Cannot convert 0 to a BigInt error when loading uint64 Zarr data

[dchen116]

Description

When loading Zarr v2 segmentation data with uint64 dtype, I get a JavaScript TypeError:

Cannot convert 0 to a BigInt

Analysis

The error appears to be related to how fill_value is handled for 64-bit integer types. The fill_value: 0 in the Zarr
metadata is parsed as a JavaScript Number, but BigInt typed arrays require BigInt values (e.g., 0n instead of 0).

Reference: zarr-specs#334 discusses this broader issue with
JSON encoding of int64/uint64 values in JavaScript.

Data Specifications

• Format: Zarr v2
• dtype: uint64 (<u8)
• fill_value: 0
• Chunks: 32x32x32
• Compression: zstd
• Multi-scale pyramid: 7 levels

Notes

• The same data loads correctly in Neuroglancer
• The data passes the OME-NGFF validator
• Small dense arrays may not trigger this bug (only occurs when sparse chunks need fill_value)

[will-moore]

Hi, thanks for the bug report and apologies for the delayed response.

I created a sample file (just random pixel values) and deleted the first chunk from each resolution and can see the error is coming from the zarrita.js library used to load zarr data.

It only occurs with Zarr v2 data - It seems that this has been fixed with Zarr v3, either in the spec or zarrita.js.

I put my sample file at https://ome.github.io/ome-ngff-validator/?source=https://uk1s3.embassy.ebi.ac.uk/idr/zarr/test-data/zarrv2_uint64_sparse.zarr - The thumbnail fails (as reported) but so does the loading of the first (0, 0) chunk for each resolution, which is just using zarrita.js directly. I'll create an issue on zarrita...

[will-moore]

The fix in zarrita at manzt/zarrita.js#342 has been merged (thanks to Trevor), so once that has been released, then updating zarrita should fix this bug.