Allow unassigned rule calls after an assignment

[spoenemann]

Xtext forbids an unassigned rule call after an assignment because it uses the assignment as trigger for creating an EClass instance. We could relax this, and with that would gain a big advantage when it comes to factoring out common prefixes.

Example:

Entry:
    A | B | C;

A:
    prefix=Prefix
    'a' ...;

B:
    prefix=Prefix
    'b' ...;

C:
    prefix=Prefix
    'c' ...;

The Entry rule requires a potentially long lookahead (depending on the Prefix) rule, which can be bad for the parser performance. An alternative is to write it like this:

Entry:
    prefix=Prefix content=Content;

Content:
    A | B | C;

A:
    'a' ...;

B:
    'b' ...;

C:
    'c' ...;

The downside is that the AST then has two separate objects for Entry and Content, while the original version had a single object where Entry is a supertype of A, B and C. New idea:

Entry:
    prefix=Prefix
    (A | B | C);

A:
    'a' ...;

B:
    'b' ...;

C:
    'c' ...;

This would first parse the Prefix and then reassign the object $type to either A, B or C without creating a separate object. We have extracted the common prefix without modifying the AST.

This has limitations:

Foo:
    A B C;

What shall we do here? Parse A, then reassign the type to B, and then again to C? We could constrain unassigned rule calls to a single set of alternatives per rule.

[pluralia]

Currently, there is a workaround for Xtext and Langium:

Entry:
    Prefix (
        {infer A.prefix=current} 'a' ...
    |   {infer B.prefix=current} 'b' ...
    |   {infer C.prefix=current} 'c' ...
    );

Such grammar generates:

export type Entry = A | B | C | Prefix;

export interface A extends AstNode {
    prefix: Prefix
    ...
}

// similar for B and C

This syntax imitate behavior of the proposed syntax, however, it allows to grab the only 1 common element prefix: Prefix. There is no replacement for the case of 2 common elements like here, where A, B and C have properties prefix: Prefix and anotherPrefix: AnotherPrefix.

Entry:
    prefix=Prefix anotherPrefix=AnotherPrefix (A | B | C);