Optimizing forEach: Not so fast!

Previously I wrote about the idea of compiling JavaScript forEach function calls into equivalent for-statements that perform faster.

Now I've been playing around with my small prototype of such a compiler, and after a few initial promising successes, the complexity of the whole problem is really starting to hit me.

Lets take a simple example:

var sum = 0;
array.forEach(function(x) {
    sum += x;
});

This kind of code is really trivial to convert, and my own little script produces something like that:

var sum = 0;
for (var $i1 = 0, $len1 = array.length; $i1 < $len1; $i1++) {
    var x = array[$i1];
    sum += x;
}

Great, but let's take this a bit further, say, a nested loop over 2D array:

var sum = 0;
array.forEach(function(x) {
    x.forEach(function(y) {
        sum += y;
    });
});

That's also easy to handle. We just have to apply the optimizer recursively to the loop body. Here's the result similar to what my script gives:

var sum = 0;
for (var $i1 = 0, $len1 = array.length; $i1 < $len1; $i1++) {
    var x = array[$i1];
    for (var $i2 = 0, $len2 = x.length; $i2 < $len2; $i2++) {
        var y = x[$i2];
        sum += y;
    }
}

Again, everything is quite correct. Maybe a few extra variables, but correct nevertheless. But now let's modify the forEach version a bit:

var sum = 0;
array.forEach(function(x) {
    x.forEach(function(x) {
        sum += x;
    });
});

Now the inner and outer loop variables have a same name. And it works out fine, because the inner x nicely shadows the outer x. But when we run this through my simple-minded compiler:

var sum = 0;
for (var $i1 = 0, $len1 = array.length; $i1 < $len1; $i1++) {
    var x = array[$i1];
    for (var $i2 = 0, $len2 = x.length; $i2 < $len2; $i2++) {
        var x = x[$i2];
        sum += x;
    }
}

Boom! The shadowing closure is gone and it all fails spectacularly.

So yeah... this rabbit hole seems to be much deeper that I initially thought. Simply picking unlikely names for generated variables isn't enough to avoid conflicts.

Then again, simply renaming the inner conflicting variable should solve the problem in this case. One just needs to keep track of all the variables visible in particular function.

It still feels like... how hard can it be?