// the take
doit is a Make-like task runner where tasks are defined as Python dicts and functions in a dodo.py file. It tracks file hashes (not just timestamps), skips up-to-date tasks, and handles dynamic dependency graphs. Aimed at Python developers who want a build system without leaving Python or learning a DSL.
Incremental execution is based on file content hashing by default, not mtime — avoids the classic Make gotcha where touching a file triggers unnecessary rebuilds. The calc_dep feature lets tasks compute their own dependencies at runtime, which handles generated-file scenarios cleanly without hacks. Plugin architecture is real: custom commands, reporters, backends, and task loaders all have defined extension points. Test coverage looks thorough — one test file per module, including a Dockerfile for isolation.
2064 stars after 16+ years is a small audience for a general-purpose build tool; it never broke out of the Python data-science niche, which limits community momentum and third-party integrations. Task definitions as raw dicts are error-prone — typo a key like 'file_deps' instead of 'file_dep' and doit silently ignores it with no validation error. Parallel execution uses multiprocessing/threading but there's no built-in distributed execution or worker pool concept, so large DAGs on a single machine hit real limits. The dependency backend is SQLite by default, which means concurrent doit invocations from multiple processes on the same project will corrupt state.
We read the README, the directory tree and a key source file in June 2026. Nobody pays to be listed. How we judge →