Closed talonchandler closed 1 year ago
From the brief debugging I did, the bug points to page = self.reader.reader.coord_map[coord_reorder][1]
line 458 in zarr_converter
file. It starts the count of the pages at 511
and then process page by page sequentially (i.e 1,2,3).
One thing to mention is that this dataset failed to finish.
This is caused by the multi-page reader in waveorder.io
building index map from MM metadata. It expects the pages recorded in the MM header field (generated at the start of the acquisition) to be present, but since the acquisition did not finish, the index map is larger than the actual dataset.
I'll try to fix this in iohub.
The bug appears to be deeper in the stack: the OME-TIFF reader was constructing index maps using a magic-number offset which is inconsistent with what MM actually writes data with.
Edit: Both page indexing and byte offsets have some problems, and the later seems to be originating from the former.
Ready to close when we release iohub.
@edyoshikun and @dsundarraman reported that:
fails with this trace:
@edyoshikun you said that you've done some initial debugging on this issue? And that the acquisition may be incomplete (crash before completion?)?
@ziw-liu I'm tagging you in case this issue affects you and iohub.