issue_comments: 502721965

This data as json

html_url issue_url id node_id user created_at updated_at author_association body reactions performed_via_github_app issue
https://github.com/pydata/xarray/issues/3009#issuecomment-502721965 https://api.github.com/repos/pydata/xarray/issues/3009 502721965 MDEyOklzc3VlQ29tbWVudDUwMjcyMTk2NQ== 1217238 2019-06-17T15:04:40Z 2019-06-17T15:04:40Z MEMBER

Xarray definitely had a bug with ImplicitToExplicitIndexingAdapter. See https://github.com/pydata/xarray/pull/3027 for the fix.

Unfortunately, this triggers new failures from inside dask's blockwise_meta helper: ```

  return meta.astype(dtype)

E AttributeError: 'ImplicitToExplicitIndexingAdapter' object has no attribute 'astype' ```

The bigger issue this surfaces is that I'm not sure whether ImplicitToExplicitIndexingAdapter (and xarray's other custom array types used for data loading) should be in meta or not. These array types overload indexing, but they don't implement the rest of NumPy's ndarray API. So it's a little strange to see them in meta (and of course methods like astype don't work on them).

On the other hand, they do need to be preserved rather than always cast to NumPy arrays. Otherwise repeated indexing like x[i][j] doesn't get optimized by dask, which turns out to be a really important optimization.

 
{
    "total_count": 0,
    "+1": 0,
    "-1": 0,
    "laugh": 0,
    "hooray": 0,
    "confused": 0,
    "heart": 0,
    "rocket": 0,
    "eyes": 0
}
   454168102