dosync Archive Pages Categories Tags

Transit-js Caching

19 February 2015

One of the nicest features of Transit and thus transit-js is the standard caching mechanism. Query results often contain an incredible amount of data duplication in the map keys and Transit largely eliminates this problem for you.

Periodically users have requested that the caching mechanism be extended beyond just map keys to richer data structures. It turns out this feature has existed within Transit in plain sight since its initial release. As usual, better data structures save the day, the presence of ES6-like Maps in transit-js makes the solution nearly trivial ...

Imagine you have a type Point that looks like the following:

var transit = require("transit-js");

var Point = function(x, y) {
    this.x = x;
    this.y = y;

You write your PointHandler and construct a Transit writer and write out the following bit of data:

var write = transit.writer("json", {
        Point, (new PointHandler())        

var p = new Point(1.5,2.5);

console.log(writer.write([p, p, p]));

What will we get in return?


The keys were cached, but obviously we're not successfully eliminating the real data duplication here. If you try to serialize something less trivial like application state you'll quickly find yourself in trouble.

Custom Write Handlers With Caching

We need to duplicate the idea behind Transit's key caching. Key caching works because the traversal order for reading and writing is precisely the same. With this knowledge we can build a caching system on top of Transit with the help of some Maps!

Examine our new caching PointHandler:

var PointHandler = function(cache) {
    this.cache = cache;
PointHandler.prototype.tag = function(v, h) {
    if(this.cache.toId.get(v)) {
        return "cache";
    } else {
        return "point";
PointHandler.prototype.rep = function(v, h) {
    var id = this.cache.toId.get(v)
    if(!id) {
        this.cache.toId.set(v, this.cache.curId++);
        return [v.x, v.y];
    } else {
        return id;

We construct the PointHandler with a shared cache object. This cache object has two properties toId, a transit-js Map from an arbitrary object to its cache id, and curId the current cache id.

When we go to get its representation if it's the first time we've seen it we add it to the cache and return the full representation. Because we're using real Maps we're not restricted to string based keys. We can easily and efficiently map objects to integers.

Of course if we have seen the object before we simply return its integer id.

That's it! Now let's write a cachingWrite function:

function cachingWrite(obj) {
    var cache = {
            curId: 1
        writer = transit.writer("json", {
                Point, (new PointHandler(cache))

    return writer.write(obj);

Let's try it out:


What do we get?


This should put a smile on your face. The second time we see a cached value we emit the "cache" tagged value. The third time we emit, the standard Transit key caching kicks in and we get a very succinct representation.

What about reading this back out?

Custom Read Handler With Caching

Writing cachingRead is quite a bit simpler than cachingWrite:

function cachingRead(obj) {
    var cache = {
            curId: 1
        reader = transit.reader("json", {
            "handlers": {
                "point": function (v, h) {
                    var ret = new Point(v[0], v[1]);
                    cache.fromId.set(cache.curId++, ret);
                    return ret;
                "cache": function (v, h) {
                    return cache.fromId.get(v);


Let's roundtrip!


And we'll get the following result:

[{x: 1.5, y: 2.5 , {x: 1.5, y: 2.5}, {x: 1.5, y: 2.5}]

Perhaps you're not convinced these are all the same object in memory. Let's do a quick check:

var equal = function(x, y) {
    return !x ? false : x===y && y;


We get back a non-false value:

{x: 1.5, y: 2.5}

Which means all the values in the array were exactly the same object.

Go forth and serialize your application states into efficiently packed JSON values!