Bricks not monoliths

06 Feb 2013

Chapter 32 of Tao Te Programming advises you to make bricks instead of monoliths.  Here is an example. The example is written with the syntax of R and is a data analysis, but the principle is valid no matter what language you use or what your task is.


Here is an outline of a function  reminiscent of many novice attempts:

monolith <-
function (data, col="blue", pch=21) 
        # transform data
        # fit model to data
        # plot data, uses 'col' and 'pch'
        # get desired model results
        # return desired model results

Each of these comment lines may be many lines of code so that the whole function runs to pages. On the positive side, this is a step towards abstraction from merely having a script of commands and changing values by hand within the script.


It is not that I’m against having one function that does everything.  What I’m against is all the work being done in that one function.  Here is an alternative:

brickwall <-
function (data, ...) 
        tranData <- myDataTransform(data)
        dataMod <- myModelFit(tranData)
        myPlot(dataMod, ...)
        myModelOutput(dataMod) # this value is returned

brickwall and monolith can do exactly the same things, but they do them differently.  In the brickwall setting we have written 5 functions (at least) instead of 1. There is an R-specific difference in the two functions.  brickwall uses R’s three-dots mechanism to make changes to the plots flexible and convenient.  On the other hand, monolith tries to hardcode all of the plotting arguments that the user might want to use.


There are several ways that the modular approach helps:

  • easier to understand
  • easier to debug
  • easier to change
  • code is more reusable

Easy to understand

It is easy to comprehend what brickwall is doing.  If we saw an actual implementation of monolith, it would (almost surely) not be at all expressive of what it was really doing.  If we are to understand something, it has to be simple. Additionally, when you break a problem into pieces, you are likely to think more clearly about it.

Easy to debug

If something goes wrong in the code, brickwall is going to be very much easier to debug.  We will be able to see in which sub-function it is going wrong — we’ll be able to ignore almost all of the code when we debug.

Easy to change

This is probably the most important point. brickwall allows us to easily experiment.  We can substitute an alternative approach to one (or more) of the sub-functions with minimal effort. Code evolves, make that evolution as easy as possible.

Reusable code

The sub-functions may well find uses in other settings.  Instead of rewriting the code or mucking about extracting bits of code from a monolith, you can just use what you already have.


All in all you’re just another brick in the wall

from “Another Brick in the Wall” by Roger Waters

1 reply

Trackbacks & Pingbacks

  1. […] Bricks not monoliths – Burns Statistics […]

Leave a Reply

Want to join the discussion?
Feel free to contribute!

Leave a Reply

Your email address will not be published. Required fields are marked *

© Copyright - Burns Statistics