Add memory order conversion to handle non-contiguous input arrays

[maweigert]

This PR fixes current unexpected behaviour for non-contiguous input arrays.

[william-silversmith]

Thank you for the PR! I like what you did here with one exception. It's in theory possible for an array to be in F order with a C layout (e.g. if someone wrote to it as arr[ z, y, x ] or acquires it from a C extension). Maybe it would be better to have the default order be "auto" or "detect" which does what you want here, but allows one to specify "C" or "F" manually if that's the behavior they want?

Would that be too confusing?

Happy new year!

[william-silversmith]

Looks like 'A' is the numpy version of 'auto'. 'K' is pretty similar.

https://docs.scipy.org/doc/numpy/reference/generated/numpy.array.html

Something like this maybe?

  if not data.flags.c_contiguous and not data.flags.f_contiguous:
    order = 'C' if order != 'F' else 'F'
    data = np.copy(data, order=order)
  elif order == 'A' and data.flags.c_contiguous:
    order = 'C'
  elif order == 'A' and data.flags.f_contiguous:
    order = 'F'

  if order not in ('C', 'F'):
    raise ValueError("order must be 'A', 'C' or 'F'. Got: " + str(order))

[william-silversmith]

Thanks for the contribution! I ended up doing it in #22 to allow people to choose how they wish to interpret the array. I'm always happy to get feedback and discuss additional changes. :) Thanks for being the impetus for making dealing with Fortran arrays easy!

[maweigert]

Sure! Btw, we're using edt in https://github.com/mpicbg-csbd/stardist and its so much faster than the scipy.ndimage version! Thanks for the package! :)

[william-silversmith]

Cool! I have a very informal release process, so I'll need to get all my computers together tomorrow to release new binaries.

By the way, I was glancing through your code, and realized I might have another library you might enjoy: https://github.com/seung-lab/fill_voids

fill_voids is a replacement for binary_fill_holes which is MUCH faster and lower memory. Check it out!

[maweigert]

Thanks! I love issues turning into knowledge :)