Guid properties being boxed?

[AlgorithmsAreCool]

Been doing some profiling and I noticed some boxing of Guids occurring during serialization. Is boxing of value types expected? If so is there a workaround?

[agocke]

At the moment, yes. See here:

serde/test/Serde.Test/generated/SerdeGenerator/Serde.SerdeImplRoslynGenerator/Serde.Test.AllInOne.IDeserialize.cs

Line 101 in 9c35b8a

_l_guidfield = typeDeserialize.ReadBoxedValue<System.Guid, global::Serde.GuidProxy>(_l_serdeInfo, _l_index_);

The reason is that reading generic types requires either GVMs (generic virtual methods) or a special-case method. In my benchmarks, I found that GVMs are actually slower than boxing, so I removed all GVMs and went with boxing. All of the primitive types, meaning types that may have special handling by the underlying format, have dedicated methods.

The only workaround I can think of would be to implement Deserialize for the parent type manually, read out the primitive type directly (string, byte array, whatever) and then directly create the Guid inline.

The broader feature that I could add would be adding Guid as a primitive type, which would give it a dedicated ISerialize/IDeserialize method. I'm just not sure it's worth it. As far as I can tell there are very few/no formats that have a dedicated Guid representation, so it would be somewhat awkward to make a primitive.

I guess the question is: how much does the boxing matter?

[AlgorithmsAreCool]

Hmm, I didn't realize the penalty of GVMs was so high... interesting

[agocke]

Ah I just had an idea for a general API that could be used for things that can be encoded as fixed size integers. I’ll see if I can prototype it.