Python runner

RUNNER_NAME = "python" (the default). Executes a metapipeline directly from this Python process: for each workflow in dependency (DAG) order, it builds and runs the corresponding nextflow run or nextflow module run command as a subprocess, streaming its output.

Caching and --resume

Each workflow's resolved params (after substituting any ${...} references) are cached under --cachedir (default .nf-meta-cache/), keyed by the workflow and a content hash of its definition. nf-meta run --resume:

• skips a workflow entirely if its working directory already recorded a successful prior run
• still passes -resume through to Nextflow for workflows that do run, so a partially-completed step doesn't restart from scratch either

This means a failure partway through a long chain doesn't require re-running everything that already succeeded; only the failing step onward, once fixed.

Subsetting with --start/--target

--start <id> / --target <id> run only the sub-DAG between two workflow ids (inclusive). Cross-boundary references (a workflow inside the subset referencing one excluded from it) are validated before anything runs: if the excluded workflow has no successful cached run to read a resolved value from, the runner refuses to start rather than fail partway through.

--stub

Passes -stub to Nextflow, replacing each process's script with a no-op stub, which is useful for validating a metapipeline's structure and parameter wiring without real compute. Note: nf-meta places -stub after the pipeline's URL/revision in the generated Nextflow command deliberately, working around a Nextflow CLI argument-order bug (reported upstream) where -stub placed immediately before another option can cause Nextflow to resolve the wrong pipeline.

Global params merging

globals.params is merged into each workflow's own params before the run (the workflow's own values win on key collision). This happens after caching, so the cache only ever holds workflow-specific resolved values.