Shouldn't we permit NamedTypes without typechecking?

[JoshuaSCochrane]

According to the readme: "The built-in types String, Bool, Number, Elem, and Event are also supported. Additional user-defined types can be used by name (e.g. User or Post). Such types are treated opaquely by the type checker and will not check the fields of the object."

But we throw an exception for typechecking when we encounter user-provided NamedTypes - shouldn't we just return true in that case? Or have some mechanism to register type checkers? (one could dynamically append a key to runtimeCheckers with their own type I suppose..., e.g. runtimeCheckers["User"] = v => true;, but I'm not sure that's what's intended here)

https://github.com/efritz/arrows/blob/13be431cebd7681717b9390a3b4609ed54b4feb2/src/types.js#L104-L112

[efritz]

Now that I see it, the following is pretty suspicious as well.

https://github.com/efritz/arrows/blob/13be431cebd7681717b9390a3b4609ed54b4feb2/src/types.js#L130

Can you come up with a small example where the types should be equal (as they have the same name) but have no checker - something that doesn't work in practice? I guess I've not done these in practice very frequently (or recently).

[JoshuaSCochrane]

Well if the checker returns false, super.check(value) will simply throw a TypeClash: https://github.com/efritz/arrows/blob/13be431cebd7681717b9390a3b4609ed54b4feb2/src/types.js#L6-L8
Which was perhaps the intent - perhaps a bit roundabout (vs just throwing a TypeClash if that's what needs to happen or calling some other shared func specifically to report the error...) but I think that's ok as-is.

I ran into the associated checker error when using any custom named/opaque types (I think) - right now I'm hitting it when trying to sequence arrows with types _ ~> User and User ~> User - so the names match, but it doesn't have an associated checker for User, so it throws an exception... unless you're saying it shouldn't even get here, which may be a different problem (though I think I pulled in your recent fixes) - it's possible I've changed something somewhere though...

[JoshuaSCochrane]

So if I dynamically add a runtime checker, e.g. with runtimeCheckers["User"] = v => true;, it lets it pass through... which it turns out I can do in the nodejs stuff if I export runtimeCheckers - not so bad, but I think not intentionally necessary...

[JoshuaSCochrane]

Oh, and in this case User is actually a class created for this example code, not just a MadeUpTypeName or something... not sure if that matters or not

[efritz]

Oops, I was looking at the equals() not check() method in the base class. So the idea was to match the annotation of a named type with a value so you'd do something like this in pseudo-typescript:

runtimeCheckers['User'] = u => type(u) == 'User';

I don't think values are checked otherwise for named types, which means anything can come back from an ajax request as long as you say it's some opaque non-checked type.

[JoshuaSCochrane]

Sure. I confirmed this as a dynamic checker works too: runtimeCheckers["User"] = v => v instanceof User;

[JoshuaSCochrane]

But essentially, the readme says anything that's not a String, Bool, Number, Elem, or Event won't be checked, but it is checking other things and throwing exceptions if it doesn't understand them...

[JoshuaSCochrane]

Although "Such types are treated opaquely by the type checker and will not check the fields of the object." could also mean if you tell me it's a User, I'll check that but only that - I'll confirm it's actually an instance of User, but I won't dig into any it's fields. I took it to mean "If you tell me it's a User, I'll trust you - no questions asked." Perhaps I misunderstood the intent. In that case, it would still be neat to automatically check user-provided types to see if they match a user-defined class, but I'm not sure offhand how to make that work... it's very probably possible ;)

[efritz]

I'd be fine having something to populate the runtime checkers with class instances, but this won't work in vanilla javascript.

I'll update the readme to clarify when I have more time. I'll leave this issue open until then.