Preprocessing limits are broken

[K-ballo]

There are a number of issues with preprocessing limits:

• HPX_*_LIMIT do not always reflect the actual limits.
• The limits reported aren't always available, as some limits are shifted but still capped by HPX_MAX_LIMIT.
• For shifted limits, preprocessed files support more arguments than their name implies, but those are never used since the selector files do not consider the shifting.
• If limits other than HPX_LIMIT can be user defined, alternative symbols that are properly rounded should be used internally instead when using preprocessed files.
• Some limits are just plain wrong.

[hkaiser]

• Some limits are just plain wrong

The whole concept of using the preprocessor to define a set of variadic overloads for a function is plain wrong :-P

[K-ballo]

    Some limits are just plain wrong

The whole concept of using the preprocessor to define a set of variadic overloads for a function is plain wrong :-P

Are you suggesting we move to true variadics?

[K-ballo]

The limits branch has my take on preprocessing limits.

HPX_*_LIMIT do not always reflect the actual limits.

• Now they do.

The limits reported aren't always available, as some limits are shifted but still capped by HPX_MAX_LIMIT.

• Limits reported are always available, as preprocessed generated files are not capped by HPX_MAX_LIMIT.

For shifted limits, preprocessed files support more arguments than their name implies, but those are never used since the selector files do not consider the shifting.

• Selector files only consider the base limit, not the shifted ones, thus the correct ones are picked.

If limits other than `HPX_LIMIT` can be user defined, alternative symbols that are properly rounded should be used internally instead when using preprocessed files.

• Only HPX_LIMIT can be user defined, defining one of the other limits results in an #error.

Some limits are just plain wrong.

• Fixed

Additionally, I removed HPX_PP_ADD3 and HPX_PP_ROUND_UP_ADD3 as they were only used in a handful of places within a single file. I replaced those use cases by BOOST_PP_ADD instead.

Comments?

[hkaiser]

Looks great! Marvelous. Please feel free to merge if tests are no more broken than before :-P