LR(1) and LALR(1) parser generator by Chuck Liang.
A Tutorial with several examples is available.
The project grew out of the author's compiler construction and programming languages classes over the years and has been mainly used for implmentating modestly scaled, experimental programming languages. But it is becoming sophisticated enough to be more than just a project and will continue to improve over time.
scanner from a minimal set of grammar declarations has been added, using the built-in RawToken and StrTokenizer. Other tokenizers can still be used in the previous way, by adopting them to the Tokenizer interface.
Significant improvements required that several components are now renamed, while the older ones are retained for compatibility with parsers already created.
LBox<dyn Any> as abstract syntax type by
automatically generating runtime type casting. This means that
semantic actions for grammar productions no longer need to return values of the same
type. However, this also means that abstract syntax representations
cannot contain non-static references due to the Rust restriction that
such types cannot impl Any. An alternative approach would be to generate
a enum type that includes all possible return types, but this approach is
not compatible with allowing the lexical analyzer to be decoupled from
the parser. The ability to train the parser has been added: the Runtime::parse_train function will ask for user input to improve error reporting by augmenting the basic generated LR state machine with Error entries.
Fixed problem with Accept state; added LBox smartpointer for encapsulating lexical information into abstract syntax.
The parse function has been decomposed into a parse_core, which takes a functional argument that handles error reporting. This allows a custom parser interface to be created if one does not wish to be restricted to the supplied RuntimeParser::parse function, which uses stdio.
Training the parser now modifies the same parser file that it reads from. The ability to use LBox and LRc for non-intrusively encapsulating lexical (line/column/source) information into abstract syntax has been expanded. Fixes an error where a non-terminal symbol is declared without any rules defined for it.
parsecore has been retained but a new parsebase function is introduced that takes as input the error handler as a trait object. This should allow better flexibility in building custom parser interfaces while still using the basic state machine generated.
Constructing a parser that gives helpful error messages can be tricky, especially after a grammar has been modified and the parser is re-generated, which changes the state transition table. Interactive training with the parsetrain function now produces, in addition to an augmented parser, a training-script that records each error encountered along with the line, column numbers and the unexpected token. It's the user's responsibility to keep track of the sample input used during interactive training and the script that was created from it. A parser can be retrained from the script, given the identical input (and tokenizer) using the RuntimeParser::trainfrom_script function.
Future releases of rustlr will further enhance the training feature.
We also hope to identify a robust, generic lexical tokenizer tool for Rust so that the parser generator can also automatically generate a lexical analyzer from additional specifications in the grammar. Another potential feature to be explored is the ability to generate an abstract syntax type structure from the grammar itself.
This version's main enhancements are pattern labels. In a grammar production, the value attached to nonterminal and terminal symbols can be extracted by specifying a pattern, which will cause an if-let statement to be automatically generated. For abstract syntax with many layers of enums and structs, but which shares a single "absyntype" for the grammar. For example, if Exp and Expl are variants of a common enum, one can now write rules such as
Exprlist --> { Expl(Vec::new()) }
Exprlist --> Exprlist:@Expl(mut ev)@ , Expr:@Exp(e)@ {ev.push(e); Expl(ev)}
This capability was used to construct a parser for a scaled-down version of
Java and is included in the examples directory of the repository.
Abilities for using LBox were also extended, which allows LBox<dyn Any> to
be used as the abstract syntax type, with functions and macros for
up/downcasting.