Choosing a configuration format: JSON, YAML or TOML

What it is

JSON (RFC 8259) is a minimal data-interchange format with objects, arrays, strings, numbers, booleans and null, and no comments. YAML 1.2 is a superset of JSON in data model terms with indentation-based syntax, anchors and multi-document streams; its 1.2 core schema resolves unquoted scalars by pattern, so no, on and 1.0 can become booleans or floats depending on the loader. TOML 1.0 is a table-based format designed for configuration, with explicit types, comments and a strict specification.

Why it matters

Configuration files are edited by people and parsed by programs; the format decides how many errors happen at each end. Implicit typing and indentation sensitivity cause a class of YAML incidents that TOML and JSON avoid; missing comments make JSON poor for hand-maintained config.

How to apply

• Machine-to-machine data: JSON, validated against a schema.
• Human-edited configuration with moderate nesting: TOML (Python's pyproject.toml, Cargo, many tools).
• YAML where the ecosystem mandates it (CI pipelines, Kubernetes); then quote strings, use a schema validator, and prefer a loader that follows the 1.2 core schema.
• Whatever the format, validate the parsed result with a typed model and fail fast on unknown keys.

Pitfalls

YAML's on/off and 1e3 surprises; JSON files with trailing commas that some parsers accept and others reject; TOML's lack of nested inline structures beyond a point, which pushes very deep configuration toward another format or a split.