Swift 5.5 including async/await.
C# Support it.
TypeScript support it.
Kotlin support it.

On Friday, 24 September 2021 at 03:16:36 UTC, zoujiaqing wrote:

And Nim, and Rust, and Zig. I think of it as a lot like garbage collection:

1. deadlocks (memory errors) are too hard to deal with! Why can't software handle this instead of us poor programmers?

2. async funcs (GC-using functions) are cleaner and easier to write. Don't mind the invisible extra work.

3. huh, this new cure-all comes with surprising downsides like 'colored functions' (GC-assuming APIs) that are annoying to expose to the C ABI, and that want other functions to share their 'color' (also use/avoid GC), and extra overhead, and--

Past the trivial client-side examples and synchronous callback hell, you start caring about the actual concurrent architecture of your program, but you no longer have any control to architect anything with. That's my limited experience, anyway. If you know of a good salespitch, "Erlang devs! Stop letting mailboxes hold you back! Join the new order!", I'll watch it.

Meanwhile:

https://code.dlang.org/packages/dawait

That's using core.thread.fiber directly, but by removing that you could make it compatible with a simple import vibe.core.concurrency; and then setConcurrencyPrimitive(ConcurrencyPrimitive.workerTask); to use vibe-core's fiber scheduler.

On Friday, 24 September 2021 at 03:16:36 UTC, zoujiaqing wrote:

Python and C# went the generator => async route and if the compiler can automatically rewrite your code to state machines async becomes a runtime feature.

That rewriting (ala C++ coroutines?) is the useful/enabling thing I'd see usefull for D. Would wonderfully interop with ranges as well.

On Friday, 24 September 2021 at 06:00:30 UTC, jfondren wrote:

adding vibe as dependency = make your project much much slower to compile

dependencies are bad by design, you don't want to depend on 3rd party, you want use language to empower your project

dependencies should serve to play nice with services, for ex: postgresql driver, gpu abstraction, aws api etc etc

if people define language constructs as "dependencies", consuming them feels bad, because there will not be a common denominator between "dependencies"

you'll end up like Rust, where certain io operation will require different paradigms because it is implemented as crate

then you have crates that depends on crates that depends on crates etc

you end up watching crates downloads for 5 minutes when you want to "try the language"

same story for typescript and NPM, even though they have async support already, other aspects are affected

Go doesn't have this problem, because goroutine is a language feature

One must watch other languages and learn from their successes and avoid mistakes of others

If your language can't do X, ask why it can't, if it could be doable with language, and then, and only then you can start to workaround the problem by consuming a dependency, but even then, one must always remember that it'll feel bad to depend on a dependency

On Friday, 24 September 2021 at 12:53:40 UTC, russhy wrote:

The X that I'd like is faster compilation of vibe and std.regex and such :)

VibedScheduler is just a much better fiber scheduler than the deliberate placeholder implementation in Phobos. There's still language support, with shared, immutable, thread-local storage, etc. I keep looking for this information in the spec but https://tour.dlang.org/ shows it off in the Multithreading drop-down.

On Friday, 24 September 2021 at 12:53:40 UTC, russhy wrote:

It seems like this problem (dependencies leading to ecosystem fragmentation) could be solved by a good standard library module, without necessarily requiring a new language construct.

On Friday, 24 September 2021 at 15:30:19 UTC, Paul Backus wrote:

Let's call it std.experimental.xxx where for some reason it goes there only to die.