procgen.coverage

The procgen rewrite's coverage report. Every current op is listed with the new node that replaces it (from the composition sheet), its class (exact, meaning, gone), its parity state and the existing suites that call it. Every library graph (.procGraph and .procNode) is listed with the current ops it names and the worst parity state among them.

States: pending (the new node does not exist yet), ready (it exists), pass / fail (a run compared it), no-case (the case parity.json states for the op is refused by the current op, or no fixture builds its inputs), unmapped (the sheet names no counterpart), decided (a named decision: the op is gone, for example bundle.join because the bundle surface is gone). A library graph that names no current op is no-ops.

{ run = true } runs every ready entry's comparison and gives each compared op its cost as separate numbers: the current op's CPU time, the plan time, the cook's CPU time (read-back waits left out: the number the cost ratio judges), the read-back waits, the device dispatches, the read-backs and the frames waited. state, kind and query filter the lines. The answer carries the op states tallied (summary), the same tally by class (byClass), the graph states tallied (graphSummary), one line per op and per graph the filters keep (ops, graphs, count), and how many new nodes and kernel families the package holds (nodes, kernels). The data is proc.module/parity.module/parity.json; the comparison is proc.parity.

tools.use("procgen", "coverage", { state = "pending", query = "terrain" })

Part of procgen.toolbox, called as tools.use("procgen", "coverage", ...) (zero procgen coverage in the engine shell). Returns { summary, byClass, graphSummary, ops, graphs, nodes, kernels, count }.