Patrick Peglar

Note : #9240 may be relevant, at least to point#1 in the description above

> Here's a simpler example: > ``` . . . @delayed def get_id(x): return id(x) . . . get_id(a).compute() . . . with dask.config.set(scheduler='threads'): print(get_id(a).compute()) . . . ``` Actually,...

> Would you be interested in submitting a PR for this? P.S. I certainly don't mean to disparage the docs, which have really got so much better in recent years...

> how the dask scheduler can "replace" this dask.Array object with something else in the forked process **UPDATE:** I just discovered that the Dask scheduling is in fact copying other...

> I think serialisation really is involved ,after all, and is applied to the whole Python environment in some way. Update : I still don't get quite how+where this happens,...

That's interesting. I made a little test of mirroring + changing the config file for selective updating. A couple of things emerged : **Firstly** if a package is not in...

Rebased + targetting new feature-branch "FEATURE_split_attrs"

**UPDATE:** I have since converted the rough plan outlined in this comment into tickets on a new [project to cover all of this work](https://github.com/orgs/SciTools/projects/5) ---- *Note :* a rough plan...

I've now created a [project](https://github.com/orgs/SciTools/projects/5) for this entire effort, and re-targetted this PR onto a [dedicated feature branch](https://github.com/SciTools/iris/tree/FEATURE_split_attrs). So this PR is now represents only the very first step of...

NOTE: rebased onto updated feature-branch + fixed, following [the mergeback](https://github.com/SciTools/iris/pull/4997) which contained #4803