feat: add dask_to_root

[zbilodea]

No description provided.

[lgray]

Just a stylistic comment but I think the function would be better situated within uproot as uproot.dask.to_root, thoughts?

I think it would match dask-awkward and awkward better in that regard.

@jpivarski

[jpivarski]

You mean the function would be an attribute of another function, uproot.dask?

I've gone back and forth on that—there used to be an ak.to_parquet.dataset for converting already-written Parquet files into a dataset (the directory with metadata convention). But by not being in the same level of hierarchy as the rest of the ak.* functions, it was hard to discover.

I understand the logic of writing-from-Dask being associated with reading-from-Dask. Uproot doesn't already have the flat bundle of functions that Awkward Array has, so there's less to maintain. But still, locating functions as attributes on other functions is unusual enough in Python as to be an antipattern—if someone is looking for it, they're unlikely to think of looking there.

Perhaps it should just be a naming convention: uproot.dask_write? From a typing point of view, that only turns a dot into an underscore, but from an organizational point of view, there's no nesting of functions. (I don't think it needs "to ROOT" because this is Uproot—it's only ever going to read from or write to ROOT.)

[lgray]

dask shouldn't be a top level dependency of the uproot package? @jpivarski

I suppose below we should import dask within dask_write such that it complains when dask is not available, but doesn't impede more standard functioning of uproot.

[jpivarski]

Correct: dask and dask-awkward should not be strict dependencies for Uproot. These two lines should be removed.

[lgray]

@jpivarski ok that naming scheme makes sense, and I am fine with it!

[jpivarski]

In Uproot and Awkward, name the tests by PR number:

tests/test_1085_dask_to_root.py

[lgray]

OK I checked it with nanoevents, repartitioning, concatenating, all using a distributed Client and it seems to be robust. It touches data correctly and fills the output TTree well.

Looks good to me :-)

[jpivarski]

This is good; just be sure to

• remove version.py from this PR, now that it's auto-generated
• remove dask and dask-awkward from the list of dependencies

After that, request a review from me and this will be mergeable.

[jpivarski]

Correct: dask and dask-awkward should not be strict dependencies for Uproot. These two lines should be removed.

[jpivarski]

Also remove the 7 output files before merging:

src/uproot/my-output/*.root

But, after they're removed (as well as the if __name__ == "__main__" part), this is ready to merge!

[jpivarski]

Nevermind; I was looking at the wrong diffs. The temporary output files have already been removed.