C++ Programming Best Practices in Scientific Computing

This is a presentation by Alfredo Correa (LLNL) on good C++ programming practices.  C++ techniques from AFQMC.pdf. Alfredo is supported in part by the QMCPACK ECP project, which also supports Nichols A. Romero (Unlicensed).

(Reproducing two of the main slides)

When is a good idea to use OO (virtual, new,>, etc)? (spoiler: almost never to do math or a simulation)

Perhaps if:

• Immutable (but polymorphic) state
• Shared/singleton objects/representations of logic
• `In-the-heap-anyway'
• Open hierarchies (e.g. open behavior, cross compilation-boundaries)
• Exceptions (std::runtime_error)
• Resources (std::pmr::memory_resource)
• Devices (std::iostream)

Perhaps not if:

• - - - Value Semantics
• Mutability
• Local reasoning

Remember, virtual was never used in the STL and it still went a long way:

$ grep -R virtual /usr/include/c++/*

Conclusions

• Try not to add new virtual functions
• Give your arithmetic types value semantics
• Use constructors and assignment instead of (pointer-) factories
• (make * functions are ok)
• OO can coexists with Generic Programming and slowly fade away
• Use the stack for objects, or be stack-friendly
• Use dynamic memory as an internal implementation detail
• Use algorithms (100+ STL algorithms)\
• (chances are your loops are in essence existing algorithms)
• Use containers, even custom containers/iterators/pointers
• Use dynamic memory generically through allocators