LightTable Arch

Posted by moodyharsh on 2016-06-30

LightTable follows Behavior Object model(BOT), which itself is inspired from
Entity Systems model(ES).

In a typical OOP + GUI program we abuse the Observer Pattern and harcode most of the design
and Event types.

With BOT/ES the design is stored as an in-memory data structure,
the event-listeners(behaviors) are stored as metadata and
the events themselves are generated on the fly.

Every datum is identified by a unique ID and is typically struct-ish.

This means, the entire program can be extended runtime.

In LT,

  • object/create makes a new datum. Each datum stores behavior names.
  • object/raise triggers events with payload.
  • behavior macro stores “event(s) <-> reaction” mapping. Reaction receives the payload.
  • object/update, object/merge manages the in-memory data structure.

LightTable is built on node.js so, most of the actual work is done by the node.js library wrappers in reactions.
The ordering of the events, is mostly on a first-come basis. Think list iterations.

An alternate way of looking at this model is imagining a Relational Database + Stored Procedures.

Can this model be applied for client side applications, as an alternate to Backbone.js ?

Consider a TODO list,

Name – Data Structure :: [Behaviors]

item – Hash :: [view, add, remove, update, list]

Event – Behaviors :: Notes

  • $() – list
  • - list
  • add-button-click + data payload – add :: add, after object/create triggers
  • remove-button-click + id palyload – remove :: remove, after object/remove triggers “
  • save-button-click + add payload – update
  • todo-clicked – view

What about GUI ?
By dividing the behaviors carefully, GUI can be separated from the NON-GUI.
LT uses hiccup to generate HTML, making GUI another runtime convinience.