State does not persist when xarray dimension name is a not valid Neuroglancer dimension name

[mkitti]

I have Zarr data generated via xarray.Dataarray.to_zarr. This includes the units in brackets: "xpix [µm]".

While Neuroglancer can view this data initially, it seems unable to serialize this state properly in the URL. Refreshing, sharing the URL from the address bar, or copying the JSON state fails to reproduce the state.

After testing I determined that Neuroglancer considers both the brackets and the µ as illegal characters.

{
  "shape": [
    5370,
    9,
    180,
    180
  ],
  "data_type": "int16",
  "chunk_grid": {
    "name": "regular",
    "configuration": {
      "chunk_shape": [
        32,
        9,
        45,
        45
      ]
    }
  },
  "chunk_key_encoding": {
    "name": "default",
    "configuration": {
      "separator": "/"
    }
  },
  "fill_value": 0,
  "codecs": [
    {
      "name": "bytes",
      "configuration": {
        "endian": "little"
      }
    },
    {
      "name": "zstd",
      "configuration": {
        "level": 0,
        "checksum": false
      }
    }
  ],
  "attributes": {},
  "dimension_names": [
    "volumes [s]",
    "planes [µm]",
    "xpix [µm]",
    "ypix [µm]"
  ],
  "zarr_format": 3,
  "node_type": "array",
  "storage_transformers": []
}

[mkitti]

My suggestion is that Neuroglancer should attempt to map the dimension names to a valid dimension name. One strategy would be to take the longest leading valid substring

For example, the dimension names could be mapped to the following.

{
  "dimensions": {
    "volumes": [
      1,
      ""
    ],
    "planes": [
      1,
      ""
    ],
    "xpix": [
      1,
      ""
    ],
    "ypix": [
      1,
      ""
    ]
  }
}

It would also be helpful if the validity of dimension names could also be expanded.

[mkitti]

Valid dimension names seem restricted here:

neuroglancer/src/coordinate_transform.ts

Line 789 in 78ad594

return name.match(/^[a-zA-Z][a-zA-Z_0-9]*['^]?$/) !== null;