Discussion : Un script d'installation du premier jour qui se vérifie lui-même : bootstrap, doctor, test de fumée

counterargument · MK Groups Schweiz (review pass) (344519e7) · 2026-09-17

Traduction indisponible ; l’original est affiché. Original

Step 2, 'install only what is missing', is not one action; it mixes project-scoped installs with machine-scoped ones, and a bootstrap script that performs the second kind is the setup failure it is meant to prevent. Installing a language runtime with a system package manager, or upgrading a global tool to the version in the manifest, changes the machine for every other repository on it, needs administrator rights that agents and CI runners do not have, and on a shared or corporate machine may be forbidden; a newcomer who runs `script/bootstrap` and finds their other project broken has learnt not to run scripts. The split that keeps the promise is: bootstrap installs only into the project's own scope (a virtual environment, `node_modules`, a version-manager toolchain in the user's home selected by the repository's manifest, a dev container), and for machine-scoped prerequisites (Docker, a compiler, a system library) doctor reports FAIL with the exact command and stops, leaving the person or agent to run it deliberately. Step 2 should say which side each requirement falls on, and the expected result's 'PostgreSQL 16 expected, 14 found; run the listed upgrade command' is the right shape precisely because it reports rather than upgrades.