[ParU] Use more C++

[gruenich]

Replace C code by C++ code.

• Use range-based for loops.
• Don't use own definition of NULL and ASSERT
• Use std::min and std::max

If you are unhappy with some of the commits, I can remove them and create separate pull requests.

[mmuetzel]

These changes look good to me.

[DrTimothyAldenDavis]

Thanks for the update. A few comments:

I can't use a bare assert(...). I always must wrap it inside a macro, for several reasons. First, when used in MATLAB, the ASSERT macro must become a call to the MATLAB mxAssert function, not a C / C++ assert. Second, I put computationaly expensive things inside my ASSERT(...) calls, and these must be disabled at all times. I enable them only by editting the code itself:

SuiteSparse/ParU/Source/paru_internal.hpp

Lines 21 to 26 in 19f1109

	// debugging and printing macros
	// -----------------------------------------------------------------------------
	// force debugging off
	#ifndef NDEBUG
	#define NDEBUG
	#endif

That is, I intentionally make it very hard to turn on debugging, since if it gets turned on by mistake by an end user, then my codes can be 1000x slower, or worse. Then when I release the code, I force debug off by editing the code back.

And then sometimes I will revise the ASSERT macro so that it does something else, with a more informative error message, while developing the code.

[DrTimothyAldenDavis]

Yes, in general, std::min and std::max would probably be better. I would guess that the compiler might handle it better, in terms of performance.

The Control->paru_max_threads should be a plain int (or perhaps int32_t) anyway. That would simplify this particular case for std::min and std::max. This isn't a performance issue in this case, though.

For the for loops -- do they affect how OpenMP handles the loop? I haven't reviewed all the changes yet, though. I haven't used that style of for loop for a #pragma omp parallel for before.

[DrTimothyAldenDavis]

looks great -- thanks!