Literal assertions

[gcnew]

Fixes #14156

Disallow unrelated literal type assertions. E.g.

0 as 0; // OK
0 as 1; // Error

false as (false | string); // OK
false as (true | string);  // Error
false as (boolean | string);  // OK

'hello' as 'str'; // Error
'hello' as 'hello'; // OK
'hello' as ('str' | 123); // Error
'hello' as ('hello' | 123); // OK
'hello' as ('str' & 'hello'); // OK
'hello' as ('str' | 'hello'); // OK
'hello' as (1 | 2 | string); // OK
'hello' as ('str' & { _brand: any }); // Error
'hello' as ('hello' & { _brand: any }); // OK
'hello' as ('str' & { _brand: any } | 1); // Error
'hello' as ('hello' & { _brand: any } | 1); // OK

('string' as string as 'literal'); // OK

[mhegazy]

closing as per #14156 (comment)

[Ghabriel]

Shouldn't this be reopened, considering the discussion at #22038?

[laughinghan]

#14156 (comment) makes sense, but only applies to expressions where there's potentially something the compiler doesn't know. #22038 could be ameliorated with a rule like this that only applies to actual syntactic literals, not just expressions that evaluate to a literal type. In other words:

var x: "asdf" = "asdf";
var y = x as "lol"; // fine, allow this I guess for reasons described in #14156

var z = "asdf" as "lol"; // don't allow this!!!