Remove special case for invariant

[petehunt]

I'm finding it difficult to add type annotations to my function named invariant() because of this little guy:

flow/src/typing/type_inference_js.ml

Line 2058 in 0f56d17

let msg = "unsupported arguments in call to invariant()" in

It seems that many of these functions are special cased. I'm curious as to why, because I can't use any functions that share these names anymore.

[avikchaudhuri]

Eventually the type system will be powerful enough to express the "signatures" of these functions, at which point their names will not need to be special-cased any more (since they can simply be declared to satisfy the necessary signatures).

In the short-term, we'll try to scope these names somewhat, so that they don't interfere with other things with the same names.

[samwgoldman]

The proposal for refinement carrying boolean types in #34 isn't powerful enough, AFAICT, to express the type of invariant. Since TypeScript has already implemented that, it's probably a good time to start thinking about how we'd express something like invariant in the type language.

[leebyron]

Here's a rough working but not long-term stable proposal:

invariant(predicate: any): $Asserts<predicate is true>

This uses the Ident "is" Type grammar proposed in #34. It uses @mroch's recent addition of boolean literals as Types. And it uses Flow's "this is an uncommon and undocumented" pattern of $Thing for communicating specific typing behavior.

[samwgoldman]

I like this syntax and I think the syntax does communicate all the necessary information. I feel like the implementation might get tricky around the $Assert type, which should result in an abnormal event. I haven't looked into this enough yet to say whether it's realistic or not (maybe @avikchaudhuri can weigh in), but I'll give this some more thought.

[mroch]

3e35bd4 should get the rest of the static_upstream/core special cases out of the way, except for invariant. Renamed this task to preserve the discussion of predicate types.

[gajus]

invariant(predicate: any): $Asserts<predicate is true>

Was this ever implemented?

[leebyron]

No, %checks was built which is for predicate functions but not assertion functions. If this is a model the flow team is happy with then maybe something like this could be built such as %asserts?

[gajus]

For others reference, here is the documentation for %checks https://flow.org/en/docs/types/functions/#toc-predicate-functions.

[gajus]

This issue is duplicate of #34.

[mattconde]

So I think I've caught up on this issue, am I right to think I still can't flowtype my own "invariant" module correctly because of this?

[goodmind]

TypeScript now has this microsoft/TypeScript#32695