Aside from my personal distaste for his backing up the semantic truck and dumping it into ES6, I think it's a bit annoying -- to the point of being disrespectful -- that Brendan doesn't mention V8 in his history of JavaScript. Without V8, there is no JavaScript on the server-side (sorry, Rhino and SpiderMonkey), there is no Chakra and there is no TraceMonkey/JagerMonkey/IonMonkey: given that JavaScript had survived for a decade in its pre-V8 state of abysmal performance, it's entirely reasonable to assume that it would have slept away another decade had it not been jarred out of its slumber by V8. So it would be nice to see some respect where it's clearly due...
> I think it's a bit annoying -- to the point of being disrespectful -- that Brendan doesn't mention V8 in his history of JavaScript
Bullcrap. Why would V8 deserve any more mention than Spidermonkey (which had none, by the way) in a talk about the semantics and language evolutions of javascript when it's Gecko/Spidermonkey which pioneered and implemented roughly 95% of these evolutions in the first place? Why would it deserve more than Trident, which — through xmlhttprequest — is the one responsible for the vast majority of the language's actual popularity? Because it's your pet runtime and you dont like others?
> So it would be nice to see some respect where it's clearly due...
You may want to take this advice for yourself, your comment is dismissive, insulting, contemptuous and contemptible.
It was, and still is. It chides Eich for not including a blurb about something which has little to no relation with the presentation itself (the presentation's core was not javascript runtime performances or even javascript runtimes in general), and claims "disrespect" over that non-inclusion, all the while — as other commenters also noted — getting most if not all of its assertions wrong.
v8 might exemplify the JS performance trend, but it didn't start it. Mozilla, Google, and Apple (with squirrelfish extreme) independently developed JS JITs at around the same time, driven by the increase in JS-heavy web apps.
So no, JS performance wouldn't have "slept away another decade"; the changing web situation demanded the change (as evidenced by the fact that all 3 orgs independently moved in the same direction at essentially the same time).
Edit: Actually, I just realized I'm off here. The June 2009 release of FF was a final release, while the September 2008 release of Chrome was a beta. From what I can tell, though, the beta release of TraceMonkey was a day after the beta release of Chrome.
>TraceMonkey shipped before v8 did.
No, that's not true. TraceMonkey shipped in June 2009 with FF 3.5. V8 shipped with the first release of Chrome in September 2008. At the time, TraceMonkey was in beta, but it's unclear which began development first.
That said, the idea that V8 sparked the JS arms race is preposterous.
Where he says "[...] TraceMonkey, which we launched ahead of Chrome and V8".
Maybe "shipped" was the wrong word. I suppose there's some definition of "launched" that makes the statement true; tracemonkey landed, and was announced, in August 2008. But you're right, the Chrome beta (Sept. 2, 2008 according to wikipedia) did precede FF 3.1 beta 1 (Oct 18, 2008).
TraceMonkey was announced on 2008-08-23, after 2 months of development[1].
V8 had its first public release together with Chrome 2008-09-02 as you already mentioned, but development appears to have started in 2006 according to some of the copyright notes in the initial SVN export[2].
Yes. I did not take Bryan to task on that, but it really gives V8 a bit too much credit for it to cause Andreas Gal to work on trace-JIT before 2006 (on Java, for his UCI PhD; then on JS in collaboration with Adobe and Mozilla).
Tracing was a good rocket to strap on SpiderMonkey-the-2008-era-interpreter but it fell to a combination of the PIC-based approach V8 championed and Brian Hackett's Type Inference work (PLDI 2012, http://rfrn.org/~shu/drafts/ti.pdf).
He meant server-side JS was relatively non-existent before node.js, he even listed alternate server-side JS impls that did use server-side JS, nothing of which is anywhere close to the phenomenon that node.js is now.
> I think it's a bit annoying -- to the point of being disrespectful -- that Brendan doesn't mention V8 in his history of JavaScript.
Chill. It's a timeline of language evolution - I don't see how V8 is relevant. Mentioning the AJAX revolution isn't strictly relevant either, but serves the purpose of separating orthodox from next-gen Javascript.
Btw, while work on V8 started 2 years before work on TraceMonkey, apparently the latter was announced first.
The latest date I can pin down for the beginning of the JS arms race was Safari's JavaScriptCore in March 2008, followed by SquirrelFish in June. V8 has had a huge impact in driving innovation and setting the pace of the race, but it did not fire the starter pistol.
First, it's not all about _moi_. As with any mature language with multiple implementations, there is a committee, Ecma TC39. It has reps from all the bigs plus PayPal (Doug Crockford) and Yahoo!. We avoid design by committee, instead focusing on paving the cowpaths and in a few cases working with champions -- single innovators or RPG's "resonant dyads" _a la_ ken&dmr -- to do focused design. The best example in my view is the Proxy design by Mark Miller and Tom Van Cutsem, followed by Modules by Dave Herman and Sam Tobin-Hochstadt.
We have a few goals, including minimizing kernel semantics and extending syntax for usability only where we have experience and positive user-testing results (e.g., destructuring, let, generators, classes as prototypal sugar a la CoffeeScript).
In this light, we aren't dumping new semantics into ES6. If you count fairly, we are filling the big semantic gaps in ES5 (AKA JS as we know it). For instance, where is client-side synchronous "require" without the ES6 module system? No fair that Node's require is sync (against all sound doctrine!). Sync XHR is from the jank-devil, and we cannot do sync require as an API client side.
Second, as others have noted, V8 was stealth'ed for two years (I learned of it in 2006 but Google withheld, see https://brendaneich.com/2011/06/new-javascript-engine-module...) and then unleashed just as Apple and Mozilla got their JIT acts together. At the time some said V8 was 3x faster, but the actual ratio on stupid benchmarks was more like 1.3x at first.
V8 is an amazing piece of work, but it did not contribute to the language evolution that my "very brief history" slide diagrammed. It certainly contributed by being the fastest, but until after 2010, that was the extent of its influence, for several specific reasons.
For the first few years as open source, V8 foreswore diverging from JavaScriptCore (the first WebKit engine, Apple's fork of KJS from KHTML, which V8 in some ways forked WebKit to replace) on any new ES5 or ES6 proposals, to try to keep on (or get back on, in view of the fork) Apple's good side. Fat chance!
Even now, V8 still keeps prototyped ES6 implementations hidden behind a flag. I expect this flag to be lifted before ES6 is ratified.
Another thing that hurt: Google has moved V8 to Munich because the Aarhus team wanted to do Dart, Google explicitly chose to invest in Dart over JS, and the no-remote-teammates rule hurt. IMHO the Munich team is quite strong but only ~10 people. This limits its effectiveness compared to Chakra (~60 heads?) and even SpiderMonkey (~12, more with interns).
Finally, whoever thinks I disrespect V8 clearly does not listen to my talks. The video will be up, give it a view when it's live if you have time. There is no disrespect from me toward V8 or the team that built it. I do mention it favorably in all talks, including Strange Loop (if my memory serves).
Lars Bak did the Strange Loop first night closing keynote, which my insane schedule prevented me from attending. He and I have had no meetings since that one in 2006, although he met with some Mozilla colleagues in 2009. I bear no bad feelings, although Lars did tell a gratuitous whopper around 3:50 in this talk: http://www.youtube.com/watch?v=T2TJYBmDZHI (see my blog cited above for what happened with Mozilla being "offered" V8 in 2006 -- Google reneged). He is the best VM hacker I know of. (Mike Pall is another superhacker of note.)
I hope this sheds some light. I respect V8 while noting that, as with anything human, it and the technical/product/open-source/standards politics around it show some significant flaws. That is inevitable.
Mozilla has flaws too, but they tend to be different in kind. For example, we do open from the start and try to partner even against long odds (e.g., Tamarin/ES4). I'm not bragging, because I think that can be a mistake. I've learned from V8 and modified Mozilla strategies accordingly.
Brendan, I appreciate the thoughtful reply. I think the only reason for my criticism (and the word "disrespectful" was too harsh -- I should have left it at "annoying") was that your last slide ("Always bet on JS") implies (to me, anyway) that JavaScript's high performance is its manifest destiny. And that, to me, understates the contribution of V8: the high performance of JavaScript was not a foregone conclusion; it required guts, innovation and hard work. As for the assertion that a trace-based JIT beat V8 to market, a trace-based JIT also has well-known failings, so it's not a terribly meaningful data point -- other than that it speaks to the aspirations perhaps for a higher performing JavaScript. But we clearly agree that Lars and team developed a terrific VM in V8, and that V8 was important in the history of JavaScript -- which was my only point.
Never mind TraceMonkey -- the JavaScriptCore kids rallied in 2008 and did SquirrelFish Extreme, which held its own going into late 2008.
The chosen measure of these new VMs was a set of benchmarks, SunSpider from Apple and the V8 Benchmarks from Google. While V8 had the best GC and most optimizations, on these suites at least, for about two weeks for TraceMonkey (and longer for SFX), V8 was not that far ahead.
You can find the charts via Google still.
V8 had the longest lead time, not just working on what was released with Chrome but trying other approaches first, learning from them, and starting over. That's huge and it has paid off well.
But I don't agree that any architectural failing of one VM counts more than public, reproducible benchmark scores. Even V8 had to do Crankshaft.
Architectures evolve and supersede one another, but the developer and user benefit -- the public benefit -- comes from the competition. V8 was not alone in driving competition.
I wonder what he would have said about TypeScript, it seems that it was launch after this conference (though there is a large chance he was in the loop, no pun intended)
> I wonder what he would have said about TypeScript
Likely nothing. The presentation has a slide on compiled-to-JS languages, and a link to a listing site. Eich doesn't seem to mind languages compiled to JS so far.
