API reference

Part of magmalake: data lake building blocks in Mojo.

from avro import DataFileReader

var r = DataFileReader.open("manifest.avro")
while r.has_next():
    var rec = r.next()
    print(rec.field("manifest_path").as_string())

Value is the general reader. When the schema is known and the file is large — a scan planner walking manifests, say — RecordCursor compiles the schema into a decode plan once and then reads records into reused buffers, about ten times faster and with no per-record allocation:

from avro import RecordCursor

var c = RecordCursor.open("manifest.avro", ["manifest_path"])
var path = c.plan.slot_of("manifest_path")
while c.next():
    print(c.get_str(path))

Everything reachable from this module works on the null and deflate block codecs, which need nothing but the sibling deflate.mojo tin — raw DEFLATE, pure Mojo, no FFI. snappy and zstandard live in avro.ext_snappy / avro.ext_zstd, which pull in snappy.mojo and zstd.mojo.

inflate and deflate themselves are no longer re-exported here. They moved to deflate.mojo in avro 0.4.0, because parquet.mojo and iceberg.mojo wanted raw DEFLATE without wanting Avro. Import them from deflate.