Hi folks,

As I mentioned yesterday, here's the release version of Mason, a Squeak build system. It relies on OSProcess to build images, but it should be possible to build other artifacts, such as directory trees, zip files, SAR files etc without OSProcess. The download is available here:

http://www.wiresong.ca/downloads/Mason-1.0.0.sar

There's no documentation yet, but I've pasted below a brief description that I sent to the Pharo folks a little while ago. 

Cheers,

Colin

---

The core model is a hierarchy of targets. Targets represent the artifacts that Mason can produce, such as files, directories, images, packages etc, and the hierarchy represents the desired state, which Mason will work to create. There's a Canvas-like API for creating target hierarchies, and Products are objects that know how to specify targets using that API.

A Product is an abstract thing, which represents a high-level concept such as "the latest version of a given package" or "the HTML documentation for a project." The role of a Product is to reduce that abstraction to a set of concrete targets which it specifies using the canvas API. A Product probably has to examine the environment to decide what targets to specify - it might scan a repository to determine the latest version of a package or scan a directory of creole files to determine which HTML files need to be generated.

Given a target hierarchy, Mason can generate a Build. A Build contains a number of jobs, which collectively create the artifacts specified by the targets. Jobs essentially MethodObjects, and each do small amounts of work. The Build is responsible for running the jobs and managing dependencies between them. The build runs the jobs in parallel where possible, and since many jobs involve IO or spawning other processes via OSProcess, so there's actually a pretty decent performance advantage.

Those are the broad outlines. There's a few issues here still, both architectural and with the current implementation.

One is that building a target tree may be an unnecessary intermediate step. It would feasible to have targets just be transient entities the way TagBrushes are in Seaside - Products would specify a Build directly. Another is that I haven't tried to do any optimization along the lines of make - everything is built fresh every time. The targets -> build converter could try to examine the disk and see if it could avoid building things that are already up to date.

I have an idea to address both these issues by having a step that walks over the target tree and optimizes it before the Build is created. This could involve identifying and merging common intermediate steps, and then doing the make trick of eliminating any targets that already exist and are up-to-date.

For now though, I've been quite happy with even the unoptimized version. I've used it to build distribution SAR archives for OB, MC2 and Filesystem, development images, load-order and package dependency testing etc.