linuxfacts 1.0.0

Typed, testable library for reading Linux system state — disks, memory, systemd units, ports and packages — with no side effects.

linuxfacts

Typed, testable library for reading Linux system state — disks, memory, systemd units, ports and packages — with no side effects.

Problem

Reading Linux system state in Python is repetitive and brittle. Every project reimplements parsing of systemctl, reading /proc, psutil calls and permission handling. The result is coupled to I/O and almost impossible to test without a real machine.

This library is a layer above psutil, opinionated about three things: everything is typed, everything is injectable (so consumers test offline), and absence of information is never confused with absence of a problem.

Features

Planned for v1.0 — implemented incrementally:

• Facts for disks, memory, CPU and load, processes, systemd units, listening ports, packages and host info
• Every result is an immutable, typed Fact[T] with an explicit ok/unknown state
• A public FakeSource so consumers test their own code offline and deterministically
• Read-only by design: no shell=True, subprocess with a fixed argument list, never requires elevated privilege
• One runtime dependency (psutil); py.typed so consumers get the types

Requirements

• Python 3.11, 3.12 or 3.13
• Ubuntu 22.04+ or Debian 12+

Installation

Not published yet. For local development:

uv sync
uv run pytest

Usage

import linuxfacts

fact = linuxfacts.disks()
if fact.is_ok:
    for disk in fact.unwrap():
        print(disk.mountpoint, disk.percent_used)
else:
    print("could not read disks:", fact.reason)

Every fact returns a Fact — either ok with a value, or unknown with a reason. A reading that could not be performed (permission denied, a missing command) is unknown, never an exception you must catch and never a silent zero.

The headline: your code, tested offline

Because every fact reads through an injectable Source, the tool you build tests without a real machine. The same call works in production (real system) and in tests (a FakeSource):

# your_tool.py
import linuxfacts
from linuxfacts.sources.base import Source


def disk_warning(source: Source | None = None) -> str | None:
    fact = linuxfacts.disks(source)
    full = [d for d in fact.unwrap_or([]) if d.percent_used > 90]
    return f"{len(full)} disk(s) over 90%" if full else None

# test_your_tool.py — offline, deterministic, no real machine
from linuxfacts.models import DiskUsage
from linuxfacts.testing import FakeSource
from your_tool import disk_warning


def test_warns_when_full():
    source = FakeSource(disks=[DiskUsage("/", 100, 95, 5, 95.0)])
    assert disk_warning(source) == "1 disk(s) over 90%"

Full documentation, including the testing guide and the compatibility policy, lives in docs/ (built with MkDocs).

Testing

make check      # ruff + mypy (strict) + pytest with coverage, one Python version
make matrix     # nox across Python 3.11–3.13 (skips missing interpreters)

Every test runs offline. Facts are exercised through FakeSource; parsers are fed versioned fixtures of real command output. Minimum coverage: 90% — it is a library, the bar is higher.

Design decisions

• docs/extraction-scope.md — what is extracted from ubuntu-doctor and what stays, and why
• docs/compatibility.md — the SemVer and deprecation promise
• Architecture decision records in docs/adr/: Fact envelope over exceptions, protocol-based source injection, testing utilities are public API

Limitations

Stated up front:

• Debian/Ubuntu only
• Read-only — never writes to or changes the system
• Synchronous — no async API in v1.0
• No journald reading in v1.0

Roadmap

v1.1 adds journald reading and per-interface network facts; v1.2 an optional cache. The public API is stable within a major version; deprecations warn for at least one minor version before removal.

Contributing

Issues and pull requests are welcome. Please read the PR template first.

License

MIT — see LICENSE.

Contact

David Oliveira — [e-mail]