Is there anyone with a deep knowledge of Kotlin generics who kotlinlang #announcements

Is there anyone with a deep knowledge of Kotlin ge...

dmcg

03/27/2019, 10:32 AM

Is there anyone with a deep knowledge of Kotlin generics who could help me out with the API for #CCYE00YM6? I’m trying to find the least worst solution to a co/contravariance conundrum.

robstoll

03/27/2019, 10:39 AM

what's your problem?

dmcg

03/27/2019, 10:42 AM

I want to annotate nodes in a tree with transforms that will be applied to that node.

dmcg

03/27/2019, 10:43 AM

The NodeTransform is https://github.com/dmcg/minutest/blob/14365f101a74b805ea0d406c0fd28968f4b3511d/core/src/main/kotlin/dev/minutest/NodeTransform.kt - invariant in T

robstoll

03/27/2019, 10:44 AM

do you modify the node during transformation or do you create a new node? If you don't change the node, then use can use site variance =>

next: (NodeTransform<out F>

dmcg

03/27/2019, 10:45 AM

I don't change the node

dmcg

03/27/2019, 10:45 AM

I think I did try site variance, but maybe not there!

dmcg

03/27/2019, 10:45 AM

The annotations need to be in

interface TestAnnotation<in F>

dmcg

03/27/2019, 10:46 AM

as the transform can only apply to more general types

robstoll

03/27/2019, 10:47 AM

I guess you could even use

out

for NodeTransform if they don't modify

robstoll

03/27/2019, 10:48 AM

or what is the reason that made

invariant?

dmcg

03/27/2019, 10:48 AM

Is this something that you have time to screen share over - I can give you a tour?

robstoll

03/27/2019, 10:50 AM

nah unfortunately I don't (at least not now)

dmcg

03/27/2019, 10:52 AM

I'm in no rush. If you can pick a ZonedDateTime I can fit around you and make sure that I'm up to speed on what has worked and what hasn't

robstoll

03/27/2019, 10:53 AM

OK, I see and let you know. In the meantime, could you show me a use case where you are using the transformation

dmcg

03/27/2019, 10:59 AM

That's grand. You could look at https://github.com/dmcg/minutest/blob/master/core/src/test/kotlin/dev/minutest/examples/experimental/SkipAndFocusExampleTests.kt - that shows the test annotations SKIP and FOCUS. The essential issue is reconciling the contravariant TestAnnotation with the invariant NodeTransform.

dmcg

03/27/2019, 11:00 AM

https://github.com/dmcg/minutest/blob/master/core/src/main/kotlin/dev/minutest/experimental/TransformingAnnotation.kt is where my lack of prowess hits the fan.

robstoll

03/27/2019, 11:02 AM

is the type of the node of any interest during transformation?

robstoll

03/27/2019, 11:02 AM

the F in https://github.com/dmcg/minutest/blob/master/core/src/main/kotlin/dev/minutest/experimental/TestAnnotation.kt#L12

dmcg

03/27/2019, 11:04 AM

I would be nice - I have places like https://github.com/dmcg/minutest/blob/master/core/src/main/kotlin/dev/minutest/experimental/flattening.kt where knowing the type allows some clever transformations

robstoll

03/27/2019, 11:12 AM

just asking, sometimes we make the problems harder then they need to be 😄

dmcg

03/27/2019, 11:13 AM

That sounds like my career

dmcg

03/27/2019, 11:14 AM

I wonder if I should start from

interface NodeTransform<F, F2: F>

robstoll

03/27/2019, 11:15 AM

(node as? Context<Sequence<F>, F>)

does not seem like you really use F at all. Or am I wrong?

robstoll

03/27/2019, 11:16 AM

meaning you could replace

with

and have the same functionality

dmcg

03/27/2019, 11:17 AM

Probably true there - but it's the Sequence<F> that corresponds to the F in the NodeTransform

robstoll

03/27/2019, 11:18 AM

Well, theoretically but you are using an unchecked cast which means

could also be something completely different

dmcg

03/27/2019, 11:19 AM

That's why it's private within

TestContextBuilder<Sequence<F>, F>.flatten()

- it's safe there

dmcg

03/27/2019, 11:20 AM

F isn't important in the flattening, what is important is that the Context has a parent fixture of Sequence<F> and yields individual Fs

robstoll

03/27/2019, 11:21 AM

but you don't know this. if you cast to

Context<Sequence<F>, F>

you only know that it is

Context

but it could also be

Context<Int, String>

and would still succeed

dmcg

03/27/2019, 11:21 AM

But you're asking excellent questions - definitely the person for the job

robstoll

03/27/2019, 11:23 AM

meaning it would fail at runtime

robstoll

03/27/2019, 11:24 AM

and thus you could also remove the type parameter all together, you don't get any additional safety like this

dmcg

03/27/2019, 11:24 AM

Ah yes, another logical step -

TestContextBuilder<Sequence<F>, F>

builds the Context

dmcg

03/27/2019, 11:24 AM

that the transform is applied to

dmcg

03/27/2019, 11:25 AM

So within that extension method we do know that the type of the Context is Context<Sequence<F>>, F>

dmcg

03/27/2019, 11:28 AM

https://github.com/dmcg/minutest/blob/master/core/src/test/kotlin/dev/minutest/experimental/AnnotationVarianceTests.kt shows some of my experiments in the area

robstoll

03/27/2019, 11:35 AM

I would try to get rid of all UNCHECKED_CAST. If you cannot, then there is no point of having a type parameter in the first place

robstoll

03/27/2019, 11:35 AM

I'll see, maybe I get some time off in the afternoon so that I can have a closer look

dmcg

03/27/2019, 11:36 AM

I'm trying 😉

dmcg

03/27/2019, 11:37 AM

Really appreciate it. I work from home, so ping me any time that's convenient

2 Views

Open in Slack

Previous Next