Shorthand nullable-type syntax

[SomeRanDev]

Allow T? as shorthand for Null<T>.

@:nullSafety(Strict) {
    var MyList: Array<Int?>? = null;
    MyList = [20, null, 32];
}

Rendered version

[nadako]

This looks fine to me and I agree with the motivation.

Originally I was against and I'm still a bit undecided regarding adding syntax sugar for nullable types, because I believe that usage of nullables should generally be minimized. However the current situation is that having Null<T> even where it makes sense is so tedious (both to write and read) that people would rather not use null-safety feature at all, so my gut feeling today is that making this more concise will actually help adoption of null-safety, which in turn will make people think more about what is nullable and what is not, which will minimize the unnecessary nullability. :)

Regarding type-as-expression, I think it's the same situation as with the type params - we just don't parse that at all currently, so ? would be a ternary there. This might conflict with the potential EComplexType proposal, if we ever reconsider it, but it sounds like it could be solved with some parenthesis.

[Gama11]

I'm generally in favor of this since it would reduce the visual clutter when working with null safety by a lot.

However, I wonder about the "consequences" considering the the existing optional argument and object field syntax. The following would all be possible, but look confusing:

function foo(?i:Int?) {}

typedef Foo = {
    var ?i:Int?;
}

typedef Foo = {
    ?i:Int?
}

Putting a ? on the type in these cases is redundant, since it's already implied by the optionality of the argument / field. That means it would probably be considered bad style / should maybe be grayed out by IDE diagnostics? In Either case, it probably couldn't be disallowed entirely.

Similar to other languages, there can be whitespace between the type and the ?; however, no whitespace should be the standard

I think we should consider simply forbidding to put whitespace there, similar to the var ?name syntax.

[nadako]

Regarding the proposal, maybe it's worth showing some awkward possible situation with implicit Null<T> for optional arguments and fields, e.g. function f(?a:Int) and {?a:Int} - this is a place where "Nullable reduncancy" can appear, along with the slight syntatical weirdness if you do function f(?a:Int?) or {?a:Int?}. Also maybe add a note that the printing of Null<T> in compiler messages should probably be consistent and use the new syntax with the question mark as well, but also be careful about printing optional argument and fields to maybe avoid adding ? to the type there...

[RealyUniqueName]

I think we can emit an error if a redundant ? is spotted on parsing. And just ignore/merge it if Null<Null< is the result of typing or inlining.

[Gama11]

@RealyUniqueName Which redundancy are you talking about? The optional ? with Type? or the Type?? mentioned in the proposal?

[SomeRanDev]

Thanks for the quick feedback!

For starters, the "type-as-expression" was a bit ambiguous when I wrote it due to lack of in-depth compiler knowledge, but I was also thinking about parenthesis as a solution. I'll go ahead and remove some of the uncertainty in the section and commit to explicitness through either parenthesis or Null<T> there.

For the whitespace, I just committed to allowing it since both C# and Kotlin allow it, but quite frankly it looks horrible, even with just a single space and would actually much prefer it force no whitespace. Would be more than happy to change that as well.

In regards to printing, do you mean there should be some differentiation between Null<T> and T? and the output should reflect which style is used, or should all output of Null<T> print with the T? syntax?

And finally, in regards to {?a:Int?}, I'll go ahead and add that to "Drawbacks" (or perhaps it's something to be resolved within "Detailed design"?) Either way, don't know if I'm involved enough with Haxe design to answer that one. Would there be a reason to do it in the first place? ~Do null-safety features not trigger with {?a:Int} alone?~ They do; so I guess if there really is no difference, just throwing a warning/error would be fine, right?

[nadako]

In regards to printing, do you mean there should be some differentiation between Null<T> and T? and the output should reflect which style is used, or should all output of Null<T> print with the T? syntax?

The latter. We don't retain the details of the original syntax anyway, especially in the typed tree.

[nadako]

Regarding {?a:Int?}: while it looks funny, I don't consider it particularly confusing to read, and this is really a rare edge case as there's no need to write like that, since optionality already implies nullability in these situations. What I'm thinking is that when the compiler prints such a type in e.g. its error messages maybe some care needs to be taken to print {?a:Int} instead of {a?:Int?} :)

[Simn]

I don't like how this looks. IMO Null<T> is actually more readable, especially if we start nesting things.

I acknowledge that this is easier to type, but in that case it should be solved at IDE-level.

[back2dos]

Since we already have ?Type (via ComplexType.TOptional) I wonder why we should have Type? as well.

[kaikoga]

TL;DR: While I'm not against Null<T>s are treated special and granted a T? syntax, I don't expect this proposal would come true.

---

Null<T> (and possibly T?) is clearly a valid Haxe type. And surely ?T is also a valid Haxe type syntax, but in reality it is something that works outside "type" context.

I'm afraid to say that we don't have optional types but actually optional arguments/fields that are coincidentally marked in ComplexTypes (because things like TFunction(args:Array<ComplexType>, ret:ComplexType) and TField relies on ComplexType to carry information about whether it is optional or not). Proof: We can turn first arg of Int->Void into optional by writing ?Int->Void. So ?Int looks like an optional type. But we cannot take (a:Int)->Void and turn it into (a:?Int)->Void. We must write (?a:Int)->Void, or the compiler complains Optional type not allowed here. While both ?Int->Void and (?a:Int)->Void are valid Haxe types, there is no convincing reason (except compiler implementation detail) that we should wrap the optional named type as TOptional(TNamed(_)) in this exact order. And we cannot write ?Type anywhere else, which gives a question whether ?Type is a first class type.

Not to mention that optional and nullable works differently (i.e. optional args can be skipped but nullable args cannot), I think we should at least try to not mix up optional and nullable, they are meant to work in a completely different layer. TypeScript is one example which has optional properties and arguments, just like we do, but (unsurprisingly) lacks shorthand nullable-type syntax.

And I'm sorry if I summoned Nic's veto, as this state would confuse beginners even more (hey I've just defined a typedef with nullable field in var notation, wait where should I put the ? to make that field optional?).

Summary: We should first clean up and break optional, which would not happen, before making another use of the ? character, I regret.

P.S. I personally welcome a haxe-evolution that deletes the ternary (I don't mind typing if (a) b else c).

[back2dos]

We must write (?a:Int)->Void, or the compiler complains Optional type not allowed here.

That's circular reasoning. For the parser (a:?Int)->Void is perfectly valid syntax already and the typer may well interpret it as (a:Null<Int>)->Void instead of emitting said error.

For the legacy function syntax, there's indeed some ambiguity, as ?Int->Void makes the argument not just nullable but also optional. It should remain as is, since fans of legacy syntax can (and for reasons of consistency should!) still use the old school Null<Int>->Void if they want nullable mandatory args. When using the new function type syntax, users can explicitly distinguish between mere nullability and full optionality via ?-placement.

Maybe postfix ? is better still, but adding syntax often means a breaking change in the macro API. So if there's a sane way to reuse existing syntax, it's worth considering, because it makes it easier to move forward.

P.S. I personally welcome a haxe-evolution that deletes the ternary (I don't mind typing if (a) b else c).

Many (most?) people would, because it often stands in the way of adding sane syntax. Unfortunately it's quite widely used, so it's doubtful such a breaking change will ever come ;)

[Simn]

We have rejected the proposal in the haxe-evolution meeting today.

The syntax has no clear advantage other than potentially being shorter. There were also concerns about possible syntactic ambiguities that could arise from this.