- concentrate on expression evaluation, other things to make faster
- explain new expression evaluation + reasoning
- performance numbers
- what's JITing
- what's LLVM
  - IR / Bitcode
  - Modules
- explain how basic expression JITing looks like
- performance numbers
- things to think about:
  - syncing struct/type definitions
  - resource management
  - planning consideration
    - comments welcome
  - function reuse
- next step: function calls
  - function call overhead
  - function inlining
  - reuse clang -emit-llvm
  - extensibility?
- next step: expressify more
  - aggregate transitions
  - hash agg/join hash computations?
- next step: deforming
  - skip deforming if not necessary
    - planner support / executor structure
  - access columns at constant offsets directly, skip deforming if only access
  - JIT deforming
- Order:
  - profiling support of JITed functions
    - how to name the damned things
  - LLVM infrastructure / resource management integration
  - plain expression JITing
  - function / operator inlining
  - direct access to deformed columns
  - JITing deforming
  - memory-lifetime hints
  - code generation improvements

Issues for others:
  - cacheability of JITing