Better Type Unions

[neelsudhakaran]

Discussion

Disclaimer: New Crystal developer, so this issue may lack a deeper understanding of the language. (P.S.—loving it so far!)

• What aspect of the language would you like to see improved?
  Union Types are very wonky at the moment, especially in "complex" data structures.
• What are the reasons?
  If you type a variable as a complex data type—say as an array or a map—you cannot assign it to a "pure valued" data structure even if logically it is valid.
• Include practical examples to illustrate your points.
  The following example wouldn't work because you are trying to assign purely Int32 or Float64 hashes into a union hash. Logically however, all of the below hashes should be valid under the union. The same thing applies for arrays, and probably more data types that I simply haven't tried this with yet.

map : Hash(String, Int32 | Float64) = {
    "A" => 0,
    "B" => 0.5,
    "C" => 1,
    "D" => 1.5
}

map = {
    "A" => 0.1,
    "B" => 0.2,
    "C" => 0.3,
    "D" => 0.4
}

map = {
    "A" => 0,
    "B" => 4,
    "C" => 6,
    "D" => 8
}

• Optionally add one (or more) proposals to improve the current situation.
  I've never coded a compiler before, but perhaps for union types with complex data structures, rather than comparing the raw data type, you do some kind of a recursive check to see that all values within it fit the union type spec? It sounds kind performance costly, I guess, but I presume this is what other languages (e.g. TypeScript) do?

[Blacksmoke16]

You can accomplish this by being more explicit with your hash literal to tell the compiler that it allows values/keys other than those in the literal.

map = {
    "A" => 0.1,
    "B" => 0.2,
    "C" => 0.3,
    "D" => 0.4
} of String => Int32 | Float64

or

map = Hash(String, Int32 | Float64){
    "A" => 0,
    "B" => 4,
    "C" => 6,
    "D" => 8
}

Ref: https://crystal-lang.org/reference/1.7/syntax_and_semantics/literals/hash.html#generic-type-argument

[straight-shoota]

Duplicate of #9797.
The underlying issue isn't actually union types, this is an application for covariance/contravariance which is tracked and explained in #3803. If you have any questions, feel free to ask. But I'm closing this as a duplicate.

[neelsudhakaran]

Thank you! I appreciate the help :)