Doesn’t that describe the rise of JavaScript? I haven’t done or seen any specific benchmarks, but I gathered that JS was among the slowest of languages.

Like others have said, major Javascript engines are much than Ruby implementations, owing to the huge effort Firefox, Google, Apple, and Microsoft have put into those engines. In principle, I'm not sure if there's reasons that make one intrinsically harder to optimize.

But even if Javascript were slower, it wouldn't mean the prediction was right. Javascript didn't do something super amazing that let you write beautiful code (the way Rubyists feel about Ruby, or Python folks feel about Python). It just ran the web so we used it. Over time, it's morphed into a language that's about as expressive as any other good dynamic language, also with some warts. But it's not a leap ahead, which is what the original post predicted.

The only language I've felt that way about was Prolog. When it works right, it's amazing. Too bad I can rarely make that happen.

I am convinced there has to be a way to leverage the expressive power of a language like pure prolog without the strain of the 'hints' you need to give it when the depth-first algorithm can't solve the problem. My bet is that if the interpreter 'knows' enough algorithms to solve common calculations, it could leverage these to compute 90% of problems (without needing to have one algorithm for each problem). For the other 10% you would have to supply the interpreter with the necessary algorithms yourself.

JS is pretty much the fastest of the popular dynamic langs, if only because JS interpreters see highly-competitive investment from well-funded companies with armies of developers thanks to the web.

No, it's pretty fast [1].

Chrome and V8 were released in 2008, and Node.js (which leverages V8) in 2009. V8 was already considered fast back then. Mozilla and Apple also focused on optimizing their JS stack in the following years.

[1] I'm not a fan of linking to benchmarks, but that's the best I can do: https://benchmarksgame-team.pages.debian.net/benchmarksgame/... / https://benchmarksgame-team.pages.debian.net/benchmarksgame/...

Google’s marketing for Chrome and V8 was incredibly effective. We’ve basically forgotten that Mozilla, Adobe and Apple were pushing JS perf long before V8 was released. Memory usage of early V8 was atrocious and real world performance was barely improved over Safari and Firefox but Google dominated dev mindshare for a decade.

No, Mozilla and Apple didn't release their JIT engines until after Chrome (though I think at least Apple was working on theirs) and so when Chrome came out JS immediately got about 4x faster, which was a bigger improvement than had happened over the entire previous lifetime of JS. Popular web pages didn't immediately get 4x faster, because they had carefully avoided doing anything in JS if its performance mattered, but JS did. It's true that V8 used a ridiculous amount of RAM, and still does, but its existence made it possible to do things in JS that previously required native code.

As for Adobe, they never did much for JS; they tried to get people to use ActionScript instead, which was a statically-typed language with syntactic similarities to JS.

Apple built the Sunspider benchmarks in 2007, and along with Mozilla, they made dramatic real world performance improvements before Chrome was even publicly announced.

https://webkit.org/blog/152/announcing-sunspider-09/ https://webkit.org/blog/214/introducing-squirrelfish-extreme...

Chrome + V8 was already behind in real world performance, not ahead!

http://www.satine.org/archives/2008/09/19/squirrelfish-extre...

Macromedia (later Adobe) funded a whole bunch of JIT research as the Tamarin Tracing project, which was shared with Firefox as TraceMonkey. The NanoJIT assembly backend was actually shared code.

I guess you're right that TraceMonkey was announced before Chrome was released on 2007-09-02, one day after its inadvertent public announcement. TraceMonkey was announced, though incomplete, more than a week earlier, 2008-08-23: https://brendaneich.com/2008/08/tracemonkey-javascript-light.... SFX, however, was announced later, 2008-09-18.

I don't want to piss too much on SunSpider, but unsurprisingly V8 did better on V8’s benchmarks, while SFX and TM did better on SunSpider, once they were eventually released. That's because V8 was written to do well on its benchmarks, presumably, not because the benchmarks were rigged. Certainly they were in accordance with my experience at the time.

When Chrome was released, V8 was way ahead of the other browser JS engines at JS performance, partly because it had the first JS JIT. But the other browsers took a year or two to catch up, which they had because it also took websites a few years to move most of their functionality into browser JS.

I don't know why you keep emphasizing this “real-world” thing. Are you saying you think nbody and fannkuch are especially realistic benchmarks?

You're right though that Tamarin was a JS engine, not just an AS engine. We regret the error.

The point is Mozilla, Apple and Adobe did enormous work on JS before Chrome was released.

nbody and Fannkuch are meaningless for real world performance on a client side JS app. V8s benchmarks were widely considered to be unrepresentative of what browsers actually do. Sunspider and derivatives were the most realistic tests at the time and Chrome wasn’t a dramatic improvement.

https://johnresig.com/blog/javascript-performance-rundown/

Chrome’s marketing was incredibly successful in getting more people to use a browser better than IE but it didn’t have a dramatic improvement over FF or Safari.

> The point is Mozilla, Apple and Adobe did enormous work on JS before Chrome was released.

That's true relative to the work they had done on, for example, Gopher support, but it's not true relative to the work they did after Chrome was released. As I said, JS performance from 1995 to 2008 hadn't increased by even a factor of 4; then Chrome was released in 2008 and it immediately increased by a factor of 4, more than it had increased in the entire previous 13-year history of JS. It's true that there existed other optimization efforts. But they weren't successful, probably because not nearly enough effort was devoted to them. Tamarin Tracing/TraceMonkey was eventually discarded and is not part of SpiderMonkey today, although a different JIT strategy is. (LuaJIT uses the tracing-JIT strategy very successfully, though.)

> nbody and Fannkuch are meaningless for real world performance on a client side JS app. V8s benchmarks were widely considered to be unrepresentative of what browsers actually do. Sunspider and derivatives were the most realistic tests

I mentioned nbody and fannkuch because they are in SunSpider, so it seems that you are contradicting yourself in addition to, as demonstrated previously, mixing up the historical sequence of how things happened.

I had just written a parser generator and an experimental functional language that compiled to JS when Chrome came out, and the performance improvements I saw were in line with the Chrome benchmarks. My experience is not part of Google's marketing.

Safari perf had already improved more than 4x with the move from AST Interpretation to a register based bytecode VM with SquirrelFish. This happened before Chrome.

I’m working on a tracing JIT for Ruby at the moment partially inspired by LuaJIT. Fingers crossed it’ll be published research this year!

Right, but the Sunspider benchmark also tested strings, regex and date manipulation. V8s original benchmarks didn’t. Performance on stuff like a parser generator was way up, but it didn’t run jQuery based Ajax sites any faster.

Best of luck with that! I look forward to seeing it.

> before Chrome was released on 2007-09-02

This should read, "before Chrome was released on 2008-09-02".

fwiw n-body 2009

10x JavaScript V8 211.76 secs

100x Ruby 1.9 #2 34 min

106x JavaScript TraceMonkey 36 min

http://web.archive.org/web/20090705014239/http://shootout.al...

Real world performance was the opposite story, with Safari's super efficient bytecode VM out performing V8s JIT: http://www.satine.org/archives/2008/09/19/squirrelfish-extre...

No, it really wasn't. You had to use it in really specific ways and it had all sorts of gotchas. One I vaguely remember is that if an array got too big performance would tank like crazy.

Did anybody predict the slow death of jQuery or the rise of giant monster JavaScript frameworks? 2010 is when jQuery was a cult of personality. It was everywhere and any criticism of it was quickly met with desperate mob violence.

Did anybody predict a typical corporate web page would require 3gb of JavaScript code from frameworks and thousands of foreign packages just to write a couple of instructions to screen?

Applications written in JavaScript tend to be hideously slow (and getting worse) but this is more of a community problem; the language itself is now rather fast by dynamic language standards.

JavaScript is now both faster and less productive than Ruby, so it doesn't match his prediction that a slower and more productive language will become popular.

What it does show is that being a web standard trumps everything else.

Raku (formerly Perl 6) basically sits alone as the successor to the slower-and-more-expressive throne. It's a great language to study for the gee-whiz-they-thought-of-everything factor.

But, nobody is paying attention to it. Some of that is because of the Perl 6 baggage, but just as much is probably because this past decade was so heavily entrenched in proliferation of the web-cloud-mobile paradigm that new scripting systems weren't part of the hype cycle. If it didn't get the backing of the FAANGs, it didn't register.

I can imagine a day coming where scripting shines again, though. It might actually be closer than we think. There is always a need for glue code, and glue code benefits from being a kitchen sink of built-in functionality.

We've reached a point where package ecosystem eclipses the importance of language features (which, in brand new languages, typically only offer incremental improvements these days).

It's not good enough to build a slightly better language any more. People won't learn a new language without their favourite packages (or equivalents).

A solution to this is leveraging the ecosystem of another language, like elixir (erlang) and all the jvm languages do.

Julia does this by being able to call Python, R and Fortran libraries.

This has a habit of creating impedance mismatch problems though.

Python is eating the world. In many measures, Python is slower than Ruby.

Python is just a very nice interface to a large amount of optimized C code.

Don't neglect the Fortran code!

I would have agreed with you in 2013. Now it's just seems so fractured and all over the place, like it's had it's peak and is on the way back down.

Sometimes fractured and all over the place is a sign of success. Python is used by a huge variety of different communities for a myriad of use cases, which is pulling it in different directions. That’s not a bad problem to have, as long as it’s managed well and the core webs seem very tuned in to this issue.

Also the community migration to Python 3 is done now. Yes there are massive P2 code bases out there still, but for new projects P3 has been the clear choice for a long time now. It’s over.

python won't last, just like ruby / js / php

I sense a strong convergence between all of them (builtin DS, linguistic traits, bits of static typing)

JavaScript will last. There's no question there at all.