Skip to content

Latest commit

 

History

History
59 lines (41 loc) · 2.28 KB

README.md

File metadata and controls

59 lines (41 loc) · 2.28 KB
Raw

Basic Common Cpp

This is a set of code used in C++. All of this is basic but used in many projects. Author is Mathieu Dutour Sikiric (e-mail: [email protected]), but anyone can contribute.

Access to the source code

Since this repository uses submodules, the cloning command is

$ git clone https://github.com/MathieuDutSik/basic_common_cpp.git --recursive

In order to update the submodule the command is

$ git submodule update --remote

Design

Namelist

Forran provides a quite powerful system for doing input files. It is named Namelist and the extension of those files are nml. We provide this functionality in C++ with a more rigorous way: An entry can be defined only one time for example.

Scalar types

By scalar types we mean the one used for matrices, vectors, etc. Example could be integer, mpz_class, mpq_class, etc.

We want to use the C++ template capabilities for having multiple types. This requires some strict decisions about how things are done.

The conversion is done via UniversalScalarConversion<To,Ti> and similar with To the output type and Ti the input type. The types are always strict, there is never any implicit conversions.

The operation are of the kind x > y with x and y of the same type or x of a constructed type and y an integer. The integer type is always second. The code looks like

  • friend bool operator>(T const& x, T const& y)
  • friend bool operator>(T const& x, int const& y)

This goes the same way for ==, !=, <=, >=, < and >.

The operator<<, operator>>, hash(), load/save (for boost serialization) have to be defined for all the types.

The rules for the constructors are not as well defined. But it looks like the constructor from integer is forbidden because it leads to inconsistencies with x > 0 because either operator>(T const& x, T const& y) or operator>(T const& x, int const& y) could be used.

For the numerical types, some types allow the constructor from operator=(int) but some do not allow that which means that T sum = 0 will not compile. The solution is then to write T sum(0). This is especially the case for rational for which the T can be a int64_t and this cause some implicit conversions.