Conversation
| data_shape = (100, 120) | ||
| proxy = mock.Mock(dtype=np.dtype('f4'), shape=data_shape, | ||
| spec=pp.PPDataProxy) | ||
| spec=pp.PPDataProxy, ndim=len(data_shape)) |
There was a problem hiding this comment.
OK, but I'm really surprised by this development.
Hitherto it was sufficient for a wrapped object to have just a .dtype and a .shape.
As ndim is a trivial consequence of shape, I wonder if this is a bit of a mistake in dask, and should be queried ??
Or am I missing something special about this particular item ?
There was a problem hiding this comment.
@pp-mo From what I understand there has been a bit of a refactor for dask version 2.0.0, so for whatever reason they have started to access the ndim attribute of dask.array like objects.
Perhaps raise your concerns with them in an issue, if you care, but I don't see this as a blocker to this PR. The horse has kinda bolted, as they have already cut dask version 2.0.0 and now 2.1.0.
There was a problem hiding this comment.
@bjlittle itch has been scratched
Close but no 🚬
⛈ in 🍵
|
@pp-mo Unless you've got other objections, I think this PR is good to go 😉 ... what do you think? |
|
Respect for patience @bjlittle ! |
2 participants
This PR undoes the temporary workaround implemented by PR #3350, which pinned
dask<2.The latest version of
daskversion2.1.0(only a few days old now) resolves the outstandingstack overflowissues avoided by our pin.Also see ESMValGroup/ESMValTool#1188 for further details.