Ignore errors when decoding attributes in netCDF3.py

[mannreis]

Faced some issues when using virtualizarr because of this

[martindurant]

If you have a link to an issue on vzarr, would appreciate including it here

[martindurant]

Tests appear broken by recent zarr release, which requires type to go Object->str in HDF. Would you have time to find and fix this?

[mannreis]

I don't have any open issue in virtualizarr nor here. I hot patched by environment so that my scripts don't fail when loading old netcdf datasets with non utf-8 encoded attributes.

I managed to reproduce the problem locally but by upgrading zarr to 3.0.10 the issue goes away

[martindurant]

I managed to reproduce the problem locally but by upgrading zarr to 3.0.10 the issue goes away

!!

So actually he only thing that's wrong here is the failing tests. I'll merge this and try to fix it, although this week is rammed.

[martindurant]

You are right, things appear to pass with release 3.0.10, but not with the main branch (which gives version 3.0.9+devXX). I wonder what is going on!

[mannreis]

In the meantime the zarr-python/main version is 3.1.0 so there's a change retriggering the tests will make them pass @martindurant.

I've opened #567 as side follow up on this testing debugging. Feel free to close it if it's not really desired to tests against a "stable" zarr release

[martindurant]

Now pulls zarr 3.1.0, but still fails :\ . Did you say (in the other PR) that you saw this pass?

[mannreis]

I checked now and with 3.1.0 it will fail!

[martindurant]

...but it just passed! I am so confused.

Do you happen to know what PRs in zarr are implicated?

[mannreis]

@martin I just saw zarr-developers/VirtualiZarr#676 which seems to track what we saw here but in virtualizarr. In a way it was good to revert #567 I don't have time to dive deeper into the issue but I could help reviewing or testing any changes here.