Check that compiler generates reasonable code when upper bound of for loop involves a function call

[dictoon]

E.g. in ewah.h line 908 it's worth making sure compilers compute the upper bound once, and not at each iteration.

A safer alternative, immune to optimization woes, with minimal changes:

for (size_t k = 0, e = rlw.getNumberOfLiteralWords(); k < e; ++k)

[lemire]

If these sort of function calls do not get inlined and optimized using current C++ compilers, then the whole code base is in trouble because it is filled with getRunningBit(), getNumberOfLiteralWords() and so forth all over, including within hot loops. So there is a strong assumption that C++ compilers are at least that good.

The issue you raise is totally valid and this assumption should be checked.

In Java, I do not make this assumption. Often the JDK cannot optimize these calls away. But C++ is a different beast and I assume that C++ is a lot better at providing us abstraction for free.