Jez Cope: Software Carpentry: SC Build; or making a better make

Datetime:2016-08-22 22:27:23          Topic: Python  XML           Share

Software tools often grow incrementally from small beginnings into elaborate artefacts. Each increment makes sense, but the final edifice is a mess. make is an excellent example: a simple tool that has grown into a complex domain-specific programming language. I look forward to seeing the improvements we will get from designing the tool afresh, as a whole…

Simon Peyton-Jones, Microsoft Research (quote taken from SC Build page )

Most people who have had to compile an existing software tool will have come across the venerable make tool (which usually these days means GNU Make ). It allows the developer to write a declarative set of rules specifying how the final software should be built from its component parts, mostly source code, allowing the build itself to be carried out by simply typing make at the command line and hitting Enter .

Given a set of rules, make will work out all the dependencies between components and ensure everything is built in the right order and nothing that is up-to-date is rebuilt. Great in principle but make is notoriously difficult for beginners to learn, as much of the logic for how builds are actually carried out is hidden beneath the surface. This also makes it difficult to debug problems when building large projects. For these reasons, the SC Build category called for a replacement build tool engineered from the ground up to solve these problems.

The second round winner, ScCons, is a Python-based make-like build tool written by Steven Knight. While I could find no evidence of any of the other shortlisted entries, this project (now renamed SCons) continues in active use and development to this day.

I actually use this one myself from time to time and to be honest I prefer it in many cases to trendy new tools like rake or grunt and the behemoth that is Apache Ant . Its Python-based SConstruct file syntax is remarkably intuitive and scales nicely from very simple builds up to big and complicated project, with good dependency tracking to avoid unnecessary recompiling. It has a lot of built-in rules for performing common build & compile tasks, but it’s trivial to add your own, either by combining existing building blocks or by writing a new builder with the full power of Python.

A minimal SConstruct file looks like this:

Program('hello.c')

Couldn’t be simpler! And you have the full power of Python syntax to keep your build file simple and readable.

It’s interesting that all the entries in this category apart from one chose to use a Python-derived syntax for describing build steps. Python was clearly already a language of choice for flexible multi-purpose computing. The exception is the entry that chose to use XML instead, which I think is a horrible idea (oh how I used to love XML!) but has been used to great effect in the Java world by tools like Ant and Maven.





About List