“Blazor WebAssembly supports ahead-of-time (AOT) compilation, where you can compile your .NET code directly into WebAssembly. AOT compilation results in runtime performance improvements at the expense of a larger app size.
Without enabling AOT compilation, Blazor WebAssembly apps run on the browser using a .NET Intermediate Language (IL) interpreter implemented in WebAssembly”
https://learn.microsoft.com/en-us/aspnet/core/blazor/host-an...
F# also has some issues with some IL constructs generated by C#, with features not exposed in F#, like protected, which can be consumed, but not authored.
Your second point is exactly what I said, though. F# can consume things written in C#. Whether the F# language team wants to include some specific feature is a different question. C# libraries that use the protected attribute can be used transparently in F#.
They are different languages with a shared runtime. You can't write a computation expression in C#, but an F# library function that is implemented with CEs can still be called from C#.
Only if you don't need to write data types that need to be consumed by said libraries, as to express those types you need C# features.
Another two key examples are the recent trends from .NET team to depend on Roslyn and code generators, both not supported by F#, so the language can't be fully used in such workloads without a little bit of C# glue.
Really, F# might be from Microsoft, but the .NET team handles it as if it was a 3rd party guest language.
I think this is why .NET core doesn't have typeproviders?
We found that in dotnet 6, the code was much slower, with long startup times and a much bigger download, than in js_of_ocaml. It also had a lot of issues in running in a Webworker, which wasn't the case for js_of_ocaml.
In dotnet 7, the webworker issues are better and AOT is easier, so startup is faster. Download sizes are still bad, and it's still slower than js_of_ocaml.
However, dotnet allows almost any code to run in WASM, which js_of_ocaml had large limitations. This meant a decent chunk of functionality had to be worked around to make separate js vs native targets, which also was a massive pain and took a long time. Dune's virtual targets wasn't ready at the time - I think we were one of the test cases for it.
The Fable compiler team seems to have made steady progress towards a Rust compilation target which would solve this, but I'm not clear on where that works or how.