Crate modewise_pipeline

Expand description

Pipeline engine: specs, operation registry, planning, cached execution.

Structs§

ArtifactKey: SHA-256 content address of an artifact, as lowercase hex.
OperationDescriptor
PipelineHeader
PipelineSpec: A pipeline as the user writes it, usually in TOML.
Plan: A checked pipeline: operations exist, parameters are valid, references resolve, kinds match, and steps are in dependency order.
Port: A named input or output. Inputs may accept several kinds. An output lists one kind, or several when Operation::output_kind picks one from the kinds of the inputs.
RecordingRun: Outcome for one recording. A failure here does not stop the others.
Registry: Operations available to pipelines, by name.
RunManifest: What a run did, for provenance: every step’s operation, parameters and data keys.
RunResult
StepRecord
StepSpec
StoreError

Artifact: Data that flows between pipeline steps.
ArtifactKind
InputDecl
InputSpec: input = "x" for single-input operations, or input = { port = "x" }. A reference is a pipeline input name, a step id, or step.output.
PipelineError

ArtifactStore: Where step results live. Implemented by adapters (memory, Zarr, …).
InputSource: Supplies pipeline inputs for many recordings, for example files found on disk.
Operation: One kind of pipeline step, for example a decomposition method.

resolve_params: Round-trips raw parameters through a typed struct, which applies defaults and rejects unknown fields.
run: Runs a plan. Steps whose outputs are already in the store are skipped.
run_batch: Runs a plan once per recording, workers recordings at a time (0 = all cores). on_done sees each recording as soon as it finishes, for progress or saving.
schema_of: The JSON Schema of a parameter struct.