mat_sparse_t should be extended to manage 64 bits index

[Nelson-numerical-software]

nzmax, ir, jc should be uint64 and not int32
v7.3 matfile uses uint64 and not int32 index

[tbeu]

Confirmed, the design of the mat_sparse_t structures is broken w.r.t. the dimension types.

Furthermore, there is no need to store isComplex, isGlobal, isLogical and mem_conserve as int32 in matvar_t. This would save 12 bytes out of the 72 bytes of matvar_t.

Unfortunately, both changes are not backward-compatibel and so far I am not sure if it is a good idea to break compatibility. There are tools out there (e.g., C# wrappers) that depend on a certain memory layout of both structures.

[Nelson-numerical-software]

Unfortunately, both changes are not backward-compatibel and so far I am not sure if it is a good idea to break compatibility. There are tools out there (e.g., C# wrappers) that depend on a certain memory layout of both structures.

With an compilation flags, it should be possible to choose indexing mode.
64 bit indexing for sparse is available since 2006.

[tbeu]

Which leaves the question, what to make the default case.

[Nelson-numerical-software]

It could be something as:

• current version 1.5.13
• next version 1.6.0 with compilation flag with index 64 bit disabled by default but with a warning during build indicating the next change.
• 2.0.0 64 bit index by default