7 releases

0.2.0-dev.2 Nov 8, 2020
0.2.0-dev.1 Nov 1, 2020
0.1.4 Oct 5, 2020
0.1.3 Sep 9, 2020
0.1.1 Feb 19, 2020

#34 in Parser tooling

Download history 17/week @ 2020-08-07 13/week @ 2020-08-14 15/week @ 2020-08-21 11/week @ 2020-08-28 34/week @ 2020-09-04 36/week @ 2020-09-11 8/week @ 2020-09-18 10/week @ 2020-09-25 31/week @ 2020-10-02 6/week @ 2020-10-09 20/week @ 2020-10-16 8/week @ 2020-10-23 38/week @ 2020-10-30 34/week @ 2020-11-06 9/week @ 2020-11-13 11/week @ 2020-11-20

81 downloads per month


10K SLoC


Crate docs

ANTLR4 runtime for Rust programming language.

Tool(generator) part is currently located in rust-target branch of my antlr4 fork rrevenantt/antlr4/tree/rust-target Latest version is automatically built to releases on this repository. Also you can checkout it and mvn -DskipTests install

For examples you can see grammars, tests/gen for corresponding generated code and tests/my_tests.rs for actual usage examples

Implementation status

For now development is going on in this repository but eventually it will be merged to main ANTLR4 repo

Currently, requires nightly version of rust. This likely will be the case until coerce_unsize or some kind of coercion trait is stabilized. There are other unstable features in use but only CoerceUnsized is essential.

Remaining things before merge:

  • API stabilization
    • Rust api guidelines compliance
    • more tests for API because it is quite different from Java

Can be done after merge:

  • more profiling and performance optimizations
  • Documentation
    • Some things are already documented but still far from perfect, also more links needed.
  • Code quality
    • Clippy sanitation
    • Not all warning are fixed
  • cfg to not build potentially unnecessary parts (no Lexer if custom token stream, no ParserATNSimulator if LL(1) grammar)
  • run rustfmt on generated parser
Long term improvements
  • generate enum for labeled alternatives without redundant Error option
  • option to generate fields instead of getters by default and make visiting based on fields
  • make tree generic over pointer type and allow tree nodes to arena. (requires GAT, otherwise it would be a problem for users that want ownership for parse tree)
  • support stable rust
  • support no_std(although alloc would still be required)


You should use the ANTLR4 "tool" to generate a parser, that will use the ANTLR runtime, located here. You can run it with the following command:

java -jar <path to ANTLR4 tool> -Dlanguage=Rust MyGrammar.g4

For a full list of antlr4 tool options, please visit the tool documentation page.

You can also see build.rs as an example of build.rs configuration to rebuild parser automatically if grammar file was changed.

Then add following to Cargo.toml of the crate from which generated parser is going to be used:

antlr-rust = "=0.2.0-dev.1"

and #![feature(try_blocks)] in your project root module.

Parse Tree structure

It is possible to generate idiomatic Rust syntax trees. For this you would need to use labels feature of ANTLR tool. You can see Labels grammar for example. Consider following rule :

e   : a=e op='*' b=e   # mult
    | left=e '+' b=e   # add

For such rule ANTLR will generate enum EContextAll containing mult and add alternatives, so you will be able to match on them in your code. Also corresponding struct for each alternative will contain fields you labeled. I.e. for MultContext struct will contain a and b fields containing child subtrees and op field with TerminalNode type which corresponds to individual Token. It also is possible to disable generic parse tree creation to keep only selected children via parser.build_parse_trees = false, but unfortunately currently it will prevent visitors from working.

Differences with Java

Although Rust runtime API has been made as close as possible to Java, there are quite some differences because Rust is not an OOP language and is much more explicit.

  • If you are using labeled alternatives, struct generated for the rule is an enum with variant for each alternative
  • Parser needs to have ownership for listeners, but it is possible to get listener back via ListenerId otherwise ParseTreeWalker should be used.
  • In embedded actions to access parser you should use recog variable instead of self/this. This is because predicates have to be inserted into two syntactically different places in generated parser and in one of them it is impossible to have parser as self.
  • str based InputStream have different index behavior when there are unicode characters. If you need exactly the same behavior, use [u32] based InputStream, or implement custom CharStream.
  • In actions you have to escape ' in rust lifetimes with \ because ANTLR considers them as strings, e.g. Struct<\'lifetime>
  • To make custom tokens you should use @tokenfactory custom action, instead of usual TokenLabelType parser option. ANTLR parser options can accept only single identifiers while Rust target needs know about lifetime as well. Also in Rust target TokenFactory is the way to specify token type. As example you can see CSV test grammar.
  • All rule context variables (rule argument or rule return) should implement Default + Clone.


Currently, unsafe is used only for downcasting (through separate crate) and to update data inside Rc via get_mut_unchecked(returned mutable reference is used immediately and not stored anywhere)


In addition to usual Rust semantic versioning, patch version changes of the crate should not require updating of generator part


BSD 3-clause


~14K SLoC