hive-c0re's render_flake and the pieces it needs (AgentSpec, AgentPin, pinned_git_url, the forwarded-var tables, canonical_inputs_in_lock) move into a new crate with no IO. The forwarded values and the GitHub switch are now arguments; hive-c0re passes what it read from its own unit env, as before. Nothing changes on deploy: the rendered flake is byte-for-byte the same, and hive-c0re's render tests pass unchanged. Refs #4857
436 B
436 B
hive-meta-flake
The meta flake's flake.nix text: one nixosConfigurations.<agent> per agent,
each built from hyperhive's agent-base plus the agent's config input or
archetype.
Pure string rendering, with no IO and no environment reads. The caller passes
the hyperhive pin, the agents with their pins, and the values every agent
container gets. hive-c0re and swarm-controller both call render_flake, so
there is one renderer.