bufrust 1.0.4

Fast BUFR edition 4 parser and decoder powered by Rust

bufrust

bufrust is a BUFR edition 4 parser and data decoder written in Rust, with a Python API designed to feel natural in data workflows.

It can:

• parse one or more BUFR edition 4 messages from files or bytes
• expose Section 0/1/2/3/4 metadata
• read unexpanded descriptors from Section 3
• use bundled ecCodes BUFR definitions, including WMO/local tables and code/flag tables
• load external ecCodes definitions or local/custom tables when needed
• load BUFR4-45 CSV-style Table B and Table D files
• expand Table D sequences and replication descriptors
• decode Section 4 values for uncompressed and compressed BUFR4 messages
• provide Python helpers such as bufrust.open(...), Dataset, Message, to_dict(), and pandas-backed to_dataframe()

The Rust core is intentionally small and explicit. The Python layer adds a friendlier interface for interactive use and notebooks.

Status

bufrust is intended as a usable BUFR4 decoding library for Python and Rust. The decoder is covered by the ecCodes BUFR4 numeric and descriptor reference data used during development, plus real ECMWF and JMA/RJTD BUFR fixtures included in this repository.

bufrust bundles the ecCodes BUFR definitions needed for descriptor expansion, value decoding, and code/flag-table meanings. You only need to pass table paths when you want to override the bundled tables with a different ecCodes release or BUFR4-45 CSV files.

Installation

Python, from a wheel:

pip install bufrust

Python, from a local checkout:

pip install maturin
maturin develop

Rust:

[dependencies]
bufrust = "1"

Quick Start In Python

Parse metadata only:

import bufrust

ds = bufrust.open("sample.bufr")

print(ds)
print(ds.metadata[0])
print(ds.descriptors[0])

Decode values using the bundled ecCodes BUFR definitions:

import bufrust

ds = bufrust.open("sample.bufr")
values = ds.decode()

for value in values[:5]:
    print(value.descriptor, value.name, value.data)

Work with multiple BUFR messages:

ds = bufrust.open("multi-message.bufr")

for message in ds:
    print(message.index, message.raw.number_of_subsets, message.descriptors)
    decoded = message.decode()
    print(len(decoded))

Load from bytes:

payload = Path("sample.bufr").read_bytes()
ds = bufrust.loads(payload)

Convert to dictionaries:

record = ds[0].to_dict()
print(record["metadata"])
print(record["values"][0])

DataFrame Workflow

Install the optional pandas extra:

pip install "bufrust[dataframe]"

to_dataframe() is the most convenient way to inspect and filter decoded BUFR data. It returns a long-form table where each decoded BUFR value is one row.

The repository includes real BUFR4 files for examples and regression tests:

tests/fixtures/ecmwf_cyclone_tracks.bufr
tests/fixtures/rjtd_drifting_buoy.bufr
tests/fixtures/rjtd_iucc10.bufr

Decode it directly into pandas:

import bufrust

ds = bufrust.open("tests/fixtures/ecmwf_cyclone_tracks.bufr")
frame = ds.to_dataframe()

print(frame.head())
print(frame.shape)

to_dataframe() returns a long-form table with columns such as descriptor, name, text, value, subset, position, and message.

The DataFrame keeps text-like data and numeric data in separate columns: text is filled from raw_text or raw_meaning, while value is filled from raw_value.

print(frame[["descriptor", "text", "value"]].head())

Use raw=True to include the underlying raw, raw_text, raw_meaning, and raw_value columns:

frame = ds.to_dataframe()
debug = ds.to_dataframe(raw=True)
first = ds.to_dataframe(message=0)

For ecmwf_cyclone_tracks.bufr, this produces 45 BUFR messages and more than 2.4 million decoded rows:

(2413286, 7)

Typical analysis patterns:

# All latitude rows.
latitudes = frame[frame["name"].str.contains("LATITUDE", case=False, na=False)]

# Values from one BUFR message and subset.
track0 = frame[(frame["message"] == 0) & (frame["subset"] == 0)]

# Keep only numeric values.
numeric = frame[frame["value"].notna()]

For large files, decode one message at a time:

ds = bufrust.open("tests/fixtures/ecmwf_cyclone_tracks.bufr")
first = ds.to_dataframe(message=0)

If pandas is not installed, use dictionaries:

record = ds[0].to_dict()
debug = ds[0].to_dict(raw=True)
print(record["values"][0])

Benchmark

The following benchmark uses tests/fixtures/ecmwf_cyclone_tracks.bufr, a 914 KB BUFR4 file containing 45 messages and 2,413,286 decoded values. The bufrust runs include code/flag table meaning lookup.

Indicative median times on a local Windows workstation:

Library / operation	Median time	Output checked
bufrust.open(path).decode_all()	0.757 s	45 messages, 2,413,286 values, 149,430 meanings
ecCodes Python unpack + numericValues/stringValues	0.772 s	45 messages, 2,413,286 numeric values, 90 strings
pybufrkit Decoder.process(...)	3.743 s	45 messages, 2,413,286 values
bufrust.open(path).to_dataframe()	3.111 s	DataFrame shape (2413286, 7)

The ecCodes and pybufrkit rows use their normal low-level Python decode paths and do not construct a pandas DataFrame or per-value text/meaning columns. The ecCodes row calls unpack and reads numericValues/stringValues; it is included as a strong reference point for decode throughput. The bufrust.decode_all() and bufrust.to_dataframe() rows include code/flag table meaning lookup, and the to_dataframe() row also includes pandas allocation for the long-form table. By default that table keeps display text in text and numeric data in value; use raw=True to expose raw_text, raw_meaning, raw_value, and raw.

Python API

The high-level API mirrors common Python data libraries:

ds = bufrust.open(path, definitions=None, table_dir=None)
ds = bufrust.load(path_or_bytes, definitions=None, table_dir=None)
ds = bufrust.loads(bytes_data, definitions=None, table_dir=None)

Use definitions= when you have an ecCodes definitions root containing bufr/tables/... and want to override the bundled tables. bufrust chooses the WMO and local table directories from the message header.

Use table_dir= when you already know the exact table directory containing element.table, sequence.def, and optional codetables/.

Decoded values expose a display-oriented data property. It is chosen in this order: raw_text, then raw_meaning, then raw_value.

record = ds[0].to_dict()                   # descriptor, name, data
metadata = ds[0].to_dict(decode=False)     # skip value decoding
debug = ds[0].to_dict(raw=True)
df = ds.to_dataframe()                     # text and value columns
debug_df = ds.to_dataframe(raw=True)

Important objects:

• Dataset: a file or byte buffer containing one or more messages
• Message: one parsed BUFR message and its original bytes
• DecodedValue: one decoded value with descriptor, name, data, raw_text, raw_meaning, raw_value, and raw
• TableSet: loaded BUFR Table B/Table D definitions plus code/flag tables
• Descriptor: an F/X/Y descriptor helper

Low-level functions are also available:

msg = bufrust.parse_file("sample.bufr")
messages = bufrust.parse_all_bytes(payload)

values = bufrust.open("sample.bufr").decode()

Table Definitions

Bundled ecCodes BUFR definitions

For ordinary use, no table path is required:

ds = bufrust.open("sample.bufr")
values = ds.decode()

The bundled definitions are available for inspection:

print(bufrust.builtin_definitions_path())
tables = bufrust.tables_for_message(ds[0])

The bundled files are copied from ecCodes 2.47.0 and include BUFR templates, WMO/local Table B and Table D files, version aliases, and codetables used for code/flag-table meanings. The full ecCodes project is licensed under Apache-2.0; its license and notice are included under python/bufrust/definitions/ECCODES_LICENSE and python/bufrust/definitions/ECCODES_NOTICE.

External ecCodes-style tables

Load one concrete table directory:

tables = bufrust.TableSet.from_eccodes(
    "external/eccodes/definitions/bufr/tables/0/wmo/42"
)
print(tables.expand([307080]))

Decode using an external ecCodes definitions root:

ds = bufrust.open("sample.bufr", definitions="external/eccodes/definitions")
values = ds.decode()

The definitions root should look like:

definitions/
  bufr/
    tables/
      0/
        wmo/
          42/
            element.table
            sequence.def
            codetables/

BUFR4-45 CSV tables

If you have BUFR4-45 CSV files, load the directory containing files named like BUFRCREX_TableB_en_*.csv and BUFR_TableD_en_*.csv:

tables = bufrust.TableSet.from_bufr4_45("external/BUFR4-45")
print(tables.get_element(42001))

Rust API

use bufrust::{decode_values_with_builtin_tables, parse_message, TableSet};

fn main() -> bufrust::Result<()> {
    let bytes = std::fs::read("sample.bufr")?;
    let message = parse_message(&bytes)?;

    println!("edition {}", message.edition);
    println!("descriptors {:?}", message.unexpanded_descriptors);

    let values = decode_values_with_builtin_tables(&bytes)?;

    println!("decoded {} values", values.len());
    Ok(())
}

Descriptor expansion:

use bufrust::TableSet;

let message = bufrust::parse_message(&std::fs::read("sample.bufr")?)?;
let tables = TableSet::from_builtin_definitions(&message)?;
let expanded = tables.expand(&[307080])?;

Development

Run the self-contained test suite:

cargo test

Build the Python wheel:

maturin build --release

Install the local Python package for development:

maturin develop
python -c "import bufrust; print(bufrust.__version__)"

Update versions before tagging a release:

python scripts/bump-version.py vX.Y.Z
git add Cargo.toml Cargo.lock pyproject.toml python/bufrust/__init__.py
git commit -m "Release vX.Y.Z"
git tag vX.Y.Z
git push origin main vX.Y.Z

Optional ecCodes Reference Tests

The default tests do not require files outside this repository.

For deeper compatibility testing, place an ecCodes source tree or extracted definitions/test-data tree inside this repository, then set BUFRUST_ECCODES_ROOT:

$env:BUFRUST_ECCODES_ROOT = "external\eccodes-2.47.0"
cargo test
cargo run --bin check_eccodes_numeric -- external\eccodes-2.47.0

The helper script downloads the BUFR test payloads referenced by ecCodes into external/eccodes-2.47.0/data/bufr:

.\scripts\download-eccodes-bufr-tests.ps1

During development the BUFR4 numeric checker reports:

numeric refs: passed=23 failed=0 unsupported=109

unsupported includes non-BUFR4 fixtures and uegabe.bufr, which ecCodes' own reference script excludes because its numeric reference is incorrect.

Notes And Limitations

• BUFR editions other than edition 4 are rejected.
• ecCodes BUFR definitions are bundled. Pass definitions= or table_dir= only when you want to override the bundled tables.
• to_dataframe() is optional and imports pandas only when called.
• The current decoded value model is long-form. Higher-level xarray-style dimensions and coordinates can be built on top of this API as the decoder matures.

License

Apache-2.0.

The bundled ecCodes-derived BUFR table files are also Apache-2.0 and retain the ecCodes notice files in python/bufrust/definitions.