Breaking changes in 3.0.0

[nedtwigg]

Hi users! We've found a bug that requires some breaking changes to fix. Since we've gotta break compat anyway, this seems like a nice opportunity to clean up a few other warts. I'm looking for comments on the things we're currently planning on breaking in 3.0.0:

• Fixed multidimensional array indexing (see #10).
  • For arrays with dimension 1 or 2, there is no change.
  • For arrays with dimension 3 and up, matfilerw 2.x did not order the dimensions in the column-major format used by MATLAB. In 3.x forward, matfilerw uses the same column-major format as MATLAB.
• Complex and non-complex arrays now have separate methods.
  • MLNumericArray has methods getReal/setReal, getImaginary/setImaginary, and get/set.
  • In 2.x using the imaginary methods on a non-imaginary matrix would fail silently, and get/set were synonymous with getReal/setReal
  • In 3.x, you can't use getReal/setReal on a non-complex matrix, and you can't use get/set on a complex matrix.

I'm tagging a few here who might be interested: @tmrn411, @mikaelgrev, @sinjax, @no-man-is-an-island, @gradusnikov.

[nedtwigg]

After a little more thought, I'm rethinking the getReal/setReal, getImaginary/setImaginary, and get/set division. Complex arrays should really be a different type, but that's a bigger change than we can handle. I think I'll remove the restrictions on getReal/setReal vs get/set so that it has the same behavior as 2.x.

Edit: done in 73c3a63

[tmrn411]

Just as point of reference, in Matlab real() and imag() work just fine on non-complex variables. imag() just returns 0 in that case. So I think the 2.x behavior is reasonable.

[nedtwigg]

Thanks for the info @tmrn411. 12858a6 makes sure that imaginary will return 0 for all non-complex arrays.

[nedtwigg]

Shipped 3.0.0