Isn't it cleaner to only have sections when there are more than 1 definitions in a file?

I had the same thought, kinda just feels like clutter when there is nothing else in the file except the item in question.

I usually put these in to separate the "imports" section from the main code. It is a kind of a visual clue for "important code starts after this line".

I was thinking a KernelLibrary would only wrap a CompiledLibrary (and would make all its checks in the constructor). Then the ModuleInfo and Kernel can be derived from the CompiledLibrary in methods. The advantage of that approach would be that the struct would be simpler to understand, since it wouldn't store any duplicate information.

This is how I had it originally but then noticed that the constructor and into_parts() method were almost identical (e.g., we need to build a Kernel at construction time to validate that it can be built without errors). So, given that additional memory footprint is likely to be minimal (kernels can have at most 256 procedures), decided not to throw out all the stuff we built in the constructor and save it with the struct.

I do agree that we could probably refactor things a bit here (e.g., maybe Kernel should be just a newtype over ModuleInfo) but I'm leaving this to future PRs.