Hey there, im asking myself if there is a possibil...

# kotlin-native

[Julian Hille]

Hey there, im asking myself if there is a possibility to compile kotlin to objectiv c on linux instead of on macos? So far i did not find any documentation about changing the compiler from c/cpp to objc on the linux target. Is this even possible?

[Big Chungus]

No, apple stuff can only be compiled on apple hardware, unfortunately. Might I suggest using gh actions macos runner for that if you don't gave access to apple hardware?

[Julian Hille]

The idea is not to cross compile on linux for apple, the idea is to use objective as framework and compile on linux a lib for linux

[Big Chungus]

Even so, you need apple hardware. I know that it's possible to cross compile objc and swift, but sadly kotlin native does not support that.

[Julian Hille]

thats what i was trying to get to know. thanks

[Julian Hille]

So there is also no option to force to build the objc files?

[Big Chungus]

Not that I'm aware of, but I'm not an expert in K/N. @kpgalligan might know one if it exists afterall.

[Big Chungus]

Out of curiosity, why are you so set on using objc for Linux?

[Julian Hille]

im searching for a way to use native libs on linux, mac and windows and use suspend fun on each platform

[Julian Hille]

wasm is not an option to compile to as the libs i have as depenencies do not support that. JS is no a good option. native to c / cpp does not support suspend funcs

[Julian Hille]

thats why i try to see if there is an option to use objc on all platforms

[Big Chungus]

One option could be to expose non-suspend functions to c that invoke suspensions on its behalf

[Julian Hille]

yes indeed. i could wrap all the funcs into runBlocking and throw a naming convention on them, that would at least export them in the lib / header files

[Big Chungus]

Not saying it would be pretty nor pleasant 😀

[Julian Hille]

but im afraid / unsure of the performance impact if i remove all the threading and multiprocessing and do all the io bound stuff sync

[Big Chungus]

You can avoid runBlocking by having a dedicated custom coroutine scope for these functions to use

[Julian Hille]

basically im tryng to get all the options onto the table to make a profound decision on which route to go. The objc option is still open

[Julian Hille]

not sure if that would work, as no suspend function is exported in the native shared or static lib

[Big Chungus]

I'm not sure how concurrency works in c, but since coroutines are just a callback loop, you might be able to wrap c concurrency primitives into coroutines via suspendingFunction{}

[Big Chungus]

not sure if that would work, as no suspend function is exported in the native shared or static lib

You write non_suspend kotlin functions that enter suspend scope via that custom scope and invoke appropriate fun on c behalf

[Julian Hille]

to use the function at all they are not allowed to be suspend func

[Julian Hille]

yes. isnt that what runBlocking essentially does?

[Big Chungus]

You then return a wrapper to whatever c uses for async/deferred values

[Big Chungus]

No, run blocking blocks until the value is ready. I'm suggesting to return some deferred c value if it has something like that (not familiar with c concurrency so I'm just guessing here)

[Big Chungus]

Basically c equivalent to Promise in js

[Julian Hille]

i guess if that would be possible kotlin lang would have done it 😛

[Big Chungus]

Surely c must have something for concurrency

[Julian Hille]

https://github.com/Kotlin/kotlinx.coroutines/issues/3296

[Big Chungus]

Though luck https://github.com/Kotlin/kotlinx.coroutines/issues/3296#issuecomment-1132020716

[kpgalligan]

im searching for a way to use native libs on linux, mac and windows and use suspend fun on each platform

We don’t use the suspend functions on Objc. In fact, I’m going to push for a way to suppress generating them. They aren’t really “suspend” functions. They also wrap the suspend calls with a callback mechanism, but it’s done by the compiler rather than you. If they work with Swift async/await, its because they conform to a specific convention, but they aren’t, strictly speaking, “suspend functions”. We generate our own interface, or suggest https://github.com/rickclephas/KMP-NativeCoroutines. That won’t work in your case, as I assume https://github.com/rickclephas/KMP-NativeCoroutines doesn’t support linux C output (correct me if I’m wrong @Rick Clephas)

[kpgalligan]

We don’t use the suspend for objc because you can’t control the lifecycle, and there are other restrictions (need to call on the main thread, can’t do Flow regardless). Once you get past the boilerplate, controlling your own interface is better (IMO)

[kpgalligan]

Having said all that, you can compile objc on linux (I believe), but the Kotlin compiler only offers objc for Framework output, and that’s very much tied to Apple/Darwin platforms unfortunately. I was hacking around trying to build a static lib, then see if we could use objc headers, but you simply can’t.

[kpgalligan]

So, summary, since the objc is just callbacks, you should just do the same with the C that’s output for Linux. As there’s less Linux support for these kinds of libraries, you’ll need to do that yourself, but It’s not super complicated. I was going to post an example, but our sample app seems to have grown more layers, so I’d need to find a simpler one…

[Julian Hille]

that would be great. i;d love to see a small sample to get a grasp on it

[Julian Hille]

thanks for all the deep insights which would take some days for me to get my head around

[Rick Clephas]

Correct currently KMP-NativeCoroutines won’t do anything on a linux target (only on Apple targets). However the code generation is all Kotlin, so it could easily be enabled for other targets as well.
The only thing needed for that would be another way to map `Exception`s (which are currently mapped to `NSError`s).

[Rick Clephas]

In fact, I’m going to push for a way to suppress generating them.

I have actually been working on that (for ObjC/Swift): https://github.com/JetBrains/kotlin/pull/4818

[Julian Hille]

interesting will take a look

[kpgalligan]

That PR definitely looks interesting. Will dive in. Reading it is giving me flashbacks: https://github.com/JetBrains/kotlin-native/pull/2850

[kpgalligan]

The concept is interesting. I’d also like to be able to pass a compiler arg that basically suppresses all suspend call generation, but an annotation will work as well, and will allow us to be more precise. We’re currently going deep on Swift code generation and renaming/removing methods from the interface was on the roadmap if possible.

[Rick Clephas]

Compiler arg would also be nice indeed! For now I focused on the annotation as it will help with more use cases/limitations (and I kind of want the annotation for KMP-NativeCoroutines 🙈).

[kpgalligan]

Not exactly related, but I want to get the name map out of the compiler and/or be able to control names rather than the auto-mangling that happens for name collisions. Anyway, pretty interesting. Swift private with __ is new to me, but that’s great.

[kpgalligan]

I’m giving a talk about this soon. Just gave one that was less detailed. Essentially efforts to shape the api surface. Not being able to selectively remove methods has been a real pain point.

[Rick Clephas]

😇 https://github.com/JetBrains/kotlin/pull/4815

[Rick Clephas]

Cool! Any more details about that talk? Like where and when 😁

[kpgalligan]

That’s cool! We started talking about some of this stuff a few years ago, but adding annotations to Kotlin seemed like it would be rather rare. Controlling the name is great.

[kpgalligan]

For Swift gen, we really need a way to pull the names out too, but that’s a whole different discussion.

[Rick Clephas]

Ah you mean a predictable way to know the Swift name of a Kotlin declaration from a compiler plugin or something?

[Rick Clephas]

About the mangling. I guess there should be a compiler arg to make any mangling an error (as soon as you can controle the names of course).

[kpgalligan]

Yeah, so if we generate swift code, what’s the Swift name of the Kotlin thing. Obviously its usually the same, but not for dependencies (unless you export them, which is a whole different discussion), and not if you have name collisions. Being able to override specific classes is good, as you can work around the naming collisions.

[kpgalligan]

For now, our plan is to just provide a way to config those names if they don’t align. We’re still sort of in the “explore” phase on Swift code gen. I can see where it might be useful, but I want to get feedback from folks doing it before we jump in with both feet. Trying to poke a hole into the compiler to pull that map, or have the compiler write that map, etc, is a bigger change, so for now it’s more of a hard-coded workaround.

[kpgalligan]

Mangling as an error definitely makes sense to me.

[Rick Clephas]

Got it! Yeah let me know if you have any thought on those PRs.

[kpgalligan]

We’re all poking around them now…

[kpgalligan]

On the talk, the short version I gave was more complaining about the problem, longer version will be Droidcon Berlin (if it’s accepted). I was about to tweet about the difficulty of writing talks on topics that are actively changing, and your PRs are an example 🙂 Part of my recent talk was about compromises in Kermit to support a better Swift api, but after the talk I basically rewrote it (https://github.com/touchlab/Kermit/tree/kpg/api_reformat2), and if we can figure out the Swift gen (with klibs in the middle), then everything from that talk will be old news, and I only gave it a few weeks ago.

[kpgalligan]

I have to find something to talk about now that the memory model has changed…