DSL

The poly-bench-dsl crate is the front-end of the poly-bench pipeline. It converts .bench source files into a structured Abstract Syntax Tree (AST) that downstream crates consume. This page covers the lexer, parser, AST types, validation, and formatter.

---

Overview

The DSL pipeline:

1. Lexer — Converts character stream into tokens (keywords, identifiers, strings, numbers, punctuation)
2. Parser — Parses token stream into AST nodes
3. Validator — Semantic checks (fixture references, language block completeness, configuration)
4. Formatter — Pretty-prints .bench files for consistent style

---

Lexer

The lexer (poly-bench-dsl/src/lexer.rs) tokenizes source text. It maintains line/column tracking for error reporting and handles:

• Keywords — suite, bench, setup, fixture, declare, use, import, helpers, after, before, each, globalSetup, performance, timeBased, sameDataset, baseline, warmup, targetTime, count, hex, data, etc.
• Identifiers — Suite names, benchmarks, fixture names, language tags (go, ts, rust, python, c, csharp, zig)
• Literals — strings (double-quoted), hex strings, numbers (with ms suffix for durations)
• Punctuation — {, }, (, ), :, ,, @, =

The lexer produces a Vec<Token> with TokenKind and Span for each token. Errors include source location for diagnostics.

---

Parser

The parser (poly-bench-dsl/src/parser.rs) consumes tokens and builds an AST. The grammar is defined in poly-bench-grammar (Tree-sitter) and the parser produces:

• File — Top-level container with use directives and suite declarations
• Suite — declare suite or suite blocks with configuration, setup blocks, fixtures, benchmarks, and after blocks
• Setup — Per-language blocks with import, declare, init, and helpers
• Fixture — hex, data, encoding, format, or selector properties
• Bench — Benchmark blocks with language-specific expressions and optional before/each/after hooks

The parser is recursive-descent and handles error recovery. Invalid input produces ParseError with span information.

---

AST Types

The AST (poly-bench-dsl/src/ast.rs) defines:

• Lang — Enum for supported languages (Go, TypeScript, Rust, Python, C, C#, Zig)
• Span — Line/column/byte range for error reporting
• File, Suite, Setup, Bench, Fixture — Node types with serialization
• CodeBlock — Embedded language code (Go, TypeScript, Rust, Python) with language tag

AST nodes are serializable (Serialize/Deserialize) for caching and tooling.

---

Validator

The validator (poly-bench-dsl/src/validate.rs) runs semantic checks:

• Fixture references — Benchmarks reference defined fixtures
• sameDataset consistency — When sameDataset: true, all benchmarks use the same fixture set
• Setup completeness — Each language used in benchmarks has a setup block
• Configuration — baseline, warmup, targetTime, count are valid

Validation produces ValidationResult with ValidationError and ValidationWarning for each issue.

---

Formatter

The formatter (poly-bench-dsl/src/formatter.rs) pretty-prints .bench files:

• format_file(source, filename) — Returns formatted string
• format_file_with_options — Configurable indentation and line width

Used by poly-bench fmt and the LSP for incremental formatting.

---

Related Crates

---

Further Reading

• DSL Reference — Complete syntax documentation
• Architecture — Pipeline overview
• LSP v2 — Editor integration