>If using threads means the language is asynchronous to you,
Goroutines are not threads. It's named after a "coroutine," which is an async function (like a generator) where you can pause execution of a task (to wait for IO, for instance) and then resume it at a later time. Coroutines give you implicit async/await style programming, without having to use extra keywords. The language Lua supports coroutines natively, for instance, including explicit yields to use them as generators, if you need. Goroutines give you coroutines with a bonus: Actual thread hopping (and multiple channels of communication possible, so you can effectively have more than one "yield" channel).
If you have 400,000 Goroutines running on 4 CPUs, your Goroutines can task switch between those 4 CPUs using 4 OS threads (or 8, or whatever number you determine is best for your app) as their IO events queue up. It is async, and each Goroutine has around 10k of RAM overhead, not counting data you're storing yourself (it varies by architecture, but that's a reasonable estimate; I've seen between 6k and 16k on different architectures).
Go is better async than JavaScript, because a Goroutine started on one thread can get processed on another, so a few thread-hogs won't block a task. A single Goroutine could cycle between all your worker threads, in fact, though I think it tends to stick with a single worker thread ("thread affinity"). It's probably the most asynchronous you can get in a language (only the Erlang VM/Elixir is in the same class of language, as far as I know).
Do you think PHP could handle 400,000 concurrent WebSocket connections on one server? Go can. How much RAM would you need to handle 400,000 CPU threads? Maybe 64Gb instead of ~8Gb in Go? Wouldn't CPU switching alone be starving most of the threads and redlining the CPU at that point? In my experience you start seeing serious slowdown around 5,000 threads. I'd guess you'd need 50x as many servers to handle that many connections on PHP. Maybe 100x. You need to support a million users? Is it better to have 3 servers or 300?
> The things you complained about in another language were fixed/removed years ago but you say it's still just as bad.
PHP is architecturally single-thread-per-request. Full stop. Even PHP 7 and HHVM. That simply doesn't scale as well as async, as I described above. The complaints I made were why I bailed on it years ago. I'm sure PHP has gotten a lot better than it was when I used it, but given that it still misses the mark architecturally, I have zero motivation to give it a second chance.
But given that NodeJS has far surpassed it in community support (as well as by many performance measures -- HHVM is quite fast at raw processing speed now, though not as high throughput because of the lack of true async), why is PHP still relevant except to continue to maintain existing code bases? (I've worked with several of those as well -- Joomla and Drupal in particular -- and neither was something I'd consider worth using, having looked at their internals and terrible performance. Joomla was a complete disaster; Drupal only slightly better.)
>Every failure or less than great situation in your favoured languages has just been fixed in the last few years.
Your point? Even if they all were only better as of yesterday, they're still better. Or are you defending your choice of a few years ago?
And actually, Promises have been available using polyfills for years. The proposal was made at least six years ago; I can't find the exact date, but I found a reference added to the CommonJS wiki in 2010 [1]. The Q library dates back to 2010 as well. [2]
I concede that there may have been a period of time where PHP was better than JavaScript, by the criteria I'm using today, because PHP did improve a lot after I first encountered it, while JavaScript only more recently developed its compelling advantages. It looks like Node was released in 2009? In 2009 I was using OpenResty (Nginx+Lua) to do my (small amount of) server work, which can still be more performant than both Node and PHP. But NodeJS has shot past in terms of community support, which is critical, TypeScript is awesome, sharing code on client and server is awesome, isomorphic rendering is awesome, and performance is good enough (compared to Lua) that NodeJS just wins for me, big time. Except when I need the extra speed, and for that I use Go.
Go is a brand new language, relatively speaking. So of course they're still making major improvements. Performance has already surpassed HHVM, even for raw compute tasks. It's a better fundamental architecture, giving it an edge, and performance will likely improve still more, though they're already hitting diminishing returns: Some of the worst case performance they see now may get 2x-4x faster, but typical app performance is probably only 10-20% short of optimal. I mean, they're already beating C++ on the server. How much better can they go?
Use whatever language you want; if your background and/or current job is PHP, so be it. It's not as bad as it was 10 years ago, for certain. Just don't expect anyone to pick it up based on its merits. There are too many other, stronger options available today.
[1] http://wiki.commonjs.org/wiki/Promises/A
[2] https://github.com/kriskowal/q/graphs/contributors