antlr-rust-runtime 0.7.0

High performance Rust runtime and target support for ANTLR v4 generated parsers
Documentation

ANTLR4 Runtime for Rust

Crates.io Version ANTLR Runtime Testsuite

antlr-rust-runtime is a pure Rust runtime and metadata generator for ANTLR v4 lexers and parsers. It is a clean-room implementation written from scratch from the public ANTLR runtime contract; it does not vendor or fork an older Rust ANTLR runtime.

First Steps

1. Install ANTLR4

Follow the ANTLR getting-started guide and install the ANTLR tool jar. The runtime tests currently validate against ANTLR 4.13.2.

2. Install the Rust ANTLR runtime tools

Each ANTLR target language needs a runtime package used by generated parsers. For Rust projects, add the runtime crate:

[dependencies]
antlr-rust-runtime = "0.7"

The library crate is imported as antlr4_runtime:

use antlr4_runtime::{CommonTokenStream, InputStream};

Install the companion generator binary:

cargo install antlr-rust-runtime

This installs antlr4-rust-gen, which turns ANTLR .interp metadata into Rust lexer and parser modules. During generation it also compiles the lexer's DFA ahead of time and embeds the tables in the generated lexer, so tokenization runs at full speed from the first character with no per-process warmup.

3. Generate your parser

The current release uses a metadata-first generation path:

  1. run the official ANTLR tool to produce .interp files,
  2. run antlr4-rust-gen to emit Rust modules,
  3. compile those modules against antlr4_runtime.

For a split lexer/parser grammar:

antlr4 MyGrammarLexer.g4 MyGrammarParser.g4

antlr4-rust-gen \
  --lexer MyGrammarLexer.interp \
  --parser MyGrammarParser.interp \
  --out-dir src/generated

The checked-in ANTLR RustTarget/StringTemplate shell is kept in tool/ and will be expanded around the same runtime contracts.

Alternative: Generate metadata with antlr-ng

antlr-ng is a TypeScript/npm parser generator based on ANTLR 4.13.2. It does not currently ship a Rust target, but it can produce the same .interp metadata that antlr4-rust-gen uses.

Install it with npm or run it through npx:

npx -y antlr-ng -Dlanguage=Java -o build/antlr --exact-output-dir true JSON.g4

The -Dlanguage=Java option selects one of antlr-ng's bundled code-generation targets only so the tool emits grammar artifacts, including JSONLexer.interp and JSON.interp. The Java files can be ignored; Rust code still comes from antlr4-rust-gen:

antlr4-rust-gen \
  --lexer build/antlr/JSONLexer.interp \
  --parser build/antlr/JSON.interp \
  --out-dir src/generated

For local tooling, antlr-ng requires Node.js 20 or newer. See the antlr-ng getting-started guide for CLI installation and option details.

Complete Example

Suppose you are using the JSON grammar from antlr/grammars-v4/json.

Fetch or copy JSON.g4, then generate ANTLR metadata:

antlr4 JSON.g4

Generate Rust modules:

antlr4-rust-gen \
  --lexer JSONLexer.interp \
  --parser JSON.interp \
  --out-dir src/generated

Declare the generated modules in your crate:

mod generated {
    #![allow(dead_code)]

    pub mod json;
    pub mod json_lexer;
}

Call the generated parser helper for the compact path:

use generated::json::{self, Json};
use generated::json_lexer::JsonLexer;

fn main() -> Result<(), antlr4_runtime::AntlrError> {
    let tree = json::parse(r#"{"a":1}"#, JsonLexer::new, Json::json)?;

    println!("{}", tree.text());
    Ok(())
}

Use parse_with_parser when you want the compact setup path and also need the parser afterward for diagnostics or the owned token stream:

use antlr4_runtime::Parser;
use generated::json::{self, Json};
use generated::json_lexer::JsonLexer;

fn main() -> Result<(), antlr4_runtime::AntlrError> {
    let output = json::parse_with_parser(r#"{"a":1}"#, JsonLexer::new, Json::json)?;
    let syntax_errors = output.parser.number_of_syntax_errors();
    let tree = output.result;
    let tokens = output.parser.into_token_stream();

    println!("{} errors across {} tokens", syntax_errors, tokens.tokens().len());
    println!("{}", tree.text());
    Ok(())
}

Or construct each layer explicitly when you need to set source names, parser options, or custom error handling before invoking the entry rule:

use antlr4_runtime::{CommonTokenStream, InputStream};
use generated::json::Json;
use generated::json_lexer::JsonLexer;

fn main() -> Result<(), antlr4_runtime::AntlrError> {
    let lexer = JsonLexer::new(InputStream::new(r#"{"a":1}"#));
    let tokens = CommonTokenStream::new(lexer);
    let mut parser = Json::new(tokens);
    let tree = parser.json()?;

    println!("{}", tree.text());
    Ok(())
}

Choosing Parser Entry Rules

Generated parsers expose one public method per grammar rule. Call the method that matches the grammar's intended top-level rule for the input; the generator can identify rules that are not called by other rules, but it cannot infer the semantic choice between multiple top-level forms. The generated parser rustdoc lists likely entry methods first, followed by all rule methods.

For the JSON grammar above, json() is the natural entry. Larger grammars may have several top-level forms, so confirm the intended entry rule against that grammar's documentation. Calling the wrong rule can still recover and return a parse tree with error nodes, so check parser diagnostics when adding a new input form.

Technical Notes

  • Pure Rust runtime implementation.
  • Written from scratch as a clean-room implementation.
  • Supports ANTLR serialized ATN deserialization.
  • Supports lexer and parser execution through generated Rust wrappers.
  • Supports real split lexer/parser grammars, including Kotlin smoke builds.
  • Passes every upstream ANTLR runtime-testsuite descriptor discovered by the harness: 357 passed, 0 failed, 0 skipped, 357 run.
  • Licensed under BSD-3-Clause for compatibility with ANTLR's runtime licensing pattern and downstream open-source applications.

The runtime contains:

  • IntStream and CharStream
  • UTF-8 input as Unicode scalar values
  • Token, CommonToken, token factories, and TokenSource
  • buffered, channel-aware CommonTokenStream
  • Vocabulary
  • recognizer metadata and error listener plumbing
  • parse tree node types, rule contexts, terminal nodes, error nodes, and walkers
  • ANTLR v4 serialized ATN deserialization
  • lexer ATN recognition with longest-match/rule-priority behavior and lexer actions
  • ahead-of-time compiled lexer DFA tables, built by antlr4-rust-gen and embedded in generated lexers, with per-token escape to ATN interpretation for constructs a finite DFA cannot represent (semantic predicates, recursive lexer rules)
  • parser ATN rule recognition with backtracking over token stream indices
  • antlr4-rust-gen, a Rust generator that consumes ANTLR .interp metadata and emits Rust modules
  • antlr4-runtime-testsuite, a harness for running upstream ANTLR runtime-test descriptors through the Rust metadata path

See docs/kotlin-build.md for the Kotlin smoke workflow. See docs/runtime-testsuite.md for the upstream runtime-testsuite harness.

Runtime Testsuite

On the maintainer checkout, where the ANTLR jar and upstream runtime-testsuite live under /tmp/antlr-cleanroom, run the full sweep with:

cargo run --quiet --bin antlr4-runtime-testsuite

Run a specific descriptor:

cargo run --bin antlr4-runtime-testsuite -- \
  --antlr-jar path/to/antlr-4.13.2-complete.jar \
  --descriptors path/to/antlr4/runtime-testsuite \
  --case LexerExec/KeywordID

Performance

tools/parse-bench/ benchmarks parse throughput of the generated Rust parsers against the upstream Go runtime (github.com/antlr4-go/antlr/v4) — and optionally the reference Python runtime and tree-sitter — on real-world Kotlin, C#, Java, and Trino SQL fixtures. See tools/parse-bench/README.md for setup (the ANTLR jar, the grammars-v4 sparse checkout, and the Python dependencies).

Run the Rust-vs-Go comparison across all fixture languages:

python3 tools/parse-bench/run.py \
  --languages kotlin,csharp,java,trino \
  --runtimes rust-antlr,go-antlr \
  --quick \
  --json target/parse-bench/results.json \
  --markdown target/parse-bench/results.md

The report prints min/avg parse time and a ratio against rust-antlr for every fixture. Drop --quick (or add --iters/--warmups) for longer, lower variance runs; add --runtimes rust-antlr,go-antlr,python-antlr,tree-sitter to include the other runtimes.

Current results

Relative parse speed of this runtime versus the Go runtime, summarized as the geometric mean of the per-fixture go ÷ rust parse-time ratios in each language group (> 1.0 means Rust is faster than Go; < 1.0 means slower):

Language Fixtures Rust vs Go (parse time)
Kotlin 4 ~18× faster
Java 4 ~1.8× faster
C# 4 ~1.2× faster
Trino SQL 5 ~1.1× faster

Rust is faster than Go on every fixture in all four language groups, with Kotlin leading dramatically (expression-ladder memoization in the generated walker). Lexer DFAs are compiled at generation time and embedded in the generated lexer, so tokenization needs no warmup at all; learned parser decision DFAs are shared across parser instances, so repeated parses of the same grammar — the common case for a CLI tool or language server — skip relearning entirely. Numbers are warm-parse minimums on an Apple M3 Pro and are indicative — re-run the benchmark on your own hardware for authoritative figures.

Useful Information