C++ Style

Prefer nullptr to 0 or NULL

0 is an int not a pointer. Almost the same goes for NULL, though implementations of the language can differ in the details. If you want to overload on pointer types and/or use pointer types with templates, use nullptr to signal the null pointer. The correct overload/template parameter will then be deduced. Using nullptr also makes the code more readable, especially if auto is used consistently throughout.

Reference: Item 8 in [Effective Modern C++]

Prefer std::make_shared to direct use of new

Using std::make_shared:

  1. Reduces code verbosity, especially when coupled with auto:

    1
    2
    3
    4
    5
    6
    7
    8
    // Type information given 3 TIMES!!!
    std::shared_ptr<Matrix> F = std::shared_ptr<Matrix>(new Matrix("Fock matrix", nso, nso));
    
    // So much typing...
    std::shared_ptr<Matrix> F = std::make_shared<Matrix>("Fock matrix", nso, nso);
    
    // Much better!!!!
    auto F = std::make_shared<Matrix>("Fock matrix", nso, nso);
    
  2. Ensures exception safety and prevents resource leaks.

  3. Improves efficiency:

    1
    2
    3
    4
    5
    // Performs TWO allocations
    std::shared_ptr<Matrix> F = std::shared_ptr<Matrix>(new Matrix("Fock matrix", nso, nso));
    
    // Performs ONE allocation
    auto F = std::make_shared<Matrix>("Fock matrix", nso, nso);
    

Reference: Item 21 in [Effective Modern C++]

Prefer auto to explicit type declarations

Using auto reduces and/or avoids:

  1. Verbosity in variable declarations:

    1
    2
    std::shared_ptr<Matrix> F = std::make_shared<Matrix>("Fock matrix", nso, nso);  // So much typing...
    auto F = std::make_shared<Matrix>("Fock matrix", nso, nso);  // Much better!
    
  2. Problems with uninitialized variables. auto works like template type deduction, hence the right-hand side of the declaration needs to have an initializer:

    1
    2
    3
    int x1;  // fine, but uninitialized :(
    auto x2;  // WON'T COMPILE!!!
    auto x3 = 1;  // fine and initialized
    
  3. Problems with unintended type casts and type mismatches that are hard to impossible to catch:

    1
    2
    3
    4
    std::vector<int> v;
    // !!! The size of a vector is of type std::vector<int>::size_type and is compiler- AND architecture-DEPENDENT
    unsigned sz = v.size();  // might not be correct on some compiler/machines
    auto size = v.size();  // size is ALWAYS of the correct type
    

Reference: Items 2 and 5 in [Effective Modern C++]

Prefer GiB for memory printing

As memory sizes get larger, we should work in giga (requires decimal printing to not round to zero) rather than mega units. As it’s what we’re computing anyways, we should work in 1024-based (mebi, gibi, etc. https://en.wikipedia.org/wiki/Gibibyte) rather than 1000-based units. As it’s a unit, put it in brackets. Note that users can supply MiB, GB, bytes, or whatever; this guideline is for output printing.

1
2
outfile->Printf("  DFHelper Memory: AOs need %.3f [GiB]; user supplied %.3f [GiB]. ",
                (required *  8 / (1024 * 1024 * 1024.0)),