Phantom coding philosophy/developer guidelines
What Phantom is not
Phantom is not a code for testing algorithms. It is a “take the best and make it fast” production code for high resolution simulations in 3D.
What Phantom should be
No repeated code
Avoid repeated code. Do not cut-and-paste. Use subroutines and functions.
Group related functionality together. New physics = changing one file.
Actively delete old/obsolete code.
Add unit tests for important modules.
Automate testing as much as possible.
Please follow the house style in your coding
Try to write subroutines and modules with minimal dependencies. Send options as arguments rather than bringing phantom-specific modules as dependencies.
Write PURE subroutines and functions as much as possible.
Write your code so it can be made into a library.
Subroutines ideally should not be longer than a few hundred lines of code.
Much of what I try to follow comes from Joel on Software:
At some point I would like to follow these as much as possible:
Rambling justification for above
ndspmhd, is the code I use for testing algorithms. That code has many different options, all specified at runtime. It has dynamically allocated arrays. It can be compiled in 1, 2 or 3 dimensions. It’s very flexible. But it is bloated, has everything I’ve ever tried in it, and would be hard to optimise or parallelise. Phantom is meant to take the “best” of things I’ve tried in ndspmhd and implement them fast and parallelised in 3D.
In the most important routines like densityforce, this means I will sacrifice readability for speed. The initial design goal was to have a very low memory footprint. That is memory should only be allocated if it is strictly necessary for the physics. Hence most of the physics is chosen at compile-time rather than runtime. Hence also Phantom uses static rather than dynamically-allocated arrays.