ceos/suse-professional-package-installer
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity Actions
suse-professional-package-i.../RATIONALE.md

5.9 KiB
Raw Permalink Blame History

Rationale

Why a ~400-500 line orchestrator/subscript script set is worth more than a one-line sudo zypper install X Y Z, and when it isn't.

The core argument

The complexity here isn't overhead wrapped around a trivial task — the logging and verification machinery is the actual value being built. A one-liner installs packages. This project installs packages and produces a record of having done so correctly.

What the pre-flight/test machinery actually buys you

  • Audit trail. Pre-flight state, exactly what was already installed vs. newly installed, post-install verification per package, and cache-cleanup outcome — all captured permanently in ~/.local/logs/ceos_installer/, rather than scrolling past in a terminal and gone.
  • Transparency. Nothing happens invisibly. Every check, skip, install, and verification is logged as it happens.
  • No silent failure. A bare zypper install X Y Z gives no structured signal when one package among several fails or is skipped — you find out later, indirectly, when something that depended on it doesn't work. test_package_available and the post-install run_test check surface that class of problem at the point it happens, not downstream.
  • Correct handling of naming and dependency quirks. Not hypothetical — see ~/.local/opt/INSTALLER_NOTES.md's tinkerpad-stt entry: ffmpeg doesn't exist as a bare package name on Tumbleweed (only ffmpeg-8 etc. do), and a naive zypper install ffmpeg just fails with no clear diagnostic path. The warn-not-fail design here exists specifically to surface that kind of problem cleanly instead of aborting the whole run or failing silently.
  • Reproducibility. The package list itself is a declarative spec of "what should be installed here." Rebuilding this machine, or standing up a second one needing the same tooling, is "run this list" instead of trying to recall what got installed by hand over months.
  • Disaster recovery. After a wipe or hardware failure, the list plus its log history is the recovery runbook — not just what to reinstall, but a record proving the last known-good state and confirming each package actually verified present afterward, not just that the install command didn't error.

Who this is actually for

A single, private user is entirely free to use the one-liner to their heart's content — nothing here claims that's wrong for personal use. The case for this project's thoroughness is specifically:

  1. A regulated business context, where a documented, auditable change process is a real requirement, not a preference. This isn't just opinion — it lines up with how established frameworks are actually written:

    • NIST SP 800-53, CM-3 (Configuration Change Control) requires documenting configuration change decisions and retaining records of configuration-controlled changes for a defined period.
    • SOC 2, CC8.1 expects a reconstructable change history and evidence of testing — the run_test PASS/FAIL/WARN lines in this project's logs are exactly that evidence.
    • ISO 27001, Annex A Control 8.32 requires changes to be planned, assessed, authorised, tested, documented, and communicated.

    These frameworks scope themselves to organizations under audit or compliance obligation — a private individual has no external party requiring this evidence, which is exactly why the one-liner remains a legitimate choice for that case.

  2. A machine you intend to keep documenting and maintaining long past this session — the specific motivating case for this repo is tinkerpad13, John's first SUSE machine, being deliberately built as an opinionated Dev/SysAdmin OS. Every install is scripted specifically for the documentation this produces, not incidentally. The payoff comes from repetition and time, not the first run — it's overkill for a machine you're about to reinstall next week anyway.

Removal is not just install-in-reverse

Apps that get installed and later turn out not to be worth keeping should have their removal documented with the same rigor as their install — an incomplete history that only records what was kept, not what was tried and rejected, isn't a complete build record.

This is not yet built (only 10_zypper-packages.sh, the install subscript, exists as of this writing). When it is, it must not simply mirror the install subscript with zypper remove swapped in — removal is inherently more dangerous than install, because it can break dependencies that other, still-wanted packages rely on. The removal subscript's design needs a zypper remove --dry-run-equivalent step that surfaces what else would be pulled out as a dependent, shown for confirmation before the real removal runs — not just a bare y/N on the package name itself.

Supporting sources