The main function is sdp(...) in ../deps/src/Csdp-6.1.1/lib/sdp.c, starting in line 68.
It takes about 40 parameters, including a lot of structs.
In contrast, the easy_sdp(...) method has only 11 parameters, because it computes an initial solution and working memory.
Csdp works always with block matrices. For sparse matrices, I don't see the benefits, yet.
Question: Is it necessary that all the matrices C, A1, and A2 in the example have the same block structure? Yes, it is: They need to be multiplied or added.
The sparseblock.nextbyblock pointer is always NULL in the example.
Another peculiar aspect is, that the C code uses Fortran like arrays, i.e. the first index is 1, leaving an empty entry before.
Only two dimensional arrays are accessed via the ijtok macro which starts at 0.
To overcome the issues, one has to always add another element to the array or one should add -1 to the pointer; that is done by the fptr function which returns the pointer to “before that object”.
A major problem seems to be the the question whether blockmatrix should be immutable or not:
On the one hand, in some C functions the blockmatrix has to be handed over; so the memory alignment, which is better for immutable's, is important.
On the other hand, blockmatrix has to be allocated by init_sol; that is why it has to be mutable.