Python API

mgnifam is a command-line tool first. The generated API pages document every module so that contributors can browse the docstrings, most of which explain an invariant, but only three names are a stable interface:

• the mgnifam console script and its subcommands, equivalently python -m mgnifam;
• mgnifam.__version__;
• mgnifam.generate_families.build_ssi_index, for building a shared index once upstream of many chunk tasks.

Everything else may change in any release without notice.

build_ssi_indexfunction#

View source

def build_ssi_index(fasta: str | os.PathLike[str], ssi_path: str | os.PathLike[str]) -> None

Build an Easel SSI index for fasta, replacing ssi_path atomically.

fasta must be uncompressed: Easel cannot seek within a gzip stream. Its sequence names must be unique, or DuplicateSequenceName is raised, and non-empty, or ValueError is – a bare > line otherwise reached split()[0] as an IndexError, which is the one malformed input that produced a traceback instead of a message.

The index is built inside a private directory on the destination filesystem and moved into place only on success, so a concurrent reader never observes a partial index and a crashed build leaves nothing behind. That directory is not merely tidiness: SSIWriter.close() raises before Easel closes its own writer, so on a duplicate key the native handle and any external-sort scratch files are still open and cannot be cleaned up by unlinking a known path.

Safe against PID collisions and cooperating concurrent writers. Not a defence against an actor with write access to the destination directory.