Type predicates with regex are unsound

[blutorange]

Type predicates generated by langium that include a regex are unsound. This is because type predicate are an if-and-only-if test: it returns true if and only if the value is of the given type. But the regex is not included in the TypeScript type. See e.g. microsoft/TypeScript#15048.

A type guard such as

export type A = string;
export function isA(item: unknown): item is A {
    return (typeof item === 'string' && (/[a-z]+/.test(item)));
}

will return false even when the item is in fact an A (as far as TypeScript is concerned), e.g. isA("?").

Possible solutions

Hard to solve without breaking compatibility. I didn't create an issue for now since I don't think anything can/will be done currently. I mainly wanted to point this out for now.

In principle, solutions could be to (a) drop the regex check or (b) drop the type predicate and make the return type only boolean, or (c) generate both (a) and (b).

Reproduction

Consider this grammar

grammar Foo
entry Entry: value=A | value=B;
terminal ID returns string: /[a-z]+/;
terminal INT returns string: /[0-9]+/;
A returns string: ID;
B returns number: INT;

This results in:

export interface Entry extends langium.AstNode {
    readonly $type: 'Entry';
    value: A | B;
}
export type A = string;
export function isA(item: unknown): item is A {
    return (typeof item === 'string' && (/[a-z]+/.test(item)));
}
export type B = number;
export function isB(item: unknown): item is B {
    return typeof item === 'number';
}

Now with these types, the following program type checks, but produces a runtime error due to mismatching types:

function processEntry(entry: Entry): string {
  if (isA(entry.value)) {
    return entry.value.toUpperCase();
  } else {
    return entry.value.toFixed(2);
  }
}

processEntry({
  $type: "Entry",
  value: 999
});

processEntry({
  $type: "Entry",
  value: "___"
});