[dagmx]

I have such a love hate relationship with Swift. It genuinely has some great ergonomics and I see it as a sister language to rust.

I just wish it was more cross platform (I know it technically works on Linux and windows…but it’s not a great experience) and that it didn’t have so much churn (though they’ve stopped futzing with the core components as much with Swift 4+).

I also wish it was faster. I did an Advent of Code against some friends. I picked Rust, they picked Swift. The rust code was running circles around their Swift ones even when we tried to keep implementations the same.

Anyway, that’s a rant, but I think to your point, I feel like Swift could have been as big as Rust..or bigger given the easier use, with many of the same guarantees. I just wish the early years were more measured.

[embedded3]

>The rust code was running circles around their Swift ones even when we tried to keep implementations the same.

I've done Advent of Code a few years -- even Javascript implementations, if using a good (optimal) algorithm, are highly performant, so I'm suspicious of the claim. In most AoC problems, if your code is observably different between languages, it's the fault of the algorithm, regardless of language. But perhaps you are referring to well-profiled differences, even if there are no observable differences.

That said, in projects other than AoC I've compared Swift to C++ and it's hard to deny that a low-level language is faster than Swift, but Swift itself is certainly fast compared to most dynamically typed or interpreted languages like Python, JS, Ruby, etc. which are vastly slower than anything in Swift.

[dagmx]

Swift is fast but it’s both easy to accidentally write slow code and there is a not-insignificant amount of overhead to the reference counting and constant dynamic table lookups.

When I say Rust ran circles around it, I mean the difference was not noticeable unless timing, and was the difference of 200ms vs 400ms or a 3 seconds vs 5 seconds , so nothing to write home about necessarily.

[irae]

That is not the first time I want to understand a bit better the performance difference today between the approaches of Rust, without a garbage collector, ARC on Swift with the reference counting and other garbage collected languages, such as Javascript.

I know Javascript have an unfair advantage here, since the competition between V8 and the other javascript cores is huge over the years and garbage collection on JS is not often a problem. At least I see more people struggling with the JVM GC with its spikes in resource usage.

I've also heard that the erlang VM (be it written in elixir or erlang itself) implements GC on a different level, not to the global process, but on a more granular way.

Is there a good resource that compare the current state of performance between those languages or approaches?