Saturday, August 04, 2007

callbacks within callbacks

Oh, the travails of an Ajax the beginning the requirement was straightforward. Grab some data asynchronously by passing along an "update-display" callback function (a function with a 'data' parameter). The callback function could even be anonymous, or refer to variables in enclosing function scopes at the time of definition--nothing out of the ordinary.

Then someone has the nerve to change his mind about the interface, so now one display needs to show what is actually the patched-together result of several invocations of the same data call, all at once. At first glance, this is still easy--just fire off all the asynchronous requests required and fuhgeddaboutit, because each one has a callback to update only its part of the display.

Wrong! In this case, the parts of the display are ordered and the necessary extra data calls to fill in "missing" points aren't known until some of the data has been gathered, meaning the part of the display generated from call C must be finished after the part of the display generated from call B is finished, and so on. But all the calls are asynchronous, so there is no guarantee whatsoever of the time or order in which individual calls will return.

Flashbacks of semaphores and mutexes and Dining Philosophers dance before your eyes. No--that way lies madness. Since the display is ordered, better to just get all the data pieces, assemble them, then display the whole as if it had been from a single data call. Unfortunately, the "update-display" function can't immediately run after firing off the asynchronous calls, of course, because then it might run before all the data was gathered! So the "update-display" function must be in a callback of some kind. Yet if the calls are asynchronous, which callback should contain the overall "update-display" function? The last one to finish, of course. But which one will that be?

Puzzle it till your puzzler is sore, and one answer, not necessarily the best, may emerge. The time spent studying the techniques of functional programming may generate (yield?) a payoff. Consider the following almost-real code, in which a callback function takes its own function parameter.

function baseCallback(data,update-display-function) {
var afterFunction = function(dataparm) {
update-display-function(dataparm); };
function addAnotherAfterFunction(currentAfterFunction,
patching-info) {
return function(dataparm) {
var innerCallback = function(innerData) {
// process data...
// patch data using patching-info
// (data patch location passed by reference)
// process data in a loop, and within the loop,
// when there's a need to make another data call...
afterFunction =
// after the data processing loop
function dataCallInvoke(originalParams,upd-disp-fn) {
var normalCallback = function(dataparam) {

The idea is that addAnotherAfterFunction repeatedly operates, as needed, on the function in the afterFunction variable. One callback becomes embedded inside another callback inside another callback. When the outermost callback (the final contents of the afterFunction variable) runs, it makes a data call and passes in the next-inner callback. The data call runs, then executes the next-inner callback that was passed to it. The next-inner callback perhaps makes its own data call...and so on, until the "true original" callback, the "update-display" function, takes the fully-constructed data and does what it will.

One downside is that the invocations of the data calls can't run in parallel (although since order explicitly matters so much in this situation, parallel execution won't work anyway). Also, tossing around so many functions with their own individual closed-over scopes surely eats up some memory (although I wonder if some "copy-on-write" implementation could be a workaround). Honestly, I'm not sure what terminology describes this technique. Um, mumble, mumble, continuation-passing-style, mumble, combinator, mumble, monadic bind, er...

No comments:

Post a Comment