Add notes on why anonymous, generic lambda wrappers are not possible in our context.

NOTES.md

@@ -4,6 +4,29 @@ A collection of stuff that we noticed during development.
 Useful later on two write a project report and to go back
 in time to find out why certain decisions where made.
 
+## 11.04.2019 - Lambda Pointer Abstraction
+
+The question is if we could use a pointer to a lambda without
+needing templating (for the type of the lambda) at the place that
+we use it.
+
+We looked into different techniques to achieve this:
+- Using std::function<...>
+- Using custom wrappers
+
+std::function uses dynamic memory, thus ruling it out.
+All methods that we can think of involve storing a pointer
+to the lambda and then calling on it later on.
+This works well enough with a simple wrapper, but has one
+major downside: It involves virtual function calls,
+making it impossible to inline the lambda.
+This property broke the technique for us in most places,
+as inlining is crucial, especially in small functions like loop
+iterations. See `invoke_parallel_impl.h` for an example where we
+did this (wrapper with virtual function call), but we only did it there,
+as the generic `fork_join_sub_task` requires the virtual call anyway,
+thus making us not loose ANY performance with this technique.
+
 ## 11.04.2019 - Notes on C++ Templating
 
 After working more with templating and talking to mike,