Pyresample cannot be used for gapfilling due to `geometries_are_the_same` check

[ameraner]

I have a use case where I would like to use the nearest neighbour approach to fill some gaps in my data. In other words, I would like to resample a Scene to its own same area again. If I do this, currently the resampling is skipped due to:

pyresample/pyresample/resampler.py

Lines 121 to 122 in 3c07255

	if self._geometries_are_the_same():
	return data

which is (rightfully) implemented for optimisation purposes.

A solution could be to add a kwarg skip_if_same_geometry=True to resample, in order to be able to force the resampling when False is passed.

[djhoese]

Hhhmmm good point. 🤔

I don't like the idea of yet another keyword argument, especially one with such a long name. It also doesn't help that the "equivalent" check is such a long complicated method. I wouldn't be super against requiring the user to do the equality check and skip resampling themselves if it was as easy as src == dst.

The keyword argument could be as simple as "force=True" and would represent the "amount of work" an algorithm should do to get to a solution. I could see this being expanded in the future to limit how long an algorithm should try to find the "best" result or rather choosing a "short" path versus a "long" path algorithm.

Another option, that I think would cause more bad than good, would be to make the destination area being passed to the resampler optional. If it isn't passed then we know that the user wants to resample to the same source area and fill gaps.

Another other option, would be to interpret some existing keyword arguments to mean this same thing. Like if the geometries are the same but mask_area=True. Ugh that might have unintended consequences. Maybe we do just go with a keyword argument for now...or maybe we make the equality check easier to access? Like an src.equivalent(dst) and make callers responsible for skipping resampling if they want?

[ameraner]

yeah, I think a force kwarg could do as well, is clear and could be useful indeed also for other purposes. I wouldn't remove the check since I guess it is optimising quite a bit of other use cases.. For reference, those lines were added with #520

[djhoese]

Well and I added that exact check after Martin had used a src == dst and it ended up computing dask arrays on swath definitions. That's why that equality check is so complicated now, to avoid dask computations.