Marshalling time of custom types is dominated by hash computation

[Imberflur]

flamegraph.svg.zip

This seems like something that could be vastly improved by caching the type_guid computation for StructRef but there may be a reason that this is not being done.

impl ReturnTypeReflection for StructRef {
    type Marshalled = RawStruct;

    fn type_name() -> &'static str {
        "struct"
    }
}

[Wodann]

Good find. There is no reason why ReturnTypeReflection::<StructRef>::type_guid can not be constant. I'll look into optimising this, thanks! 🙂

Also, exciting to see that you are considering this for Veloren 👍

[baszalmstra]

@Wodann You are currently working on a fix for this right?

[Wodann]

I have a branch with the improvement, but wanted to add some additional benchmarks as well. I was working on a second performance improvement based on that.

The first pass can be found at: https://github.com/Wodann/mun/tree/improve/struct-marshalling-perf

[Wodann]

Resolved by #238.