Roslyn VS SharpLab

Local functions do not support overloading

Honestly, I don't have a clue why these would be different. I found the relevant Roslyn code here. You can see on line 719 that this is definitely intentional:

compiler: https://github.com/dotnet/roslyn

Why not use Roslyn Scripting…you can pretty much do any kind of dynamic evaluation with this and it does not require any reflection. https://github.com/dotnet/roslyn/blob/main/docs/wiki/Scripting-API-Samples.md you can add the support via a Nuget.

It was discussed at one point: https://github.com/dotnet/roslyn/issues/5058

It must be a fairly large echo chamber since it has room for Clang, GCC, V8, OpenJDK, Roslyn, etc. (The Zend parser for PHP seems to use some flavor of YACC, but given PHP, I don't know if that strengthens or weakens my point.)

I'm planning to rewrite all of our company's old T4-based codegen tools into Source Generators. I've been looking at how other Source Generators are implemented and what surprised me is that everyone seems to assemble the output code manually as text with string builders or at best some simple templating engine (eg. Scriban). I expected that I would see a lot of projects assembling their output code from Roslyn tokens and then using Roslyn's formatter to produce the actual code that can be emitted into the output file(s), like this article demonstrates. No-one seem to be doing that though. Why?

Clearly nobody ever manages to use this Go thing with over 700: https://github.com/golang/go/issues. Or .net, over 5000 issues for the compiler alone: https://github.com/dotnet/roslyn. I'd guess stuff like Java, glibc would be similar if they used github

With sharplab.io you can see the C#-code which is produced during lowering. When we paste our original GetMonthIndexSwitch in there, you will see that this is rewritten into if statements.

It does not sort your issue out but if all you want to do is check out the latest language C# 11 Preview feature, you can run main branch of dotnet as of 19 Apr 2022 at sharplab.io without installing anything.

If that is the term that would help me do something equivalent to godbolt.org or https://sharplab.io/ against other runtimes.

The above example shows state machine that was generated by a source code containing only one await keyword. When we use await more than once in the given method, the state machine ends up with more complicated MoveNext version. If you would like to experiment with it, I recommend https://sharplab.io/ for viewing Intermediate Language or a downloadable tool ILSpy which has similar functionality.

I have never worked with C#, so I can't really tell you unfortunately. If I inspect the "JIT ASM" of the following code in sharplab.io, in Release mode, it seems that similar optimizations are performed. static bool IsEvenA(int n) { return (~n & 1) != 0; } static bool IsEvenB(int n) { return n %2 == 0; }

The other thing that may hint to what is going on is to use sharplab.io (https://sharplab.io) and decompile to C# to get an imperative view on your code, which will be closer to the reality of how it is executed. This may be overwhelming for the full code so you may need to target specific snippets.

Great tool to inspect the lowered C# version of your code: https://sharplab.io/

sharplab.io

There is one exception I have encountered: you're limited to a single file (e.g. because you're using https://sharplab.io/) and you need to to do something weird (typically using a preview compiler feature without running on a preview runtime) by adding a type from a specific namespace (typically System.Runtime.CompilerServices).

If your ever curious about what your code compiles to you can just throw it into sharplab.io. There you can see things like foreach loops are in fact just for loops and for loops are just while loops to the compiler. It's a great tool to use if you're ever caught wondering whether one implementation of your method is faster than another.