Skip to content

HTTPS clone URL

Subversion checkout URL

You can clone with
or
.
Download ZIP
Fetching contributors…

Cannot retrieve contributors at this time

221 lines (148 sloc) 10.948 kb

I get a weird "Function contains async calls but does not have _ parameter" error. What's the deal?

You'll get this error with code like:

function foo() { bar(_); }

Asynchronism is contagious. Function foo contains an async call to bar. So foo itself becomes asynchronous and needs an _ in its parameter list. Change it to:

function foo(_) { bar(_); }

Can I create an anonymous streamline function?

Yes:

function(_) { bar(_); }

What if my async function does not do any async calls?

function foo(_) { return bar(); }

Nothing bad will happen. But you have to call foo as foo(_). If you call it as foo() you'll get a future back, probably not what you want.

There will be a bit of overhead so you should avoid declaring sync functions with an _.

But you many need this feature, for example if you dispatch to functions that may be either async or sync. In this case you should dispatch asynchronously (pass an _) and declare all of dispatch handlers as async functions, even those that are synchronous.

Why do the built-in streamline functions (array.forEach_, map_, etc.) have _ as first parameter rather than last?

Because it makes it easier to deal with optional parameters.

It also make it easier to check that the _ is passed to every async call. You don't have to look at the end of the parameter list.

The standard node convention is to have the callback as last parameter. You should follow this convention if you design libraries for a general node audience. But you may choose to pass the callback as first parameter if you design private APIs or if you target a more restricted audience of streamliners.

Can a streamline function take optional parameters?

Yes but you have to be careful with the special arguments variable. You cannot pass it blindly to another call with apply. You have to use the streamline apply_ built-in function instead. See its documentation.

It does not work and I'm not even getting an exception. What's going on?

You probably called a buggy asynchronous function and you did not pass _. For example:

function buggy(_) { undefined.toString(); }
buggy();

The problem is that when you call buggy() without _ it returns a future. The future memorizes the exception but does not throw it. Try the following:

function buggy(_) { undefined.toString(); }
var f = buggy();
console.log("after buggy()"); // you'll see this one
f(_); // throws the exception
console.log("after f(_)"); // you won't see this one

I'm calling path.exists(fname, _) and it does not work. What am I doing wrong?

Streamline is designed to work with functions that take standard node callbacks, i.e. callbacks that have the standard cb(err[, result]) signature.

Unfortunately some libraries use a different callback signature, and node itself has a few exceptions, the most notable one being path.exists which does not have any err parameter in its callback.

The workaround is to write a small wrapper that re-aligns the callback parameters on the standard:

function exists(fname, cb) { 
  path.exists(fname, function(result) { cb(null, result); });
}

console.log(__filename + ': ' + exists(__filename, _)); // works
console.log(__filename + ': ' + path.exists(__filename, _)); // does not work

I'm calling an async function without _ and I'm not getting a future back. What's wrong?

You're calling a function which was not written with streamline, for example one of node's fs function. The workaround is easy: just wrap it with a streamline function:

// the wrappers
function readTextFile(path, enc, _) { return fs.readFile(path, enc, _); }
function readBinaryFile(path, _) { return fs.readFile(path, _); }

// testing futures
var f1 = readTextFile(path1, "utf8"); // ok
var f2 = fs.readFile(path2, "utf8"); // does not fail but f2 is undefined
var data1 = f1(_); // ok: f1 is a future
var data2 = f2(_); // fails!

Wrapping functions with optional arguments might be a bit tricky. The following wrapper will work though:

function readFile(path, enc, _) {
    if (typeof enc === 'string') return fs.readFile(path, enc, _);
    else return fs.readFile(path, _);
}

My flow control is completely broken. I am completely lost. Help!

Check your file extensions.

If you put streamline code in a .js or .coffee file it won't get transformed and your code will go wild: callbacks will be executed multiple times, etc.

Can I use the streamline extensions for files that don't contain any async code (yet).

This won't hurt. Streamline only transforms functions that have the special _ parameter. So, files that don't contain async code won't be impacted by the transformation.

The only drawback is a slower application startup because more files get transformed but you can avoid that with the --cache option.

The underscore trick is designed for callbacks but not events. How do I deal with events?

If you are dealing with stream events, you should try streamline's stream API. It wraps node streams with a simple callback oriented API and it takes care of the low level event handling for you (pause/resume on readable streams, drain on writable streams). For example:

var streams = require('streamline/lib/streams');

var inStream = new streams.ReadableStream(nodeInStream);
var head = inStream.read(_, 128); // read the first 128 bytes
var chunk;
while (chunk = inStream.read(_)) {
  // do something with chunk
}

var outStream = new streams.WritableStream(nodeOutStream);
outStream.write(_, result);

This module also contains wrappers around node's Http and Net objects, both client and server. See the streams documentation for details.

If you are not dealing with stream events, you can take a look at the implementation of the streams module for ideas. Any event API can be turned into a callback API (with a getEvent(_) call that you would call in a loop) but this can be totally counterprodutive (events will be serialized).

If the events are loosely correlated, it is better to let them be dispatched as events. But in this case, you may want to use streamline to handle the logic of each event you subscribed to. This is not too difficult: just put a small anonymous function wrapper inside your event handlers:

function handleError(err) {
    // log it somewhere
}

server.on('error', handleError);
server.on('eventA', function(arg) {
    (function(_) {
        // function has an _ parameter, you can use streamline
    })(handleError);
});
server.on('eventB', function(arg) {
    (function(_) {
        // streamline code...
    })(handleError);    
});

If the event handler takes a fixed number of parameters, there is even a lighter solution: just add an _ parameter to each event handler:

server.on('eventA', function(arg, _) {
    // function has an _ parameter, you can use streamline
}).on('eventB', function(arg, _) {
    // streamline code
});

It works because function(arg, _) will be called as a future (without _). But it is a bit more fragile because errors are not trapped and code would break if the server changes and starts passing more arguments to its event handlers.

Are there limitations? Am I limited to a subset of Javascript?

Hardly any. Streamline knows how to transform all Javascript constructs except two:

  • labelled break and continue.
  • non-empty switch case that falls into another case without break nor return.

The transformation engine could be improved to handle these constructs too but they are rather hairy and workarounds are easy so they haven't been implemented yet. You'll get a compilation error if you use these constructs.

On the other hand, you can do all sorts of crazy things, like calling async functions from object or array literals, or even writing async constructors. The following will work:

var foo = [f1(_), f2(_), f3(_)].filter_(_, function(_, elt) { return elt.g1(_) || elt.g2(_); });

function Bar(_, name) { this.name = name; baz(_); }
var bar = new Bar(_, "zoo");

Will I always get the same semantics as in normal (sync) Javascript?

The streamline compiler works by applying patterns. These patterns have been carefully crafted to preserve semantics. The only known case where streamline may diverge is the order of evaluation of subexpressions inside a given statement.

In callbacks mode, streamline evaluates the asynchronous subexpressions before the synchronous ones. So if you have foo() + bar(_), it will evaluate bar(_) before foo().

In fibers mode, streamline preserves the order and evaluates foo() first. So you should not write fragile code that relies on precise order of evaluation of subexpressions.

But streamline guarantees the ordering in the cases where it really matters: logical operators (&& and ||), ternary operator (cond ? a : b) and comma operator (a, b, c). If you write foo() && bar(_), foo() will be evaluated first and bar(_) will only be evaluated if foo() is true.

What about performance? Am I taking a hit?

In callback mode, streamline generates callbacks that are very similar to the ones you would be writing by hand. So you are only paying a small overhead. Usually, the overhead will be small in comparison to the time spent in the async functions that you are calling. For example, you incur a 50% overhead when calling process.nextTick(_), which is the fastest async call in node.js. If you call setTimeout(_, 0) the overhead drops to 18%. And on a real (but simple) I/O call like fs.stat it goes down to 3 or 4%.

The fibers mode has more overhead on I/O calls but it eliminates all the callback overhead in the layers that call low level I/O services. So depending on the thickness of the logic that sits on top of the I/O layers you may get an increase or decrease of performance. The nice thing is that you don't need to choose between callbacks and fibers upfront. You can write your code, compare performance and then choose the best mode for deployment.

Some patterns like caching can give surprising results (see https://gist.github.com/2362015).

There is also room for improvement. In callback mode the small overhead comes from the additional comfort and security that streamline gives you: sync stack traces, global context, trampoline, rigorous exception handling. This could be improved with options that disable these comfort features but it would make the tool more complex.

Future versions of V8 will likely support harmony generators. Streamline could then provide a third transformation mode that takes advantage of this language feature and may benefit from additional V8 optimizations. But all this is still speculative at this stage.

Jump to Line
Something went wrong with that request. Please try again.