stdf-convert

Read STDF V4 and V4-2007 semiconductor test data files as Arrow tables, one per record type. Parsing is done in Rust by rust-stdf.

import stdf_convert

tables = stdf_convert.tables("results.stdf.gz")    # {"FAR": pyarrow.Table, "PIR": ..., "PTR": ...}
ptr = stdf_convert.tables("results.stdf.gz", ["PIR", "PTR", "PRR"])["PTR"]
for record_type, batch in stdf_convert.batches("results.stdf.gz", batch_size=65_536):
    ...                                            # for files too large to hold at once
stdf_convert.schema("PTR")                         # the schema, without reading a file

Both tables and batches parse the file in Rust without holding the GIL. Gzip (.gz), bzip2 (.bz2) and zip (.zip, first member) files are read directly. Errors raise stdf_convert.StdfError; an unknown record type in a filter raises ValueError.

Installing the package also installs the stdf-convert command, which converts files to JSON Lines (stdf-convert --help), or run it without installing: uvx stdf-convert results.stdf.

Tables

Every table starts with the record's position and header, then has one column per record field, in specification order:

The schema comes from the STDF type rust-stdf declares each field with (kept in the field metadata as stdf_type), so it is the same for every file, whatever values the file holds:

Optional fields are nullable and null only where the record left them out; NaN stays NaN. Records outside the specification are kept: RESERVED (types 180 and 181) and UNKNOWN (unknown types), each with TYP, SUB, BYTE_ORDER and its payload as RAW_DATA.

VUR has UPD_CNT (uint8) and UPD_NAM (list<string>), for both layouts in use: the V4-2007 specification's single name (UPD_CNT null), and the later count and array of names, e.g. 2, ["V4-2007", "Scan:2007.1"]. A VUR that is neither is an error.

Record order across tables is given by sequence_number, e.g. to find each result's part:

import duckdb

con = duckdb.connect()
for record_type, t in stdf_convert.tables("results.stdf.gz").items():
    con.execute(f"CREATE TABLE {record_type.lower()} AS SELECT * FROM t")
con.sql("""
    SELECT pir.sequence_number AS part_start, ptr.*
    FROM ptr ASOF JOIN pir
      ON ptr.HEAD_NUM = pir.HEAD_NUM AND ptr.SITE_NUM = pir.SITE_NUM
     AND ptr.sequence_number > pir.sequence_number
""")

Development

cd python
uv venv && uv pip install maturin pytest duckdb
uv run maturin develop --release
uv run pytest