I have come across a fair few books and talks on C++ which teach us how to write good, maintainable code using good programming practices. Scott Meyers, Bjarne Stroustrup, Chandler Carruth, all have great ideas of efficiency, error-proofing, et al.
However, books on scientific C++ use naked pointers *a instead of smart pointers, C-style arrays instead of C++ vectors/arrays, and many more such. This leads to a situation where books on scientific C++ programming teach very bad programming practices/styles, and books on good practices/style don't really focus on scientific computing (PDE solving, nonlinear optimization, linear solvers...).
Welcoming all suggestions for books, talks, videos, tutorials, whatever, on good scientific C++. Even general advice on the process of learning good C++ will be greatly appreciated.
Just a few quick comments:
- We tend to stay away from eg. smart pointers since these have a negative impact on code speed and memory footprint. Concepts considered as good programming practice have to be evaluated on their performance impact. Not many programming references pay much attention to this.
- for the same reason, you will see that many codes revert to standard C or fortran in the most compute intensive parts. Using dynamic objects etc at this level is mortal for performance due to overhead in recuperating functions etc. Keep polymorphism at a very high level.
- the baseline standard template library provides many very interesting functionalities in terms of containers, but you have to pay attention again to performance aspects when deciding which container to use.
- another interesting library is boost, which provides an extremely large panel of advanced functionalities. Our experience is that these implementations are usually very well done, both in terms of api and performance, and extremely portable.
- it is important to look at templatisation as it is an efficient way to provide polymorphism and generic coding (eg. for multiple precision code), for compute-intensive parts. It also enables short hand, almost "paper notation" style operators eg. for algebra (so-called expression templates).
How about: https://www.amazon.com/Discovering-Modern-Scientists-Programmers-Depth/dp/0134383583/ref=sr_1_16?ie=UTF8&qid=1495728264&sr=8-16&keywords=c%2B%2B ?
Hope these might help.
http://www.springer.com/in/book/9781852334888
http://www.springer.com/in/book/9781447127352
http://www.amazon.in/Scientific-Engineering-Introduction-Advanced-Techniques/dp/0201533936
C++ for Programmers from Paul Deitel is a good one :)
https://g.co/kgs/SRkazo
Discovering Modern C++ by Peter Gottschling is rather nice, but also not for novices. Book Guide to Scientific Computing in C++
is a more gentle introduction to Scientific Computing with C++ if it misses some of the newest stuff.And I disagree with the comment about smart pointers. Shared pointers cause overhead, of course, but unique pointers do not since they are eventually converted by the compiler to memory addresses, like an ordinary pointer. Always use a unique pointer if the pointer owns a resource uniquely. Use shared pointers only if it is necessary to share the resource. Use basic pointers only if they do not own the resource but are just addresses to an object you may want to change run time but whose life is not handled via the pointer. Use reference if you want an alias to an object. This is my opinion.
- Badges
- Science topic
- Similar topics
- Mathematical Sciences
- Control Theory
- Control Systems