Null-Restricted and Nullable Types

[forax at univ-mlv.fr]

> From: "Brian Goetz" <brian.goetz at oracle.com>
> To: "Remi Forax" <forax at univ-mlv.fr>
> Cc: "valhalla-spec-experts" <valhalla-spec-experts at openjdk.java.net>
> Sent: Wednesday, September 4, 2024 4:44:05 PM
> Subject: Re: Null-Restricted and Nullable Types

> I appreciate the attempt to avoid the rabbit hole -- but there are just so many
> rabbits! But let me try to separate the layers. Each of these is a substantial
> discussion, and we're still gathering our thoughts on most of them.

> We have to be very careful to separate the notations we use for discussing the
> semantics from notations that are being seriously proposed as part of the user
> model.
yes, this distinction is very important. 

> Let's assume we are a long way from talking about what the surface syntax might
> look like.

> For a use of a type variable, there are four possible ways to combine a type
> variable with nullity information. Let's call "the argument type" the thing the
> user put in the pointy brackets.

> - T! -- the argument type, but null-restricted if it wasn't already
> - T? -- the argument type, but explicitly null-allowing if it was
> null-restricted
> - T* -- the argument type
> - T~ -- the argument type, but of *unknown* nullability

> I think we can agree we don't want to burden users with all of these fine
> distinctions; there is a discussion to be had about "what does bare T mean",
> but we should wait.

> The reason we even talk about T* vs T~ is because of _migration_. And, unlike
> the migration from non-generic to generic, we have to deal with three separate
> migration scenarios:

> - the client wants to use null-marked parameterizations (Foo<String!>), but the
> class has not yet considered nullity restrictions (legacy class)
> - the class wants to use null markings in its API, but the client is not
> prepared to deal with nullity restrictions (legacy client)
> - both client and class are null-aware

> So T~ is the type variable equivalent of "I have some old pre-nullity code, I
> have no idea what it thinks the nullity of this type is".
yes, and I think it can be useful to see T~ and String (raw) the same way, because it makes legacy class code and legacy client code the same thingy. 

> We of course want to get to the point where we get maximal type checking in the
> last stage, but we don't want to put impediments in people's way, and are aware
> that they may take either path.
The drawback of letting too many possible migration paths to the user is that the majority of the users may never reach the maximal type checking. 
If you have several stages, it's not obvious that moving to stage 0 to stage 1, and from stage 1 to stage 2 is easier than moving from stage 0 to stage 2 in one step, this is a classical problem with gradual typing. 

> There are many tradeoffs here about "more restrictions but sounder type
> checking" vs "lenient but more chances for runtime errors."
For me, this is another axis. I think here, we have not a lot of leeway, given we want to support a world with both pre-nullity code and post-nullity code, we have to have null pollution warnings. 
In most cases, this is okay because the pollution does not go far, it stops that the next function call (intra-function). Casting type variables or parametrized types sadly may have inter-functions side effects. 

Rémi 

> On 9/4/2024 10:06 AM, [ mailto:forax at univ-mlv.fr | forax at univ-mlv.fr ] wrote:

>>> From: "Brian Goetz" [ mailto:brian.goetz at oracle.com | <brian.goetz at oracle.com> ]
>>> To: "Remi Forax" [ mailto:forax at univ-mlv.fr | <forax at univ-mlv.fr> ]
>>> Cc: "valhalla-spec-experts" [ mailto:valhalla-spec-experts at openjdk.java.net |
>>> <valhalla-spec-experts at openjdk.java.net> ]
>>> Sent: Wednesday, September 4, 2024 3:26:01 PM
>>> Subject: Re: Null-Restricted and Nullable Types

>>> Let's take a step back. Are you looking for *answers* in the short term (say, so
>>> you can answer someone elses questions), or are you looking to open the dialog
>>> about how we will expose nullness for purposes of generics migration? Because
>>> there are many, many pages of things to say about these questions, not all of
>>> which we have answers to, and this will significantly affect how we structure
>>> the discussion.
>> I'm opening the discussion. Those questions are like anchors to me, to try to
>> avoid to go too deep into one of the rabbit holes.

>> Rémi

>>> On 9/4/2024 8:36 AM, [ mailto:forax at univ-mlv.fr | forax at univ-mlv.fr ] wrote:

>>>> Here are my questions:

>>>> 1/ Nullness of type variable
>>>> "Like other types, a type-variable type (that is, a use of a type variable) may
>>>> express nullness. T! is a null-restricted type, and T? is a nullable type."
>>>> - what does 'T' exactly mean ?
>>>> For me, there is the old 'T' which propagate the type but not propagate the
>>>> nullability and there is the new 'T' that propagate both the type and the
>>>> nullability of the type.
>>>> During the JVMLS, Dan Smith references them as 'T' and 'T*'.

>>>> 2/ If we agree that there are 4 different kinds: String/T, String?/T?,
>>>> String!/T! and T*, we now have the choice of several user facing models
>>>> - model 1, we let the user annotate using '!', '?' and '*'
>>>> - model 2, we have an opt-in mechanism that set type as '!' by default and type
>>>> variable as '*' by default.
>>>> - model 3, we only let user to annotate type variable declaration, type of
>>>> field, type of parameter types/return type, parametrized type (inside the angle
>>>> brackets) and cast, the rest is inferred
>>>> (this is the semantics of jspecify).

>>>> 3/ Nullness of wildcards
>>>> "A type variable declaration or wildcard may have nullness markers on its
>>>> bounds. A type may satisfy the bounds via nullness conversion, though, so again
>>>> these nullness markers are not strongly enforced, but may cause warnings."

>>>> => I see two questions here :
>>>> - unbounded wildcards, they are special because they represent a reified type,
>>>> so ? is nullable in List<?> because list.add(null) is valid at runtime, thus
>>>> List<?> is equivalent to List<? extends Object?>.
>>>> Do you agree ?
>>>> - super wildcard, the content of List<? super Foo!> may accept null because Foo?
>>>> is a super type of Foo!, so List<? super Foo!> is maybe nullable or maybe not,
>>>> so the content is neither a '!' nor a '?'. Do you agree ?

>>>> Rémi

>>>>> From: "Brian Goetz" [ mailto:brian.goetz at oracle.com | <brian.goetz at oracle.com> ]
>>>>> To: "Remi Forax" [ mailto:forax at univ-mlv.fr | <forax at univ-mlv.fr> ] ,
>>>>> "valhalla-spec-experts" [ mailto:valhalla-spec-experts at openjdk.java.net |
>>>>> <valhalla-spec-experts at openjdk.java.net> ]
>>>>> Sent: Tuesday, September 3, 2024 10:44:03 PM
>>>>> Subject: Re: Null-Restricted and Nullable Types

>>>>> There are many points that are not clear to us as well, but we will try to
>>>>> respond to your questions :)

>>>>> On 9/3/2024 4:41 PM, Remi Forax wrote:

>>>>>> Hello,
>>>>>> if everybody is okay with that i would like to discuss about Null-Restricted and
>>>>>> Nullable Types [1] tomorrow.

>>>>>> There are several points that are not clear to me, i will try to come up with a
>>>>>> list for tomorrow.

>>>>>> Rémi

>>>>>> [1] [ https://openjdk.org/jeps/8303099 | https://openjdk.org/jeps/8303099 ]
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://mail.openjdk.org/pipermail/valhalla-spec-observers/attachments/20240904/3492cd19/attachment-0001.htm>