Rust API reference

Feature flags

Feature	What it enables
`fmt`	Formatter
`analysis`	Analysis (unknown tables, columns, functions, lineage)
`sqlite`	Built-in SQLite dialect (enabled by default)
`lsp`	Language server protocol implementation
`serde`	`serde::Serialize`/`Deserialize` for AST nodes and diagnostics
`dynload`	Load custom dialects from shared libraries at runtime

Formatter

Type / Method	Description
`Formatter::new()`	Create with SQLite dialect and default settings
`Formatter::with_config(&FormatConfig)`	Create with custom config
`fmt.format(sql) -> Result<String>`	Format SQL string
`FormatConfig`	Builder: `with_line_width()`, `with_indent_width()`, `with_keyword_case()`, `with_semicolons()`
`KeywordCase`	`Upper` or `Lower`

The formatter is reusable; call format() repeatedly. Internal allocations are reused across calls. Defaults: 80-char lines, 2-space indent, uppercase keywords, semicolons on.

Parser

Type / Method	Description
`Parser::new()`	Create a parser for the SQLite grammar
`parser.parse(sql) -> ParseSession`	Start a parse session
`session.next() -> ParseOutcome`	Yield next statement
`ParseOutcome::Ok(stmt)`	Successfully parsed statement
`ParseOutcome::Err(err)`	Parse error (parser recovers and continues)
`ParseOutcome::Done`	No more statements

The parser yields statements one at a time, reusing internal allocations. It recovers from errors and continues parsing subsequent statements.

Tokenizer

Type / Method	Description
`Tokenizer::new()`	Create a tokenizer for the SQLite grammar
`tokenizer.tokenize(sql) -> Result<Vec<Token>>`	Tokenize SQL string
`Token`	`token_type`, `text`, byte offsets into source

Zero-copy: tokens reference byte offsets into the source string.

Analyzer

Type / Method	Description
`Analyzer::new()`	Create an analyzer for the SQLite dialect
`AnalysisContext::new(&mut catalog)`	Bundle the catalog, config, and optional module resolver for a single analysis call
`analyzer.analyze(sql, &mut ctx) -> Analysis`	Analyze SQL, returning diagnostics and lineage
`Catalog::new(dialect)`	Create an empty catalog
`catalog.layer_mut(CatalogLayer::Database).insert_table(name, cols, false)`	Register a table
`AnalysisConfig::default()`	Default config (warnings for unknowns)
`AnalysisConfig::default().with_strict_schema()`	Strict mode (errors for unknowns)
`model.diagnostics()`	Parse and semantic diagnostics

The analyzer is reusable; call analyze() repeatedly. The catalog uses a layered resolution order; populate the Database layer with your schema.

Lineage

After analyze(), the returned Analysis provides column-level lineage for SELECT statements. Lineage traces each result column back to its source table and column.

model.statements() yields one StatementAnalysis per statement; lineage methods are per-statement.

Type / Method	Description
`stmt.lineage()`	Per-column lineage: `Option<LineageResult<&[ColumnLineage]>>`
`stmt.relations_accessed()`	Relations in FROM: `Option<LineageResult<&[RelationAccess]>>`
`stmt.physical_tables_accessed()`	Physical tables after resolving CTEs/views: `Option<LineageResult<&[PhysicalTableAccess]>>`
`stmt.defined_relations()`	`&[DefinedRelation]` — targets created by DDL (`CREATE TABLE`/`CREATE VIEW`)
`stmt.unexpanded_views()`	`&[String]` — canonical names of views whose bodies weren't available for expansion
`LineageResult<T>`	`Complete(T)` — fully resolved, or `Partial(T)` — some view bodies unavailable
`ColumnLineage`	`name: String`, `index: u32`, `origin: Option<ColumnOrigin>`
`ColumnOrigin`	`table: String`, `column: String`
`RelationAccess`	`name: String`, `kind: RelationKind`
`RelationKind`	`Table` or `View`
`PhysicalTableAccess`	`name: String`
`DefinedRelation`	`name: String`, `is_view: bool`

Lineage methods return None for non-query statements (CREATE, INSERT, etc.). A column's origin is populated only when the column is an untransformed passthrough — a direct column reference or a pure rename alias. Expressions, casts, aggregates, and set-op (UNION, etc.) columns have origin: None. Result is Partial when a referenced view's body is unavailable for tracing through; unexpanded_views() lists the view names.