...no they weren't? They let you embed native code (platform-specific) or LLVM IR (vendor-specific) into a webpage. I don't see what that has to do with support for other languages. Anything NaCl can do, asm.js can do too, but one of these has multiple implementations, is backwards-compatible, is truly cross-platform, is well-specified, and is standardised. (Hint: it's not the first one.)
14paninta's comment is dead again - shadowban? Anyway, I'll repost and respond:
> asm.js doesn't support threading.
asm.js supports threading, actually, through Web Workers and SharedArrayBuffer.
> it can't run native code.
This is true, but asm.js is relatively close to the metal. It's not very far from assembly language.
> it doesn't support meaningful debugging.
This is a tooling issue. While the situation is bad at the moment, it'll surely improve.
> it doesn't have a post-DOM API surface like pepper.
Why do you need a "post-DOM API surface"? If you just want to blit stuff to the screen, the DOM is actually no less efficient than Pepper: you can directly pass it pixel buffers, or call OpenGL routines.
Asm.js could be considered close to the metal if you were running on a Pentium 166. I'd consider it close to the metal if it had features modern processors have, like SIMD, operations on integers that have other than 32 bits, IEEE-754 compliance, etc.
With the threading issue, Web Workers are boxed up so tightly that there are a lot of operations, like passing objects between threads, or letting web workers have access to portions of the DOM, are impossible. Additionally, because everything's passed by value, there's a lot of copying that needs to take place; to expand on my prior example, sending a message to another thread would involve copying the object to send to the main thread, then copying it again to send to the destination thread. The alternative, transferrable objects, completely deletes the object being passed, and still can't be transferred between threads. Finally, there is no pooling control among threads, which means that there is a lot less control over what threads are running at any moment.
These problems all come back to the fundamental issue. There's no escape from javascript. Asm.js is fundamentally hung off of javascript with bailing wire and duct tape, out of fear of "breaking compatibility". Anyone that tries to experiment with a new language is immediately accused of trying to fracture the web. Thus, we stagnate, and everyone suffers for it.
I've never used web workers, but as a developer who has written a lot of multi-threaded code in C++, I'm kind of glad that as a user I don't have to suffer poorly written non-thread-safe JavaScript. I've seen some terrible multi-threaded code by developers who just didn't get it. I suppose that's why web workers are designed the way they are: to make it easier for developers who maybe don't quite get the intricacies of multi-threading.
> I'd consider it close to the metal if it had features modern processors have, like SIMD, operations on integers that have other than 32 bits, IEEE-754 compliance, etc.
It has all of these things. It has SIMD now, 64-bit support is improving, IEEE-754 is supported (though some of the checks aren't available yet, I'll give you that). Give it time.
> With the threading issue, Web Workers are boxed up so tightly that there are a lot of operations, like passing objects between threads, or letting web workers have access to portions of the DOM, are impossible. Additionally, because everything's passed by value, there's a lot of copying that needs to take place; to expand on my prior example, sending a message to another thread would involve copying the object to send to the main thread, then copying it again to send to the destination thread. The alternative, transferrable objects, completely deletes the object being passed, and still can't be transferred between threads. Finally, there is no pooling control among threads, which means that there is a lot less control over what threads are running at any moment.
Yes, but this stuff isn't applicable to asm.js. It shares a heap between threads, it's just like your old native code, and everything is triggering race conditions^B^B^B^B^B^B^B^B^B^B^B^B^B^B^B^B^B^B^B^B^B^B^B^B^B^B^B hunky dory.
> These problems all come back to the fundamental issue. There's no escape from javascript.
What's wrong with that? If you don't like JS, you can compile to it. It's an excellent compile target.
> Asm.js is fundamentally hung off of javascript with bailing wire and duct tape, out of fear of "breaking compatibility".
It's not just a compatibility hack, though managing to work on everything with no updates is a killer and awesome feature. It also avoids unnecessarily reinventing the wheel. It reuses as much of JS as possible. That's good.
> Anyone that tries to experiment with a new language is immediately accused of trying to fracture the web.
...well, yes, they're fracturing it. And usually needlessly. There'd be no benefit to a new language on the web. There's nothing a new language can offer which JS doesn't have.
> Thus, we stagnate, and everyone suffers for it.
What stagnation? Everything is moving rapidly. Things are improving all the time.
Simd.js is only supported on one and a half browsers, is only supported on a single platform, and only supports two data types. I'd hardly call that supported, more like being hacked together. Its current implementation requires you to waste quite a few clock cycles and bytes of memory if you want to do things like graphics operations. 64 bit support is at the, "hey, you can store numbers. Have fun creating all the operations for them" stage." Furthermore, it deviates from the IEEE-754 standard in things like number handling -- Math.max(5,NaN) should return 5, not NaN. This means that libraries written in other languages and transpiled using emscripten and the like will not behave as they should.
I'd say that the weaknesses of web workers is absolutely relevant. Though you see it as protecting the coder from themselves, there are design patterns out there, like publish/subscribe, that are useful in letting threads communicate, but not generating race conditions. DOM access and the like can be done sanely, but Javascript doesn't even give you that option.
Given the reasons described, it's a rather poor compile target. It's a plodding beast of a language that mishandles numbers, requires circuitous coding to handle basic concurrency, and otherwise behaves in ways other languages don't. Transpiling just barely hides its flaws.
I will restate that it is a hack, and a rather ugly one. Its attempts to maintain compatibility leave us with a virtual machine that is increasingly obsolescent. An "assembly language" designed in 2013 shouldn't be lacking support for 64 bit integers (or 8-bit or 16), should have good support for vectorization, and should provide some sort of register and/or stack model so that you can better clue in the vm as to what items are going to be frequently used in functions. The missing features could be handled much like JSON was early in its life, by including a library as a fallback. It would probably end up looking ugly as hell, but at the same time, it would work, and give a lot more long-term benefits.
Regarding alternative languages, Javascript isn't always the best language to describe a problem in. Sometimes your program's better written in C, or Lisp, or Fortran even. Why should you force your problem to fit the language?
Regarding stagnation, you can move quickly, yet not really go anywhere useful. COBOL had 7 major versions and several minor in 14 years. Does it mean that it's a useful language today? The stagnation comes in terms of how people think of problems, and how they try to wedge those problems to fit into Javascript. The web would be a much better place if we allowed more experimentation into languages and behavior.
Well, Dart would've been one other language in the browser, sure. But it's a language not that different from JavaScript in actuality. The benefits seem few for the work needed to support two different languages. Compiling to JS, which fits Dart quite well (probably because it was designed to compile to JS well), seems more productive to me.
Some were excited Dart might lead to some sort of bytecode VM being added, but that didn't lead anywhere. And since then, JS has become more bytecode-like, so it doesn't really matter now anyway.
As more and more people write software in languages that compile to JavaScript instead of directly hand-writing JavaScript the demand for JavaScript becoming a better compile target grows.
In that sense adoption of Dart, ClojureScript, TypeScript, Elm et all will continue to accelerate JavaScript (language and runtime) improvements.
All part of the plan. I've been saying this since 2012, Dave Herman and Alon Zakai of Mozilla Research (asm.js creators along with Luke Wagner) said it before me:
> but one of these has multiple implementations, is backwards-compatible, is truly cross-platform, is well-specified, and is standardised
None of these problems would be difficult to address if the four-member vendor cartel which controls web "standards" was willing to adopt NaCl. Of course they don't want to adopt it, but that is the real problem right there. Of the four, two members (at least—maybe now three) have an active interest in holding back the Web to protect their native app platforms, while one would rather reign in Hell (through its position of shared mastery over the Web hairball) than go broke in Heaven. As to support for other languages, I only have to point out that one of Mozilla's reasons for decrying Google's intended browser support for Dart was that Dart compiled to JS would not be competitive with the native Dart implementation.
Nope! (On everything you wrote, including "the" and "and".)
Also: loser talk is loserly. PNaCl (did you mean PNaCl not NaCl? NaCl was never proposed by Google for the Web) like Dart fell to physics. All else is stamp collecting.
You didn't link to any of my comments, so let me help your reading comprehension.
My point then was that if Google had had MS's IE4-6-era market share and lack of scruples, they could have pushed Dart into Chrome and started using it in their web properties, and the lack of equivalent performance between DartVM and dart2js -- especially due to lack of bignums and other affordances in JS, which lack Google materially increased by choosing to invest in Dart over JS -- would have stunk. It would have verged on the monopoly abuse of which MS was in fact convicted.
My point was not that JS must be the only way to run Dart simply because I or anyone at Mozilla (or anywhere else, MS and Apple included) prefers JS. That would be dumb, and it's clearly not what I wrote.
Browser vendors have to keep JS improving. That's a given. Adding the extra costs of DartVM and the inter-heap GC and write barriers to glue it into the shared DOM client side is a huge tax, which would have helped kill Mozilla faster. MS and Apple wouldn't pay that tax. Why should anyone? Again it's not about JS being uncompetitive. "Competitive" is what browsers are when they run web content better and faster. Dart has very little to do with this.
JS is very, very hard to replace, for reasons I summed up as "physics" (apologies to Rutherford) and won't belabor here. Yet the Dash memo proposed to do just that, to replace JS, using lame assertions about how JS couldn't be fixed.
That was not just lame posturing and excuse-making for a Lars Bak retention program. Given Google's power, it was a disservice to the Web as a standards-based and ever-evolving platform, whose caretakers are obligated to keep working on it in the open, not try proprietary end-runs. (Yes, of course, many people at Google work hard to improve the Web. Good for them, but not relevant to what happened with Dash.)
I'm happy to say I was wrong to worry about Google succeeding in acting like MS did with ActiveX/VBScript. Not that some at Google didn't try (or won't again with other stuff), but they failed with Dart. I predicted that, too, as an ultimate outcome after wasting years and megabucks -- including real opportunity costs to V8 and JS. So too did many at Google predict this crater.
Dart's a compile-to-JS language now, not due to any preference for JS by anyone at any browser vendor.
> My point was not that JS must be the only way to run Dart simply because I or anyone at Mozilla (or anywhere else, MS and Apple included) prefers JS. That would be dumb, and it's clearly not what I wrote.
I've never suggested that this is what you were saying there. Here I pointed to it as evidence that I was right that "one of Mozilla's reasons for decrying Google's intended browser support for Dart was that Dart compiled to JS would not be competitive with the native Dart implementation", where by "competitive" I was obviously referring to performance. Since we're agreed on that, the claim that everything I wrote above was wrong 'including "the" and "and"' goes out the window. (What I did point out to you at the time is that this was hard to square with the claims being made elsewhere for JS' wonderfulness as a compile target.) Your take on Google's intentions in proposing Dart is also completely in line with my argument that the vendors of major browser engines have huge power basically not constrained by anyone besides the other three vendors, as well as motives that don't always align with the public interest.
The "and" and "the" line was from Mary McCarthy in reply to Lillian Hellman. You're Hellman :-|.
You wrote:
"As to support for other languages, I only have to point out that one of Mozilla's reasons for decrying Google's intended browser support for Dart was that Dart compiled to JS would not be competitive with the native Dart implementation."
The empty rhetorical flourish ("I only have to point out...") aside, this is rank question-begging. You assume that JS as Dart target being too slow must mean DartVM should win because of "competitive" force. Wrong on at least two counts (I'll ignore market power abuse):
First, DartVM may be faster at semantics JS supports due to optimization problems on the JS engine side. V8 in particular lost its founding team to DartVM, except for some moonlighting. Therefore "competitive" does not mean that JS cannot improve to be as fast when compiled from Dart as when the Dart source runs in DartVM.
Second, DartVM may have types and optimizations that JS lacks, bignums for example. Again "competitive" does not mean JS cannot change, and in fact bignums have been on the ECMAScript Harmony agenda since 2010, before "Dash" leaked.
In any competitive market with deep tech, there are lots of ways to meet a set of developer-facing language support/performance requirements. Even with all Google's billions, Dart and Chrome decision-makers couldn't ignore the compile-to-JS alternative and continue to justify developing and shipping DartVM + OilPan + all else not fully in hand to support JS and Dart with no performance regressions. Compiling to JS is the sole path forward for Dart on the Web precisely due to competition.
Citing my argument about bignums is perversely wrong, unless you assume DartVM is the only solution to any DartVM vs. dart2js performance disparity.
Such an assumption, like the one you seem to have made about NaCl, is anti-competitive, uneconomic, magical thinking. It requires indefinite amounts of time and money to fund all the work to have two VMs, a super-GC to collect cycles, write barriers as part of that super-GC, and more optimization to overcome the performance regressions imposed by those barriers (if possible). It assumes JS can't more cheaply be extended to be a fast-enough target language. You offer no proof.
BTW, you started out with NaCl and I still think you're mistaken about something pretty fundamental. NaCl was never proposed as a Web standard. Ok, suppose you mean PNaCl. The problem remains that PNaCl required a big API, Pepper, to reach the guts of chromium/Blink (originally WebKit) and the underlying OS. The problem with PNaCl was always Pepper, and again: competition killed Pepper as a cross-browser API.
It takes longer (it would have happened sooner with more cooperation from Google earlier, instead of the malinvestment on these two big follies), but JS is getting SIMD and shared memory threads, which are among the last few bricks still standing in the Pepper wall.
So PNaCl is evolving with Emscripten/asm.js now -- precisely due to competition -- and I would not hold my breath for PNaCl to remain based only on LLVM bitcode (with fixes requiring years and millions, to remove unspecified behavior). But feel free to hold your breath and then blame the "vendor cartel" if you like.
On that "vendor cartel" point, there are more web engines coming along, Servo (an open source project with multiple paying companies involved) among them. There's no cartel, just a big compatibility hill for any new engine to climb.
Adding gewgaws like NaCl and Dash slows down that climb. It's uncompetitive for the would-be market entrant. Why you think this requires collusion ("cartel") or Satanic pride ("reign in Hell") is unclear. Do you get many billion-dollar free lunches where you live?
