wkb_wkt_converter

A high-performance, zero-dependency Rust library for streaming conversion between WKT/EWKT and WKB/EWKB geometry formats used in GIS systems (PostGIS, GDAL, etc.).

Exposes both a Rust API and Python bindings (via PyO3/maturin).

Features

• All 7 OGC geometry types: Point, LineString, Polygon, MultiPoint, MultiLineString, MultiPolygon, GeometryCollection
• All coordinate dimensions: XY, XYZ, M, ZM
• Both EWKB (PostGIS flag-bit encoding) and ISO WKB (type+1000 offset encoding) as input
• SRID preservation (EWKT ↔ EWKB); option to split SRID from geometry
• Big-endian and little-endian WKB input; little-endian EWKB output for full conversions, with simple header rewrites preserving input byte order
• EMPTY geometry support
• Per-member EMPTY support inside MULTI* geometries
• Hex WKB convenience helpers
• Memory-safe handling for malformed input: structurally invalid input returns descriptive errors, while trusted-valid fast paths may pass through malformed geometry bodies as invalid output bytes/text.

Conversion strategy

WKB → WKT is a straightforward streaming read. WKT → WKB uses a seekable-buffer approach: count fields (ring count, point count, etc.) are written as 0 placeholders and patched in-place after the coordinates are streamed, avoiding a two-pass scan of the input.

Rust API

Add to Cargo.toml:

[dependencies]
wkb_wkt_converter = { path = "wkb_wkt_converter" }

Functions

// WKB/EWKB → WKT/EWKT, with SRID output control
pub fn wkb_to_wkt(wkb: &[u8], srid: SridMode) -> Result<String>

// WKB/EWKB → WKT, SRID returned separately (not in the string)
pub fn wkb_to_wkt_split_srid(wkb: &[u8]) -> Result<(String, Option<i32>)>

// WKT/EWKT → EWKB bytes, with SRID output control
pub fn wkt_to_wkb(wkt: &str, srid: SridMode) -> Result<Vec<u8>>

// WKT/EWKT → EWKB bytes, SRID returned separately (not in bytes)
pub fn wkt_to_wkb_split_srid(wkt: &str) -> Result<(Vec<u8>, Option<i32>)>

// WKT/EWKT → uppercase hex-encoded EWKB string, with SRID output control
pub fn wkt_to_hex_wkb(wkt: &str, srid: SridMode) -> Result<String>

// Hex-encoded WKB/EWKB → WKT/EWKT string, with SRID output control
pub fn hex_wkb_to_wkt(hex: &str, srid: SridMode) -> Result<String>

// Read only the top-level EWKB SRID header, without converting the geometry
pub fn wkb_header_srid(wkb: &[u8]) -> Result<Option<i32>>
pub fn wkb_header_srid_including_unknown(wkb: &[u8]) -> Result<Option<i32>>

Generic converters

These functions accept an explicit Input: a WKT/EWKT string, a hex-encoded WKB/EWKB string, or raw WKB/EWKB bytes. Text input is detected automatically (a non-empty, even-length string composed entirely of hex characters is treated as hex WKB; anything else, including odd-length all-hex text, is treated as WKT). Input::Wkb is always treated as raw WKB/EWKB, not encoded text.

pub enum Input<'a> {
    Text(&'a str),
    Wkb(&'a [u8]),
}

pub fn to_wkb(input: Input<'_>, srid: SridMode) -> Result<Vec<u8>>
pub fn to_wkt(input: Input<'_>, srid: SridMode, normalize_wkt: bool) -> Result<String>
pub fn to_hex_wkb(input: Input<'_>, srid: SridMode) -> Result<String>

SridMode controls SRID handling in the output of direct and generic converters:

wkb_header_srid returns None for headers without an EWKB SRID flag and for raw non-positive SRID values. Use wkb_header_srid_including_unknown to distinguish an EWKB SRID flag carrying an unknown raw value: raw 0 returns Some(0), and raw 0xffffffff returns Some(-1) under the signed i32 SRID model.

Validation notes:

• WKT coordinates must be finite. WKB coordinate payloads are treated as trusted-valid: all-NaN Point remains POINT EMPTY, while other NaN or infinity values may format as invalid WKT instead of raising an error.
• Direct WKB-to-WKT entry points still reject structurally invalid WKB such as truncation, unsupported type codes, excessive nesting, and trailing top-level bytes.
• For simple Point, LineString, and Polygon EWKB/WKB hex or raw WKB input in either byte order, to_wkb(input, SridMode::Strip | SridMode::Set(_)) and the equivalent to_hex_wkb paths patch only the top-level header. Malformed coordinate bodies or trailing bytes in those simple fast paths can pass through as invalid output. ISO-dimensional, collection, and non-canonical type headers fall back to a full normalising round-trip.
• to_wkb(Input::Text(hex), SridMode::Auto) returns decoded bytes without WKB structure validation, and to_hex_wkb(Input::Text(hex), SridMode::Auto) validates and uppercases the hex text only. Input::Wkb under SridMode::Auto is copied or hex-encoded without WKB structure validation.
• GeometryCollection dimension tags are not inherited by child geometries. An XY collection may contain heterogeneous immediate children. A Z, M, or ZM collection requires each immediate child to declare the same dimension.
• WKT and WKB parsing reject geometry nesting beyond the implementation depth limit of 128.

to_wkt accepts a normalize_wkt: bool parameter. When true, WKT input is normalised (canonical casing, spacing, coordinate formatting) via a round-trip through WKB. When false, only the SRID prefix is adjusted — no validation is performed: malformed WKT is returned without error. Leading/trailing whitespace is always trimmed regardless of this flag. Hex WKB and raw WKB input are always decoded to normalised WKT regardless of this flag. Odd-length all-hex input is not detected as hex WKB; with normalize_wkt=false it follows the same unvalidated WKT fast path.

Example

use wkb_wkt_converter::{wkt_to_wkb, wkb_to_wkt, wkt_to_wkb_split_srid, hex_wkb_to_wkt};
use wkb_wkt_converter::{to_wkt, to_hex_wkb, Input, SridMode};

// Basic round-trip
let wkb = wkt_to_wkb("POINT (1 2)", SridMode::Auto)?;
let wkt = wkb_to_wkt(&wkb, SridMode::Auto)?;
assert_eq!(wkt, "POINT (1 2)");

// With SRID embedded
let wkb = wkt_to_wkb("SRID=4326;POINT Z (1 2 3)", SridMode::Auto)?;
let wkt = wkb_to_wkt(&wkb, SridMode::Auto)?;
assert_eq!(wkt, "SRID=4326;POINT Z (1 2 3)");

// SRID split from geometry
let (wkb, srid) = wkt_to_wkb_split_srid("SRID=4326;LINESTRING (0 0, 1 1)")?;
assert_eq!(srid, Some(4326));
// wkb contains a plain (non-EWKB) LineString

// All geometry types and dimensions work the same way
let wkb = wkt_to_wkb(
    "MULTIPOLYGON ZM (((0 0 0 1, 1 0 0 1, 1 1 0 1, 0 0 0 1)))",
    SridMode::Auto,
)?;
let wkt = wkb_to_wkt(&wkb, SridMode::Auto)?;
assert_eq!(wkt, "MULTIPOLYGON ZM (((0 0 0 1, 1 0 0 1, 1 1 0 1, 0 0 0 1)))");

// Generic converters: input format (WKT, hex WKB, or raw WKB) selected explicitly/detected automatically
// Normalise WKT (casing, whitespace) — SridMode::Auto mirrors the input SRID
let wkt = to_wkt(Input::Text("point(1 2)"), SridMode::Auto, true)?;
assert_eq!(wkt, "POINT (1 2)");

// Add or override an SRID regardless of what the input contains
let hex = to_hex_wkb(Input::Text("POINT (1 2)"), SridMode::Set(4326))?;
// hex is an EWKB string encoding SRID=4326;POINT (1 2)
let wkt = hex_wkb_to_wkt(&hex, SridMode::Strip)?;
assert_eq!(wkt, "POINT (1 2)");

// Raw WKB bytes can also go through the generic converters
let wkt = to_wkt(Input::Wkb(&wkb), SridMode::Auto, false)?;
assert_eq!(wkt, "MULTIPOLYGON ZM (((0 0 0 1, 1 0 0 1, 1 1 0 1, 0 0 0 1)))");

// Strip the SRID without re-encoding (fast path)
let wkt = to_wkt(Input::Text("SRID=4326;POINT (1 2)"), SridMode::Strip, false)?;
assert_eq!(wkt, "POINT (1 2)");

Error handling

All functions return Result<_, wkb_wkt_converter::Error>:

pub enum Error {
    InvalidWkt(String),
    InvalidWkb(String),
    UnsupportedGeometryType(u32),
}

Fuzzing

cargo-fuzz support for the core crate lives in fuzz/README.md. It includes five targets covering raw WKB bytes, arbitrary UTF-8 text, generic Input dispatch behavior, bounded valid-geometry round-trips, plus a depth-boundary target for nested GEOMETRYCOLLECTION inputs. The GitHub Actions fuzz workflow is a smoke/regression check; it complements, but does not replace, longer local or continuous fuzzing campaigns.

Python API

Build and install

Requires Python 3.10 or newer, maturin, and a Rust toolchain. Wheels include PEP 561 type stubs for static type checkers and IDEs.

pip install maturin
maturin develop          # install into the current virtualenv (dev mode)
maturin build --release  # build a wheel

Functions

from wkb_wkt_converter import (
    wkb_to_wkt,
    wkb_to_wkt_split_srid,
    wkt_to_wkb,
    wkt_to_wkb_split_srid,
    wkt_to_hex_wkb,
    hex_wkb_to_wkt,
    wkb_header_srid,
    # generic converters
    to_wkb,
    to_wkt,
    to_hex_wkb,
)

WKB arguments in the Python API accept bytes, bytearray, memoryview, and other C-contiguous one-byte buffer objects. They are always treated as raw WKB/EWKB, not as encoded text.

For performance, bytes-like WKB inputs may be borrowed directly without copying. Do not mutate a writable buffer while a converter is reading it, including from native code, another thread, or shared memory. Mutating the buffer during conversion is unsupported and may produce invalid or inconsistent results.

All functions above raise ValueError on invalid geometry input. The bytes-like WKB functions raise BufferError when the Python object does not provide a C-contiguous one-byte buffer. (See to_wkt below for an exception when normalize_wkt=False.)

Generic converters

These three functions accept a WKT/EWKT string, a hex-encoded WKB/EWKB string, or bytes-like WKB/EWKB input. String input is detected automatically: a non-empty, even-length string composed entirely of hex characters is treated as hex WKB; anything else, including odd-length all-hex text, is treated as WKT. Bytes-like input is always treated as raw WKB/EWKB, not encoded text.

The srid keyword argument on direct and generic converters controls SRID handling in the output:

Validation behavior matches the Rust API: WKT coordinates must be finite, while WKB coordinate payloads are treated as trusted-valid. Simple Point, LineString, and Polygon EWKB/WKB hex or bytes-like inputs in either byte order under srid=False or an integer SRID are patched at the top-level header without scanning the body, so malformed bodies or trailing bytes can pass through as invalid output. to_wkb(source, srid=None) returns decoded/raw bytes without WKB structure validation, and to_hex_wkb(source, srid=None) validates and uppercases hex text or hex-encodes bytes-like input without WKB structure validation.

wkb_header_srid inspects only the top-level EWKB header. By default it returns None for headers without an SRID flag and for raw non-positive SRID values. With include_unknown=True, a present SRID flag carrying raw 0 returns 0, and raw 0xffffffff returns -1 under the signed i32 SRID model.

to_wkt accepts a normalize_wkt keyword argument (default False). When True, WKT input is normalised (canonical casing, spacing, coordinate formatting) via a round-trip through WKB. When False (the default), only the SRID prefix is adjusted — no validation is performed: malformed WKT is returned without raising an error. Leading/trailing whitespace is always stripped regardless of this flag. Hex WKB and bytes-like WKB input are always decoded to normalised WKT regardless of this flag. Odd-length all-hex input is not detected as hex WKB; with normalize_wkt=False it follows the same unvalidated WKT fast path.

Example

from wkb_wkt_converter import wkt_to_wkb, wkb_to_wkt, wkt_to_hex_wkb, hex_wkb_to_wkt
from wkb_wkt_converter import to_wkt, to_hex_wkb

wkb = wkt_to_wkb("POINT (1 2)")
wkt = wkb_to_wkt(wkb)
assert wkt == "POINT (1 2)"

# EWKT with SRID
wkb = wkt_to_wkb("SRID=4326;POLYGON ((0 0, 1 0, 1 1, 0 0))")
wkt = wkb_to_wkt(wkb)
assert wkt == "SRID=4326;POLYGON ((0 0, 1 0, 1 1, 0 0))"

# Hex WKB (common PostGIS text format)
hex_wkb = wkt_to_hex_wkb("POINT (1 2)")
wkt = hex_wkb_to_wkt(hex_wkb)
assert wkt == "POINT (1 2)"

srid_hex_wkb = wkt_to_hex_wkb("SRID=4326;POINT (1 2)")
wkt = hex_wkb_to_wkt(srid_hex_wkb, srid=False)
assert wkt == "POINT (1 2)"

# Generic converters: input format detected automatically
wkt = to_wkt("point(1 2)", normalize_wkt=True)  # normalise WKT
assert wkt == "POINT (1 2)"

wkt = to_wkt(hex_wkb)                           # hex WKB -> WKT (always normalised)
assert wkt == "POINT (1 2)"

wkt = to_wkt(wkb)                               # raw WKB bytes -> WKT
assert wkt == "POINT (1 2)"

hex_out = to_hex_wkb("POINT (1 2)", srid=4326)  # add SRID
wkt = to_wkt(hex_out)
assert wkt == "SRID=4326;POINT (1 2)"

wkt = to_wkt("SRID=4326;POINT (1 2)", srid=False)  # strip SRID (fast path)
assert wkt == "POINT (1 2)"

Benchmarks

Comparison against shapely 2.x. Regenerate with:

pip install ".[benchmark]"
python scripts/update_readme_benchmarks.py
# or, using an already-saved JSON:
python scripts/update_readme_benchmarks.py --json benchmark_results.json

For local performance history across commits, use airspeed velocity:

pip install ".[asv]"
asv check --python=same
asv run --python=same --quick --show-stderr --dry-run
asv run HEAD^! --quick --show-stderr --dry-run
asv run main..HEAD --skip-existing-commits --show-stderr
asv run ALL --skip-existing-commits --show-stderr
asv publish
asv preview
asv compare main HEAD --split

2026-05-05 — Python 3.12.3 — 12th Gen Intel(R) Core(TM) i7-12700KF

Times are mean latency per call (lower is better). Speedup = shapely mean ÷ wkb_wkt_converter mean.

Basic conversions

wkt_to_wkb

wkb_to_wkt

wkt_to_hex_wkb

hex_wkb_to_wkt

Generic to_* converters

to_wkb(wkt)

to_wkb(hex_wkb)

to_wkt(wkt) — normalize_wkt=False (fast path, no WKB round-trip)

to_wkt(wkt) — normalize_wkt=True (full WKT→WKB→WKT round-trip)

to_wkt(hex_wkb)

to_hex_wkb(wkt)

to_hex_wkb(hex_wkb)

Project layout

wkb_wkt_converter/          # core Rust library (zero runtime dependencies)
  src/
    lib.rs                  # public API
    error.rs
    types.rs                # GeomType, Dimension
    wkb_to_wkt/             # WKB reader + WKT builder
    wkt_to_wkb/             # WKT tokenizer + seekable WKB writer
  tests/
    wkb_to_wkt.rs
    wkt_to_wkb.rs
    generic_converters.rs

wkb_wkt_converter_py/       # Python bindings (PyO3 / maturin)
  src/lib.rs

pyproject.toml              # maturin build config

Running tests

cargo test                  # run all Rust tests
cargo clippy -- -D warnings # lints
cargo fmt --check           # formatting