Using C library with different data structure

Question

I have a question about the best way to write computing code as flexible as possible in C++, I give you this example:

I am writing a code that manipulates meshes (Finite Element computation), basically set of nodes and connectivities. We only focus on Node structure for now:

struct Node { double x,y,z; };

I have a set of algorithm implemented using this Node data structure.

However I want to use an external library that provides specific geometrical treatment on meshes (writen by a collegue), it has a different data structure for Node

struct LibNode { double x; int idx; double y,z;};

note: int idx is used in the library, I do not need/want it in my own developpement.

All the algorithms of this library are written with this structure. I am just using the library, for me it is just a set of treatment that I can apply to the mesh I am using for computing.

I want to use the algorithm in the library written in C, so here are the solutions I imagine:

• make a copy of my Nodes to some LibNode, simplest solution, but very unefficient !

• rewrite my code with LibNode, but what if I want to use several libs ?

• rewrite the lib !

I have several questions here:

• Am I missing a better solution ? I feels like whatever I do I cannot avoid using library's data structure.

• As performance is an issue, for now I think that rewriting the lib will be the best solution. Then I will write it in C++, and I do not want to make the same mistake than for the original lib. I see two options, template implementation of the lib, more "elegant" but also more difficult to write and maintain by a team of physicists stucked to F77. Or manipulating the mesh through an abstract class and delegate the data structure definition to the user of the lib. But access data through virtual method (more than a million time by code run) seems to be greatly inefficient to me, am I right ?

Thank you in advance for your response !

Answer 1

If your code is conformant C, the simplest way would be to rebuild the library using Node, just changing Node to struct Node { double x; int idx; double y,z;};. All access to members x, y and z should work correctly, and the idx node will be processed more or less as padding. After all, that is one of the best points on using structs for maintaining applications: you can add members with no or little change to existing code

Of course if you assume somewhere that &node.y == &node.x + 1 everything will break...

The only thing I can think about that could be problematic would be loading structs that would have been saved as a whole to a binary file. That part (and the save part) should be rewritten to explicitely save and restore members and not the whole whole struct - anyway saving a whole struct is non portable and should be avoided if possible.

Another problem, mainly if it is used as a DLL, is that it will be a major change because it will break compatibility, so you should considere to rebuild all programs using your library.