Architecture Overview

teckel-rs follows a strict two-crate architecture that separates domain types from parsing logic.

Crate dependency graph

teckel-parser depends on teckel-model. Nothing depends on teckel-parser at the library level -- it is the top-level entry point for consumers.

teckel-model

Pure domain types with minimal dependencies:

• serde / serde_yaml -- serialization support for data types
• thiserror -- ergonomic error definitions

Key modules:

Module	Contents
source.rs	Source enum with Input, Output, and all 45 transformation variants
asset.rs	Asset struct (wraps AssetRef + Source), Context type alias
types.rs	Newtypes (AssetRef, Column, Expression, Condition), enums (Format, WriteMode, JoinType, TeckelDataType)
error.rs	TeckelError and TeckelErrorCode covering all spec error codes
pipeline.rs	Pipeline, PipelineMetadata, Hooks, PipelineConfig, Template, Exposure, streaming types
quality.rs	QualitySuite with 10 check types (schema, completeness, uniqueness, validity, statistical, volume, freshness, referential, cross-column, custom)

teckel-parser

Depends on teckel-model and adds heavier dependencies:

• regex -- variable and secret pattern matching
• petgraph -- DAG cycle detection
• serde_yaml -- YAML deserialization

Key modules:

Module	Contents
resolve/variables.rs	${VAR:default} substitution
resolve/secrets.rs	{{secrets.alias}} resolution via SecretsProvider trait
resolve/config_merger.rs	Deep merge of multiple YAML documents
yaml/	Serde #[derive(Deserialize)] structs matching the YAML shape
rewrite.rs	Converts serde model to domain model
validate.rs	V-001 through V-008 semantic validation
expr/	Expression parser for optional expression validation
dag.rs	Tag propagation through the pipeline DAG

Design principles

1. Minimal dependencies in the model crate. Any code that needs to work with Teckel types (e.g., an execution engine) can depend on teckel-model alone without pulling in regex, petgraph, or the full YAML parser.

2. Heavy lifting in the parser crate. All I/O, string processing, and graph algorithms live in teckel-parser. This keeps the model crate fast to compile and easy to reason about.

3. Validation is separate from parsing. The YAML deserialization step produces an intermediate representation. The rewrite step converts it to domain types. Validation runs on the final domain model, ensuring that all checks operate on the canonical representation.

4. Errors carry codes. Every error includes a TeckelErrorCode from the spec (e.g., E-REF-001, E-CYCLE-001), making it straightforward to map parser errors back to spec requirements.