toml.bzl

A pure Starlark TOML 1.1.0 encoder and decoder.

Usage

For full API documentation, see the API Reference.

For installation instructions, see Releases.

In a .bzl file

load("@toml.bzl", "toml")

content = """
[database]
server = "192.168.1.1"
ports = [ 8000, 8001, 8002 ]
"""

# Decode
config = toml.decode(content)
print(config["database"]["server"])

# Encode
encoded = toml.encode(config)
print(encoded)

Performance

toml.bzl is highly optimized for the Starlark interpreter, leveraging native string operations to achieve high throughput while maintaining 100% compliance with TOML 1.1.0.

Benchmarks

Time per decode() call on an Apple M3 Pro MacBook Pro (Bazel 8.5.0, warm server):

Document	Size	Time per parse	Throughput
Cargo.lock	735 KB	61 ms	~12 MB/s
channel-rust-1.81.0.toml	824 KB	61 ms	~13 MB/s
uv.lock (apache/airflow)	3.3 MB	216 ms	~15 MB/s

To reproduce, build //toml/tests/benchmarks:benchmark_prof, //toml/tests/benchmarks:channel_rust_benchmark_prof, or //toml/tests/benchmarks:uv_lock_benchmark_prof (each parses its document 5 times during analysis) with --nobuild --starlark_cpu_profile=<file>, and inspect decode in the profile.

Compliance

The implementation is verified against the toml-test suite, passing all compliance tests (decoding and encoding).