Davis Bennett

@sumanthratna the `axes` argument is a dictionary of `{axis_index: downsampling_factor`}, e.g `{0: 2, 1: 2}` to downsample by 2 along the 0th and 1st axes, does that make sense?

I've been using dask for multiresolution pyramids with my own tiny library ([xarray-multiscale](https://github.com/JaneliaSciComp/xarray-multiscale)) for a few months. At the moment I'm using `dask.array.coarsen`, so I don't do gaussian downsampling, but...

In my application I'm only ever downscaling, never upscaling, so I would use `pyramid_reduce` if it was available.

Thinking about the pros and cons of expanding the configuration surface area to cover this. As a "pro", it would erode some of the differences between n5 and zarr, as...

If, as @rabernat suggests, s3fs is doing expensive full directory listings, then using `/` as a dimension separator instead of the default `.` might give an immediate performance boost. I.e.,...

I think you would do something like this: `store = FSStore('s3://bucket', dimension_separator = '/', storage_options = {'anon' : False, 'use_ssl' : True, ...})`, with all your other kwargs to `S3FileSystem`...

@fAndreuzzi you might want to pick up from my efforts here: https://github.com/d-v-b/zarr-python/tree/boundary_chunk_optimization The strategy I'm taking is to intercept the arguments to `is_total_slice` with a new function `trim_chunks`, which basically...

Interesting! Something is a list when it shouldn't be... I can dig in :)

This is going to take a bit, since it kind of spiraled into a potential refactoring of how slicing works :cold_sweat:

looks like CI is failing for python 3.7 due to some type annotation stuff... do we need to support 3.7 still?