What kizz is

This page explains what kizz does, the principles it follows, what it deliberately is not, and the tools it learned from.

What kizz does

Kizz is a build engine for Linux. Its work is described in kizz.md files: ordinary Markdown files whose fenced starlark blocks are tasks. Each task is a small program in Starlark. A task program runs commands, inspects their results, reads and writes files, and decides what happens next.

When kizz runs a command for a task, it traces the command and its whole process tree. It records every path that the command read, checked, listed, wrote or ran, including lookups that found nothing, apart from the paths that the task leaves out with ignore patterns. Those observations are the command's dependencies, and kizz stores them in a cache with the command's exit status and output. On the next build, kizz checks each recorded path against the file system. When nothing has changed, it normally skips the command and gives the task program the recorded result instead. Some commands run again anyway, such as one marked to run on every build or one whose last result could not be cached; caching lists every case.

These are the commands of kizz:

Planned: kizz graph, to show the learned task graph, and kizz image, to build container images. Today both are parsed but not implemented, as planned features explains.

Philosophy

Kizz follows a few principles, which the getting started page and the rest of the guide put to work.

What kizz is not

Kizz is deliberately narrow. It is not:

kizz execute is a diagnostic tool, not a replacement for strace: it reports paths and how they were used, not every call.

Prior art

Kizz belongs to a known family of tracing build systems. These are the closest: