Feature Request: Composing Grammars

[quad]

Context: I'm porting an existing handwritten parser to use parsus.

Right now, parsus is built on the presumption that a Grammar is complete. It would be helpful if Grammars could be composed.

Because it doesn't support this natively, I've resorted to this hack:

internal abstract class C<T> : Grammar<T>() {
   // ... grammar
}

val extendedGrammar = object : C<Expression>() {
    // Additional grammar
    override val root by parser_in_C_class
}

[alllex]

Do you have a concrete use-case for composing grammars?

The main problem with the naive approach here is that grammar represent a grouping of tokens that split the initial input stream into lexemes. Since the ordering/priority of tokens matter, it is not immediately clear how composing tokens sets from different grammars would behave or be intuitive for the users.

Grammars also serve as a grouping for parsers, but this grouping is loose and does not have a runtime effect. For instance, nothing prevents from using parsers from other grammars, as long as they agree on the token set. However, this could lead to hard-to-debug problems if the tokens do diverge.

[quad]

Do you have a concrete use-case for composing grammars?

I was porting an existing hand-written recursive decent parser into a Parsus grammar. The existing test suite tested different leaves of the grammar tree; whereas Parsus' API made it difficult (impossible?) to test a piece of the grammar in isolation.

My solution was to isolate the units of the grammar and expose them via their own roots.

[alllex]

If you need that entirely for testing, then you can rely on the Grammar.parse method overload that allows passing a different parser (than root) to perform parsing.

https://javadoc.io/doc/me.alllex.parsus/parsus-jvm/latest/parsus/me.alllex.parsus.parser/-grammar/parse.html