Skip to main content

Architecture

Each arrow is a Cargo dependency edge (A --> B means "A depends on B"). Every vendor crate depends on openmassspec-core and nothing else in the stack; openmassspec-io is the only crate that depends on more than one vendor crate at a time.

Layering rules​

  • openmassspec-core knows nothing about vendors. It owns the shared schema, the mzML byte format, the Arrow layout, and the conformance harness. Anything generic enough to be shared between Thermo / Bruker / Waters / Agilent / SCIEX / Shimadzu lives here.
  • Vendor crates know nothing about each other. Each implements SpectrumSource and a write_mzml(path, writer) helper. They depend on openmassspec-core and zero other vendor crates.
  • openmassspec-io is the only place that knows the full vendor set. Format detection, dispatch, and feature-gated re-exports live here, behind the thermo, bruker, waters features.
  • openmassspec-io-cli and openmassspec-io-py depend on openmassspec-io. They never call vendor crates directly; if they need something from a vendor, that something is promoted to the umbrella first.

Streaming model​

A SpectrumSource exposes iter_spectra(&mut self) -> Box<dyn Iterator<Item = SpectrumRecord> + '_>. Both the mzML writer and the Arrow batch builder consume this iterator without ever holding more than a single spectrum in memory at a time (except when the consumer explicitly buffers, as --validate does).

This keeps OpenMassSpec usable for the multi-gigabyte timsTOF runs that come out of dia-PASEF experiments without requiring temp files or mmap tricks.