PureYAML Design

Field	Value
Status	draft
Created	2026-06-05
Last revised	2026-06-05

Purpose

PureYAML exists to provide a dependency-free YAML implementation in Swift that can build on macOS, Linux, and WebAssembly/WASI. The first user is OpenAPIDoctor, which needs YAML parsing and mutation without a C-backed parser blocking browser deployment.

Goals

Parse a useful YAML subset with no external dependencies.
Preserve mapping order.
Emit YAML from the same model.
Validate YAML structures with path-aware issues.
Build the library target for wasm32-unknown-wasip1.
Grow toward Yams-compatible behavior through tests, not source copying.

Non-Goals

No C source in the public package.
No external SwiftPM dependency.
No JavaScript parser or build-time generator.
No claim of complete YAML 1.2 coverage until the test corpus proves it.

Current Architecture

PureYAML
├── Model
│   ├── Value
│   ├── Mapping
│   └── Pair
├── Parsing
│   ├── Parser
│   ├── Reader
│   ├── Scanner
│   ├── Token
│   ├── TokenCursor
│   ├── TokenEventParser
│   ├── EventComposer
│   ├── Event
│   ├── Mark
│   └── ParseError
├── Emitting
│   └── Dumper
└── Validation
    ├── Validator
    ├── Rule
    ├── Issue
    └── Path

PureYAML.parse(_:) and PureYAML.dump(_:) are convenience entry points. Parsing runs through the scanner, token-event parser, and event composer before returning PureYAML.Model.Value. The implementation lives in namespaced parser, dumper, and validator types.

First Milestone

The initial parser supports:

block mappings
block sequences
ordered mappings
strings
ints
doubles
bools
nulls
quoted strings
comments outside quoted strings
flow sequences and mappings
literal and folded block scalars
anchors and aliases
YAML directives and document markers for single-document streams
explicit built-in scalar tags for strings, integers, floats, booleans, and nulls

The parser layer also has an internal event contract. Parsing.Event can represent stream, document, scalar, sequence, mapping, and alias events, with marks plus scalar and collection styles. Anchors and tags are carried as event metadata so the later scanner/composer work can add YAML feature coverage without changing the event shape.

The parser layer now also has an internal reader/scanner/event contract. Parsing.Reader advances through Swift String input while tracking UTF-8 byte indexes, lines, and columns. Parsing.Scanner emits lexical Parsing.Token values for comments, indentation, block entries, mapping indicators, flow delimiters, quoted scalar starts, block scalar headers, anchors, aliases, tags, YAML directives, document markers, and source mark ranges. Parsing.Scanner also expands tag handles from %TAG directives before emitting tag tokens. Parsing.TokenEventParser consumes that token stream and emits Parsing.Event values for block collections, flow collections, aliases, anchors, tags, and scalars. Parsing.EventComposer consumes those events into PureYAML.Model.Value, preserving ordered mapping pairs, block scalar text, alias-resolved values, and explicit built-in scalar tags where the current model can represent them.

The initial dumper emits block-style YAML from the model.

The initial validator traverses parsed YAML values and reports path-aware issues. The default rule rejects duplicate mapping keys, and callers can build a blank validator plus custom rules for project-specific checks.

Planned Compatibility Work

Compatibility should be added in small, test-backed slices:

Tag-aware collection handling and merge keys.
Validation rules beyond duplicate-key behavior.
Codable-style decoding and encoding.
Yams corpus comparison tests in a separate compatibility suite.

The private PureYAMLResearch repository may be used to study Yams behavior, but the public implementation must be written in Swift and must not copy C parser source.

Attribution Boundary

Yams and its bundled CYaml / libyaml-derived C parser are compatibility references for PureYAML. They should guide test cases and behavior questions. They are not source material to copy into this package.

4.3 KiB Raw Blame History