.shape attribute of arrays created with from_h5ad generates error

[bkmartinjr]

Create an scgroup with from_h5ad, open a component array (eg, obs) and inspect its shape parameter -- generates an error.

Test case: run the following from the apis/python directory and you will get an error on the line that prints the shape parameter.

Output:

$ python temp/test1.py
0.14.1
/home/bruce/projects/TileDB-SingleCell/venv/lib/python3.9/site-packages/anndata/compat/__init__.py:232: FutureWarning: Moving element from .uns['neighbors']['distances'] to .obsp['distances'].

This is where adjacency matrices should go now.
  warn(
/home/bruce/projects/TileDB-SingleCell/venv/lib/python3.9/site-packages/anndata/compat/__init__.py:232: FutureWarning: Moving element from .uns['neighbors']['connectivities'] to .obsp['connectivities'].

This is where adjacency matrices should go now.
  warn(
Traceback (most recent call last):
  File "/home/bruce/projects/TileDB-SingleCell/apis/python/temp/test1.py", line 16, in <module>
    sys.exit(main())
  File "/home/bruce/projects/TileDB-SingleCell/apis/python/temp/test1.py", line 12, in main
    print(obs.shape)
  File "tiledb/libtiledb.pyx", line 3607, in tiledb.libtiledb.Array.shape.__get__
  File "tiledb/libtiledb.pyx", line 3014, in tiledb.libtiledb.ArraySchema.shape.__get__
  File "tiledb/libtiledb.pyx", line 2329, in tiledb.libtiledb.Domain.shape.__get__
  File "tiledb/libtiledb.pyx", line 2188, in genexpr
  File "tiledb/libtiledb.pyx", line 2188, in genexpr
  File "tiledb/libtiledb.pyx", line 2056, in tiledb.libtiledb.Dim.shape.__get__
  File "tiledb/libtiledb.pyx", line 2041, in tiledb.libtiledb.Dim._shape
AttributeError: 'NoneType' object has no attribute 'item'

Test case:

import sys
import tiledb
import tiledbsc


def main():
    print(tiledb.__version__)
    sc = tiledbsc.SCGroup("./temp/test1", verbose=False)
    sc.from_h5ad("./anndata/pbmc3k_processed.h5ad")

    obs = tiledb.open("./temp/test1/obs")
    print(obs.shape)


if __name__ == "__main__":
    sys.exit(main())

[johnkerl]

The sparse matrices in the group all have exceptions on A.shape; the dense ones do not:

#!/usr/bin/env python

import tiledb
import tiledbsc

#g = tiledbsc.SCGroup("test")
#g.from_h5ad("/Users/johnkerl/scdata/anndata/pbmc-small.h5ad")

for x in ['X/data', 'obs', 'var', 'obsm/X_pca', 'varm/PCs', 'obsp/distances']:
    with tiledb.open('./test/' + x) as A:
        print()
        print(A.uri)
        #print(A.schema)
        try:
            print(A.shape)
        except:
            print('exc')

#./test/X/data         sparse=True,
#./test/obs            sparse=True,
#./test/var            sparse=True,
#./test/obsm/X_pca     sparse=False,
#./test/varm/PCs       sparse=False,
#./test/obsp/distances sparse=False,

#./test/X/data         exc
#./test/obs            exc
#./test/var            exc
#./test/obsm/X_pca     (80, 19)
#./test/varm/PCs       (20, 19)
#./test/obsp/distances (80, 80)

I'll check on what the expected behavior is for .shape on a sparse array (throwing an exception doesn't feel right).

[johnkerl]

I checked with the team and this is a bug in TileDB-Py:

• Sparse array with int dims has a .shape
• Sparse array with float or string dims does not
  • With float dims, .shape correctly does TypeError: shape valid only for integer domains
  • With string dims, we get the less-helpful error message reported here

[johnkerl]

TileDB-Py PR: TileDB-Inc/TileDB-Py#1055

[johnkerl]

The true ask here is handled in #125.