competitive JavaScript JIT-compiling VMs from a variety of different browser vendors, I find this unlikely.

I think we are four levels into hypotheticals here. If every technical advance was rejected due to something bad that might happen four steps down the road, nothing would ever improve.

We surely never would have gotten JavaScript. That could have gone badly in tons of different ways (especially since, as the first execution engine in a browser, there was no backward-compatibility strategy). Who would have believed in 1995 that 15 years later this technology would have gotten pushed so far, and that the competition would be so healthy?

> So if Google sites force the Dart VM to be necessary to compete, that would basically mean that the other browsers cannot compete. Then there would be just Chrome, a single implementation of a web browser, dominating the market like IE used to.

The entire web platform is standardized (including Dart, which is standards-track). I would challenge you to name a standardized technology with a market comparable to the web in which one vendor is so dominant that all of the other competitors are defeated and it becomes a monoculture, despite a large demand.

If anything, I think the prospect of a web that cannot keep up with other application technologies far more likely and troubling.

> My straw-man idea was not to add the Dart VM to the browser, it was to add a Dart-to-asm.js compiler to the browser, as an AOT compilation step.

Oh ok, sorry for misunderstanding you.

That might work, but it is unclear it would be any faster than dart2js. Also, AOT is not likely to work in Dart which is optionally typed, unless a codebase is actually fully typed. Overall it seems an interesting experiment perhaps - I've suggested it myself - but an unlikely one.

> I would challenge you to name a standardized technology with a market comparable to the web in which one vendor is so dominant that all of the other competitors are defeated and it becomes a monoculture, despite a large demand.

Well, the web - back in the IE days.

Also to a lesser extent the smartphone mobile web which was a monoculture of WebKit (and to some degree still is today).

And the point is that standards become irrelevant when there is a monoculture. In theory windows supported various POSIX APIs, but in practice windows became a proprietary monoculture.

"if the VM eventually becomes much faster than the dart2js path, Google can essentially force other browsers' hands by adopting Dart on Google sites"

Or all of the browser vendors could sit down and design a new language based on the experience gained from implementing Dart. This new more easily optimizable language can become a new web standard. After all this is what happened with SPDY and HTTP2.0. This "new language" could also just be a new version of Javascript with a few backwards incompatible changes.

One interesting claim of the DartVM authors is that they have achieved a 10x start up gain over Javascript when using script snapshots. If this turns out to be true, it will add significant pressure for other vendors to implement a similar feature. The only way to get this to work in Javascript is to make backwards incompatible changes. For example only allowing a static top-level structure, as in Dart.

I see the competition from Dart as being good for end users, as it adds pressure on Javascript to evolve. It's also interesting to see the V8 team porting a number of Dart's optimisations to V8.

> To me this is like saying: "it's safe to defend Servo up to the point that it's faster single-threaded browser engines. Then Chrome and IE might fall behind in performing for web apps and then you will see how Servo will hurt the web ecosystem."

We are not implementing anything in Servo that does not have cross-vendor support and is on standards track. That's not the same as Dart.

I address this argument in the rest of my comment. Asm.js is not standards track as far as I know.

Actually let me post a thought experiment. Suppose that Dartium, instead of processing straight Dart source, processed a format that is like JS except that all the Dart is in JavaScript comments at the top of each Dart2JS-outputted function.

And suppose developers compiled their Dart to these JS-with-Dart-in-comments files, and distributed these on the web instead of Dart and JS as two separate files?

At that point, how is Dart different than asm.js?

> At that point, how is Dart different than asm.js?

It's not guaranteed that the Dart VM and the dart2js-compiled code implement the same semantics. (And, in fact, they don't, by design—bignums operate differently in JS and the Dart VM, which could break sites in non-Chrome browsers.)

No one has a problem with Dart2JS. We welcome more language choice on the web. People have a problem with Dartium because it's an attempt by one company to bully the rest of the web into adopting their own proprietary technology.

It's no different than putting an asm.js processor into a JS engine. asm.js is equally "proprietary" (actually more so, it's not on any standards track), and if it's truly faster than plain JS, then it's equally "bullying."

(Note: I actually like asm.js a lot and don't think that healthy competition is "bullying").

Are you being serious? asm.js is basically a "best practices" guide for js vm authors. It's not required to run asm.js code, or even guaranteed to be faster; other js engines have optimized for asm.js code without treating "use asm" special.

Dartium requires the end developer distribute 2 payloads, one for Dartium browsers and 1 for JavaScript browsers.

> asm.js is basically a "best practices" guide for js vm authors.

No. It is a specifically designed subset of JS that also has extra annotations. If a block of code does not adhere to the rules, it is not asm.js and will be processed as regular JS. If it validates, it is processed by a different compiler pipeline and AOT compiled.

It is much more accurate to say that asm.js is a newly designed byte-code that very cleverly happens to use backward-compatible JS as its serialization.

Native Dart is also not guaranteed to be faster than Dart2JS.

> Dartium requires the end developer distribute 2 payloads, one for Dartium browsers and 1 for JavaScript browsers.

Ah but that is only extra work for the developer. Similar to compile-to-JS languages like TypeScript, CoffeeScript, Emscripten, etc. there is no problem with new technologies that impose extra work on the developer, because it is the developer's choice to use it or not, so the person making the decision is the one paying the extra administrative cost.

It's extra work that a lot of developers aren't going to do. "Works in Chrome, closes ticket". Can't even get developers to use css prefixes properly.

Hence why I called Dartium bullying. It's an attempt to reinvent the web with Google in control. asm.js is just an attempt to make existing vm optimizations a little bit better (and again, others disagree that they are needed).