Sergio Pedri

I think there's some misunderstanding here and some incorrect statements, and also some frustration towards Microsoft that I'm not really sure is warranted in this specific case. Let me try...

> "Do you have any data to prove that, in real-world scenarios involving actual network requests, that this makes any significant difference?" Yes, I'd you're doing few very long requests...

> "This is Curious because Refit actually uses source generators because of Xamarin AOT compilation requirements on iOS, and there it has worked for nearly a decade without issues. Why...

Hey @clairernovotny, so the profiler project is basically ready, but as mentioned above I can't get the `BenchmarkDotNet` run to work correctly. That is, manually running the benchmark in both...

@MihaZupan Alright, I was familiar with some of Egor's work but I had completely missed that blog post, that's really next level! Thanks for sharing, I definitely learnt something new...

Ok I have to say I'm extremely surprised to discover that the right shift isn't working as I was expecting there, and I'm still not sure why that is to...

Yeah the resulting code seems small enough to be nicely inlined, plus the codegen is surprisingly clean despite the (arguably) ugly looking C# code with all those various hacks going...

@xoofx That's great to hear! `Markdig` is already quite efficient judging by those benchmarks you have in the readme, and it'll only get better from there! Off topic: your starklang...

@MihaZupan Oh it's literally just that gist, copy-pasted into a blank .NET Core 3.1 project referencing `BenchmarkDotNet`. But sure, if the whole project helps, here it is: [MarkdigCharacterPatternBenchmark.zip](https://github.com/lunet-io/markdig/files/4397611/MarkdigCharacterPatternBenchmark.zip). Looking forward...

Ah, yeah that's nice, as it removes that extra `movzx`, I'm doing the same in my versions too 👍 I'm curious as to whether those additions could make it into...