API inconsistency between remove_small_holes and remove_small_objects

[sciunto]

Description

In 0.16, in remove_small_holes, min_size has been replaced definitely by area_threshold. See here.

But for remove_small_objects, it is still min_size. Why that? The first one call the other internally...

[jni]

@sciunto personally I disagreed with the change to area_threshold, but didn't catch that PR in time. Either way, it's likely that remove_small_objects was simply overlooked. Thanks for flagging this. We should have a long discussion about the best name for these parameters.

[mkcor]

I think that min_size is/was fine as long as documentation is well written.

My second favourite option would be #4761 (comment) by @jni because it reads so well.

[lagru]

Having dug through the past issues and discussions on this, it seems to me that each discussed option will be a trade-off between consistency and matching the context of each function.

Since remove_small_objects and remove_small_holes are so similar, I definitely give consistency more weight. Ignoring past deprecations, min_size seems to be the least controversial option. The main argument against it was that we'd have to revert a previous deprecation.

That's not ideal, but addressing previous mistakes is precisely what skimage2 is about. So, I propose to deprecate area_threshold in remove_small_holes back to min_size and only finalize the deprecation in 2.0.0.

I think we can keep using area_threshold in area_opening and area_closing, since area is a prominent term and I anticipate less user confusion.

I'll make a PR to get this going again. Please, feel very welcome to suggest an alternate concrete action to address this issue. 😉