Hi @bouweandela
The existing test breakage is a fairly minor matter, but I think it indicates that removing line 287b is causing the da.from_array to do trial access on the data -- so I think it is probably better to continue to provide a meta key in this case.

However, I am also concerned that ideally we should also be addressing #5800 if possible (!)
In which case, I think this should have a meta of type np.ma.MaskedArray in place of np.ndarray, when appropriate.

From my reading of dask docs, I think it may also be better to pass a zero-shaped instance instead of a class, since the from_array docs don't actually mandate that usage.
Also, it actually doesn't work if I pass meta=np.ma.MaskedArray : This gives a TypeError: MaskedArray.__new__() got an unexpected keyword argument 'shape'.

So I think something like da.from_array(real, meta=real.__class__((), dtype=real.dtype)) works here, and is probably a better way.

Does this make sense to you ?

Codecov Report

All modified and coverable lines are covered by tests ✅

Project coverage is 89.78%. Comparing base (7c313ff) to head (00af67e).
Report is 21 commits behind head on main.

@@            Coverage Diff             @@
##             main    #5801      +/-   ##
==========================================
+ Coverage   89.76%   89.78%   +0.01%     
==========================================
  Files          93       93              
  Lines       22982    23004      +22     
  Branches     5006     5015       +9     
==========================================
+ Hits        20630    20654      +24     
+ Misses       1622     1620       -2     
  Partials      730      730              

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

@bouweandela I suspect that the pre-commit failure is unrelated to this pull request. When I run pre-commit run --all-files on the main branch I get the same error.

Yes, puzzling. We are looking into it...

@bouweandela I suspect that the pre-commit failure is unrelated to this pull request. When I run pre-commit run --all-files on the main branch I get the same error.

Yes, puzzling. We are looking into it...

Ok I think we know what this is now -- fix coming soon.
It's only "odd" because the flake8 error code is bugged !

UPDATE: see #5925 -- we think this fixes it

Thank you for reviewing @pp-mo! I just merged the latest main into this branch and the pre-commit test is passing fine now 🥳

However, I think we need to rethink the merge, since some changes from #5588 seem to have got lost.
I guess that means that merging back into this will cause problems
( I expect at least some tests ought then to fail, otherwise where is the testing of those new keywords ?? )
So I haven't updated with main, as you may want to resolve this in your own way.

I refactored the as_lazy_data function a bit, but the functionality is still there (see #5801 (comment)). If you prefer, I can undo that?

If you prefer, I can undo that?

Thanks no, I just hadn't understood fully what you were doing. It's good !