anmonteiro Code ramblings

Om Next internals: Incremental Rendering

The phrase "incremental rendering" is frequently thrown around in the context of Om Next's design. This post is an attempt to clarify what the concept of incremental rendering is all about, what it does for your Om Next applications and how you can take advantage of its properties.

First things first

Some of the properties behind Om Next's design make it possible to perform a number of optimizations under the hood. These optimizations are mostly driven by the fact that components are annotated with query expressions and identity. One example of such optimizations is behind the solution for issue #556: transacting large values or reading large data sets is not slow anymore 1.

Incremental rendering is another such optimization in Om Next, based on its knowledge about an application's data requirements. In plain React, re-rendering always initiates at the root component, whether its children (in the tree) end up updating or not. This means that each node in the component tree gets asked if it needs to update — via the shouldComponentUpdate lifecycle method. Om Next, on the other hand, knows exactly what data components need, which means it can always start diffing their props at the root of the concrete subtrees that need the data related to transactions. The next image compares React and Om Next's updating phases after a transaction that originated at the red node.

React rendering
React's update phase
Om Next rendering
Om Next's update phase

How can this work?

Astute readers might now be asking themselves — but how can incremental rendering be made to work, if the Om Next parser dispatches on the top-level keys in a query? Well, that's the entire purpose of path metadata and om.next/full-query. When updating a subtree, Om Next does two things. Firstly, it calls the parser with the full-query of the component rooted at that subtree — which is really just a fancy way of saying "give me a query starting at the root of my application, but narrowly focused at the current component's query". It then extracts the data relevant to the component that is updating, which is located at its data path. Cool stuff, right? Even better, this behavior is extensible through the :ui->props key in the reconciler!

If the above sounded too dense, here's a practical example:

  • Say we have an application that shows 2 tabs. We might have the components below:
(defui TabInfo
  static om/IQuery
  (query [this]
    [:info/id :info/name :info/items]))

(defui Tab
  static om/IQuery
  (query [this]
    [:tab/title {:tab/info (om/get-query TabInfo)}]))

(defui Root
  static om/IQuery
  (query [this]
    [{:tab1 (om/get-query Tab)}
     {:tab2 (om/get-query Tab)}]))
  • Now let's imagine that the TabInfo component in tab #2 has performed a transaction that adds more items to its :info/items list. Check out the differences below between the application's root query and the full-query of the transacting component:
;; Application's root query
[{:tab1 [:tab/title
         {:tab/info [:info/id :info/name :info/items]}]}
 {:tab2 [:tab/title
         {:tab/info [:info/id :info/name :info/items]}]}]

;; `om.next/full-query` of tab #2's `TabInfo`
[{:tab2 [{:tab/info [:info/id :info/name :info/items]}]}]

As you can see, the full-query is narrowly focused at the specific data requirements that TabInfo declares. Two beneficial consequences follow: for one thing, it allows the parser to dispatch on the same key as the root query; additionally, it makes om.next/db->tree's data denormalization perform faster, as we're only interested in a specific subset of the data when compared to the application's root query.

  • What about the "data path" mentioned above? What is that?

The data path of a given component is simply a vector of keywords that describe how we can get to the query of a given component starting from the query of an applications' root component. In our example above, the path of tab #2's TabInfo would be [:tab2 :tab/info]. This means that Om Next can simply use get-in in the result of parsing and pass the correct data to the updating component(s).

Wrapping up

I hope this article helps you understand some of the inner workings that make Om Next awesome :-) In a next post, I'll talk about how we can optimize incremental rendering even further with "path optimization".

Stay tuned, and thanks for reading!


1 Om Next's props are annotated with their "path" for various internal purposes. Before, we naively walked the data structure that resulted from parsing. Now we follow the data along the query path and only annotate the matches. ↩