> At the end of the day it's all JavaScript so performance is rather moot point
That's probably true, but C++ does provide very powerful compile-time evaluation mechanisms like templates, constexpr (compile-time evaluated constants and functions), and template metaprogramming. That being said, compile-time C++ can look very gross since most of the features are accidental.
However, there are a lot of huge performance benefits to moving computation to compile time: selecting the most optimal algorithm for a type, guaranteeing you don't mismatch your units, precompiling search (glob, regex) engines, eliminating dead code, and so on.
So I would be shocked if at least some javascript applications couldn't be much faster and more correct by writing in C++ first and compiling to js. That being said, it's probably not in scope for a minimum-viable product, so improving already business-critical parts of the code base might be the best place to start.
Metaprogramming and type-safety facilities of C++ are very limited compared to other languages like Haskell or Scala. And for the latter there is a JS compiler as well.
For me it's portability first, decent performance second and third the 'seamless distribution model' of the web (no installation, no app shops, no code signing, just an URL).
I wouldn't write a web-only app in C++, but if the same code needs to run on other platforms as well (e.g. native mobile, desktop or gaming consoles) then C/C++ is basically the only choice. Also, at least in emscripten, compiled code is faster then hand-written JS for several reasons (asm.js, LLVM optimizer passes, no gc), so it makes sense to write stuff like physics/pathfinding/AI engines in C/C++ and compile it to JS.
Plus, it's fun to write a desktop OpenGL demo, flip a build system switch and run it in the browser ;)
According to the Chrome guys...it isn't a thing [0][1], and I tend to agree. They've kept up, and often exceeded FF, IMO, for real code by making a better JIT compiler. They still haven't implemented asm.js and yet their "naive" approach to asm.js style code is on par.
The links you pasted show Firefox currently being 3x faster than Chrome when running asm.js code, so I'm not sure what your point is. V8 is good at generating fast code without the hints, but obviously the extra restrictions help the optimizer. Even Google admits this when they try to promote Dart.
asm.js does little for you if you write "real JS", but this thread is about cross-compiling other languages, which clearly has real world use.
(What a great post, thanks for linking. Even though I follow mraleph on twitter, I had never opened his website. Probably because it all looks like scary compiler stuff.)
Another major benefit (of emscripten too) is that is you have significant non-presentation client-side logic, it can be shared between iOS, Android, and the Web.
For example take something like Dropbox's Carousel architecture, which shares the client side cache, date model, and sync client between iOS and Android via a C++ library. Something like Cheerp or Emscripten allows you to also use this same library on the web.
That's actually not true. Emscripten compiles to code that is much faster than hand-written JavaScript. In this case they are not using Emscripten or a similar technique but are compiling to regular old JavaScript, so performance probably is the same, rather this might be for people who just like C++, like what GWT is for Java devs.
You misunderstood. People mostly write with C++ when performance is critical. Why? C++ offers deterministic memory management, Specialize to CPU instruction set and better use of CPU caches etc. Those performance benefits would not translate when you cross compile to something like JavaScript.
I get the point about converting the legacy desktop apps to web apps, but how would it handle the architectural differences ( Desktop apps don't scale)
> People mostly write with C++ when performance is critical.
Performance is a big reason people use C++, but there are many others. For example, people also write C++ when correctness is critical (bank software, safety-critical systems, etc.).
Writing for correctness in a language with no memory-safety, weak type system and, till not very long ago, no standardized memory-model, and UB in every other paragraph of the specs? Good joke. If they are really doing this, they have no idea what they are doing. There are plenty of languages better for safety-critical systems than C++; even widely hated, boring Java is much better.
It's a bit off topic, but I'm being descriptive (based on professional experience) and not normative when I say C++ (and C and Ada) are used for safety-critical software. Javascript and functional languages certainly aren't.
The vagaries of the standard aren't issues since safety critical software is validated on a per-platform and per-compiler basis. Memory safety is certainly a concern during the development process but more importantly (!) GC languages (like javascript and Scala) do not have provably (for some value of provably) deterministic execution times. Determinism is also a problem for lazy languages (like Haskell).
I say that to say the zero-cost abstractions of C++ is a huge benefit when writing safety-critical software. The level of control over emitted binaries is essential and fairly rare in programming languages.
> If they are really doing this, they have no idea what they are doing.
Well, I can't speak for entire industries, but that's a pretty sweeping generalization. You might be surprised what the challenges are when writing safety-critical systems software. That being said, I'm sure large portions of the industry are ripe for innovation.
> GC languages (like javascript and Scala) do not have provably (for some value of provably) deterministic execution times
This is not unique to GC languages, but any languages with dynamic memory management. C++ new/free and STL abstractions built on top are not provably deterministic either. And if you program without ever touching dynamic memory (statically allocating and pooling everything) then GC is not a concern.
> I say that to say the zero-cost abstractions of C++ is a huge benefit when writing safety-critical software.
You're talking about performance now, not correctness. All those benefits get lost once translated to JS.
I was replying to the statement: "For example, people also write C++ when correctness is critical". And I say if someone writes C++ in a correctness critical system, he must not know what he is doing. Even if you're using only STL, RAII and following all the good coding practices, there are still so many ways to screw things up, that C++ is among the worst choices in this regard.
NASA and military are using a whole lot of different technologies and languages, including, but not limited to assembly, C, C++, ADA, Haskell, Coq, Python and Java. Saying they use C++ for correctness, when they are using Coq or Haskell as well, is again - exaggeration. I believe they use C/C++ more for performance / low memory overhead rather than its correctness related features.
At least when compiled with emscripten, the same memory-access-related optimizations used in native code also apply to cross-compiled JS code since the memory layout is exactly the same. emscripten uses a big linear memory buffer (a single big JS typed array) as heap and dlmalloc for dynamic memory management within this heap. If your code is cache-friendly in the natively compiled version of the code it will also be cache friendly in emscripten compiled code. Cheerp seemt to use a different approach though and seems to generate 'traditional' high level JS objects (which also means it requires garbage collection, which emscripten generated code doesn't).
You do get some of those performance benefits when you cross-compile. JS VMs have different paths for different types of code and the type of code that Emscripten produces goes through a faster path because more is known about the code ahead of time.
You also get 1.4MB of code, which takes 1 hour to load on a mobile connection, and reimplements functionality that host JS/Browser provides which goes much quicker ;) Most C++ code does not care about compiled size(especially with templates!!!), but JS code often does, so is written in a smaller way.
True, you need to be sensible about compiled code size which traditionally wasn't important for C/C++ code, and it's also true that the C++ std lib or careless use oft templates can bloat the code size. On the other hand, three.js which is basically the standard lib for 3D rendering on the web comes at 0.5 MB minified, which isn't exactly small either ;)
