Hi,
@ryan-williams current example uses null as fill value, but also NaN is often used.
With our data (NaN as fill_value), if i do val zarrGroup: zarr.Group = get(Path("/home/lauri/Downloads/GFS/hycom_test").load[Group])

I get:

Not an array:
DecodingFailure(Double, List())
Not a group:
org.lasersonlab.zarr.Group$InvalidChild: Path /home/lauri/Downloads/GFS/hycom_test/lon/.zarray:
Not an array:
org.lasersonlab.io.FileNotFoundException
Not a group:
java.nio.file.NotDirectoryException: /home/lauri/Downloads/GFS/hycom_test/lon/.zarray
	at org.lasersonlab.zarr.Group$.$anonfun$apply$20(Group.scala:138)
	at scala.util.Either$LeftProjection.map(Either.scala:573)
	at org.lasersonlab.zarr.Group$.$anonfun$apply$19(Group.scala:133)
	at scala.util.Success.$anonfun$map$1(Try.scala:255)
	at scala.util.Success.map(Try.scala:213)
	at scala.concurrent.Future.$anonfun$map$1(Future.scala:292)
	at scala.concurrent.impl.Promise.liftedTree1$1(Promise.scala:33)
	at scala.concurrent.impl.Promise.$anonfun$transform$1(Promise.scala:33)
	at scala.concurrent.impl.CallbackRunnable.run(Promise.scala:64)
	at java.util.concurrent.ForkJoinTask$RunnableExecuteAction.exec(ForkJoinTask.java:1402)
	at java.util.concurrent.ForkJoinTask.doExec(ForkJoinTask.java:289)
	at java.util.concurrent.ForkJoinPool$WorkQueue.runTask(ForkJoinPool.java:1056)
	at java.util.concurrent.ForkJoinPool.runWorker(ForkJoinPool.java:1692)
	at java.util.concurrent.ForkJoinWorkerThread.run(ForkJoinWorkerThread.java:157)
Caused by: DecodingFailure(Double, List())

Same data works in Python Zarr.

Would be nice to allow e.g.

Array((1000, 1000))(1 to 1000000)

in addition to the current

Array(1000 :: 1000 :: ⊥)(1 to 1000000)

Currently Zarr arrays, and the ndarray.Vectors that typically back them, support Traverse operations as well as randomly-accessing individual elements.

Slicing along arbitrary dimensions would be good to add.

put a simple webapp on top of the zarr module that allows viewing (cloud-resident) Zarr (or HDF5) files in the browser

something like HDFView, but web-based, targeting data in the cloud, and working for multiple CDM-style formats

directory/file listings and tree-maps based on file sizes are features I've personally really missed so far while interacting with these formats

@tomwhite's old prototype for traversing HDF5 files (via NetCDF) using Spark is still in the singlecell module

It would be good to add parallelization options to the convert CLI; most likely a Spark-specific code-path will be necessary for that, though in general it would be great to plumb parallelization through cats.Traverse / using Cats' Parallel. Futures and cloud-functions might be workable there.

All dependencies of the zarr module are cross-published for JVM and JS except JBlosc.

Cross-publishing to JS would be "nice to have", either for use on the client in #11, and also to prepare for adding a scala-native target later.

Native support is currently blocked on Cats, which in turn needs ScalaCheck, and so is probably a few months out at least, but it would be really cool as a potential way to support R as well, or even be usable from Python, in case features make it in here that aren't in the reference python implementation.