async - Asynchronous programming in HaXe made easy.

RC version

The aim of the project is to use the power of HaXe macro system to make complex asynchronous operations easier to write and read.

We get seemingly synchronous code with special markings and convert it to asynchronous code. Synchronous code uses standard callback system like in node.js(and is interoperable with it, and other asynchronous functions written in this style).

Example (look at /test/Test.hx for more)

Class should implement async.Build, compiling should be done with -lib async.

In the first function we think of everything like it is synchronous, but most parts are not. Later functions are just for the sake of completeness.

class Test implements async.Build{
  @async static function foo():Many<Int, String, Dynamic>{
    //returned types are optional
    return many(222, 'string', null);
  }

  @async delay(ms:Int):Void{
    //use Void for functions which return nothing(or error)
    //...
    return;
  }

  @async static function bar(){
    //when there
    var b;
    [var a:Int, b, _] = foo(); //getting multiple values, using _ to ignore some of them
    [] = delay(100); //just waiting for callback
    trace(a);
    [
      [var c, var d, _] = foo(),
      delay(200),
    ]; // this call will be ran in parallel, so the trace will be executed at the moment both return(call back)
    trace(c, d);
  }
}

Features / Done

• Every class implementing async.Build interface will be automatically processed.
• Every function of such class with @async meta will be converted.
• write [<returns>] = call(<args>) to call asynchronous function

[a, var b, var c:Int] = call(1, 2, 3); <other code> is converted to something like

call(1, 2, 3, function(err, _a, b, c:Int){
  if(err == null){
    a = _a;
    <other code>
  }
  else{ cb(err, null); }
});

• if you put a couple of such calls into array, it will be executed in parallel

[
  [var a] = getA(),
  [var b] = getB(),
];

• for(<cond>){<code>}, if(<cond>){<code>}, if(<cond>){<code>}else{<code>}, do{<code>}while(<cond>), while(<cond>){<code>}, switch(...){case ...: <code>}

will be converted to just what you expect, but you may put asynchronous calls only into <code> blocks, not into <cond> blocks, continue and break inside loops will work.

- `throw` will be transformed to callback with error
- `return` will be converted to callback with value
- `return many(...)` will be converted to callback with multiple values

- `try{<trycode>}catch(...){<code>}` will be converted.

If there are asynchronous calls inside, it will catch asynchronous errors but not synchronous, but if there are only synchronous calls, synchronous errors will be caught.

• Mission control: if there is no return in your code - it will be implicitly added.

  All in all, there is no way @async function will never call it's callback or call it more than one time if:

  • all the asynchronous function it uses always call their callbacks, do it only once and never throw (synchronous) errors.
  • there is no implicit calls to callback

• Mode which will enrich errors with stacktrace-like information entroduced, but requires a bit more of work.

• Other

  Use -D async_readable compilation flag to make async code more readable.

Function can be implicitly converted by passing is to async.Async.it() macro.

Block of code can be converted to anonymous function of type (Dynamic->Void)->Void by using async.Async.block() macro.

TODOs

• We can allow optional arguments by setting callback as first argument(it isn't usable in pure code, but we dont care about this in generated code).
• Source mapping may be a bit broken in some cases.
• Some code can be simplified, amount of calls reduced (functions which just check error and call next function which will also check for same errors).

by Yanis Benson