1 unstable release
0.3.0 | Oct 5, 2021 |
---|
#134 in Parser tooling
465KB
10K
SLoC
antlr4rust
ANTLR4 runtime for Rust programming language.
For examples you can see grammars, tests/gen for corresponding generated code and tests/my_tests.rs for actual usage examples
ANTLR4 Tool(parser generator)
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. So if you just want to generate parser or if you want to contribute to only runtime part you don't have to do build it yourself.
But if you want to build or change generator yourself:
git clone -b rust-target https://github.com/rrevenantt/antlr4
- clone my antlr4 forkgit submodule update --init --recursive
- update Rust target submodulemvn -DskipTests install
- build generator
Implementation status
For now development is going on in this repository but eventually it will be merged to main ANTLR4 repo
Since version 0.3
works on stable rust.
Previous versions are not maintained any more
so in case of nightly breakage you should migrate to the latest version.
Usage
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:
[dependencies]
antlr-rust = "0.3"
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
otherwiseParseTreeWalker
should be used. - In embedded actions to access parser you should use
recog
variable instead ofself
/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 asself
. - str based
InputStream
have different index behavior when there are unicode characters. If you need exactly the same behavior, use[u32]
basedInputStream
, or implement customCharStream
. - 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 usualTokenLabelType
parser option. ANTLR parser options can accept only single identifiers while Rust target needs know about lifetime as well. Also in Rust targetTokenFactory
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
.
Benchmarks
Here is comparison of antlr generated XML lexer and parser
(from default XML grammar but with custom minimal Token/TokenFactory/InputStream/RuleContext) to hand-written implementations in rust ecosystem.
Keep in mind that xmlparser
and quick_xml
are much closer to being lexer than parser, so they should be compared with antlr lexer.
Also while structs used by generated lexer and parser were customized to track as minimum data as required
(which is possible by any user of antlr-rust),
internals of the lexer cannot be customized enough yet and still track quite a lot of data that might not be used in particular case.
So there is still room for improvement.
lexers:
large/large_xmlparser time: [1.8598 ms 1.8607 ms 1.8619 ms]
large/large_quick_xml time: [1.4623 ms 1.4645 ms 1.4675 ms]
large/large_antlr_xml_lexer time: [5.7866 ms 5.7877 ms 5.7891 ms]
parsers:
large/large_xmlrs time: [16.734 ms 16.748 ms 16.766 ms]
large/large_minidom time: [7.0639 ms 7.0792 ms 7.0975 ms]
large/large_roxmltree time: [4.9341 ms 4.9360 ms 4.9380 ms]
large/large_antlr_xml_full time: [10.243 ms 10.248 ms 10.252 ms]
Unsafe
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)
Versioning
In addition to usual Rust semantic versioning, patch version changes of the crate should not require updating of generator part
Licence
BSD 3-clause. Unless you explicitly state otherwise, any contribution intentionally submitted for inclusion in this project by you shall be licensed as above, without any additional terms or conditions.
Dependencies
~1–1.4MB
~23K SLoC