id,node_id,number,title,user,state,locked,assignee,milestone,comments,created_at,updated_at,closed_at,author_association,active_lock_reason,draft,pull_request,body,reactions,performed_via_github_app,state_reason,repo,type
488243328,MDExOlB1bGxSZXF1ZXN0MzEzMzA4NzE4,3276,map_blocks,2448579,closed,0,,,19,2019-09-02T15:32:42Z,2019-10-17T01:30:30Z,2019-10-10T23:44:19Z,MEMBER,,0,pydata/xarray/pulls/3276,"<!-- Feel free to remove check-list items aren't relevant to your change -->

 - [x] xref #3147, closes #3323 
 - [x] Tests added
 - [x] Passes `black . && mypy . && flake8`
 - [x] Fully documented, including `whats-new.rst` for all changes and `api.rst` for new API

This is a second attempt. It works with Datasets also in that it can pass Dataset chunks to the user's function. I haven't figured out how to reconstruct a Dataset from things returned from the user's function. For DataArrays, I currently assume that the user function returns chunks of the same shape as the provided chunk.

A few questions:
1. I wonder if we should introduce a meta xarray object that describes the final result that user expects back i.e. a template DataArray or Dataset with chunks, dtype etc. specified. We could have a utility function that makes it easy to create these templates from existing objects.

2. If the user function doesn't return anything, (e.g. it writes to a file), how do I ""compute"" the graph? In this case, we could rename this function to `apply_blocks` and release that till we figure out point 1?

3. Should we also add `Dataset.to_delayed` and `DataArray.to_delayed`? This would make use-cases like the one in point 2 easier.

ping @mrocklin ","{""url"": ""https://api.github.com/repos/pydata/xarray/issues/3276/reactions"", ""total_count"": 0, ""+1"": 0, ""-1"": 0, ""laugh"": 0, ""hooray"": 0, ""confused"": 0, ""heart"": 0, ""rocket"": 0, ""eyes"": 0}",,,13221727,pull