Current state of MVCC split version

[huangyihe]

This is how MVCC split version is supposed to work. The quoted part shows the template specialization used to define our "split policy".

The templates complicate things especially with deletes. I think there is a solution I just haven't gotten much time to think it through. The main challenge is with the current function pointer mechanism we have to invoke the deletion function, which doesn't work with the function templates used to loop around different chains. We will need some workaround on this.

[huangyihe]

I just added a RecordAccessor interface to unify client's view over split and unified records, so that we can have a single benchmark implementation for both split and non-split records. An example of how it's used is in the updated unit test.

The quoted lines here between line 10 and line 152 are highly boring boilerplate code that can be (and should be) easily generated from the schema. I'm working on such a code generator right now.

Structured binding declarations in C++17 are really a wonderful thing.