Step Paths

Every step has a path: the names of its enclosing pipelines followed by its own name, joined with /. In

(
    Node { name: "split", .. },
    Pipeline {
        name: "north",
        steps: (Node { name: "compute", .. },),
        ..
    },
    Pipeline {
        name: "south",
        steps: (Node { name: "compute", .. },),
        ..
    },
)

the two compute nodes share a name but not a path: they are north/compute and south/compute. Top-level steps’ paths are just their names (split). A flat tuple of steps adds no segment; only Pipelines (and other groups, such as the one a RecurrentNode unrolls into) do.

Paths are unique

check enforces two rules on names:

• Siblings have distinct names. Two steps in the same pipeline, or two top-level steps, may not share a name (CheckError::DuplicateStepName).
• No / in a name (CheckError::InvalidStepName), since / separates path segments.

Together these make every path in a pipeline unique, so a path identifies one step. When you build steps in a loop, give each a distinct name or, as the fan-out example does, put each in a pipeline named after its item.

Where paths are used

• Hooks receive each node and pipeline under its full path: inside a hook, n.name() is north/compute, not compute. This is what keeps per-node hook state apart — CacheHook’s cache entries, LoggingHook’s lines, VizHook’s live status. Calling name() on a step yourself still gives its local name.
• Node filters (run --nodes, --from-nodes, --to-nodes) match a path or any suffix of it that starts at a /: --nodes north/compute selects one node, --nodes compute selects both. orth/compute matches nothing.
• Viz shows full paths on the canvas and in the detail panel, and the last segment in the node tree, whose indentation shows the rest.

no_std: without an allocator there is nowhere to build a path, so on no_std hooks receive the local name. The check rules still apply.