numba.errors.TypingError: Failed in nopython mode pipeline (step: nopython frontend) non-precise type pyobject

Hi,
I am having troubles in using the scanpy.tl.neighbors function on data with more than 2^13=8191 rows while using the "backed" mode.

My machine has ubuntu 18.4, I run the code with a conda environment.
I am using these versions of libraries:
scanpy 1.4.5
hdf5 1.10.4
numpy 1.17.4
anndata 0.7rc1

This is an example of code reproducing the problem:

import scanpy as sc
import numpy as np
import anndata
import pandas as pd

data=np.ones((8192,2))

obs = pd.DataFrame()
obs['obs'] = np.arange(data.shape[0])
var = pd.DataFrame()
var['var'] = np.arange(data.shape[1])
adata = anndata.AnnData(X = data, obs=obs)
adata.filename='back.h5ad'

sc.pp.neighbors(adata, n_neighbors = 10) 
print('done with knn graph')

The returned error is the following:

Traceback (most recent call last):
  File "create_files.py", line 15, in <module>
    sc.pp.neighbors(adata, n_neighbors = 10) 
  File "/home/sara_f/anaconda3/envs/sci/lib/python3.7/site-packages/scanpy/neighbors/__init__.py", line 110, in neighbors
    random_state=random_state,
  File "/home/sara_f/anaconda3/envs/sci/lib/python3.7/site-packages/scanpy/neighbors/__init__.py", line 691, in compute_neighbors
    X, n_neighbors, random_state, metric=metric, metric_kwds=metric_kwds)
  File "/home/sara_f/anaconda3/envs/sci/lib/python3.7/site-packages/scanpy/neighbors/__init__.py", line 248, in compute_neighbors_umap
    angular=angular, verbose=verbose,
  File "/home/sara_f/anaconda3/envs/sci/lib/python3.7/site-packages/umap/umap_.py", line 269, in nearest_neighbors
    rp_forest = make_forest(X, n_neighbors, n_trees, rng_state, angular)
  File "/home/sara_f/anaconda3/envs/sci/lib/python3.7/site-packages/umap/rp_tree.py", line 715, in make_forest
    for i in range(n_trees)
  File "/home/sara_f/anaconda3/envs/sci/lib/python3.7/site-packages/umap/rp_tree.py", line 715, in <listcomp>
    for i in range(n_trees)
  File "/home/sara_f/anaconda3/envs/sci/lib/python3.7/site-packages/umap/rp_tree.py", line 573, in make_tree
    return make_euclidean_tree(data, indices, rng_state, leaf_size)
  File "/home/sara_f/anaconda3/envs/sci/lib/python3.7/site-packages/umap/rp_tree.py", line 453, in make_euclidean_tree
    data, indices, rng_state
  File "/home/sara_f/anaconda3/envs/sci/lib/python3.7/site-packages/numba/dispatcher.py", line 401, in _compile_for_args
    error_rewrite(e, 'typing')
  File "/home/sara_f/anaconda3/envs/sci/lib/python3.7/site-packages/numba/dispatcher.py", line 344, in error_rewrite
    reraise(type(e), e, None)
  File "/home/sara_f/anaconda3/envs/sci/lib/python3.7/site-packages/numba/six.py", line 668, in reraise
    raise value.with_traceback(tb)
numba.errors.TypingError: Failed in nopython mode pipeline (step: nopython frontend)
non-precise type pyobject
[1] During: typing of argument at /home/sara_f/anaconda3/envs/sci/lib/python3.7/site-packages/umap/rp_tree.py (160)

File "../../../../../anaconda3/envs/sci/lib/python3.7/site-packages/umap/rp_tree.py", line 160:
def euclidean_random_projection_split(data, indices, rng_state):
    <source elided>
    """
    dim = data.shape[1]
    ^

This error may have been caused by the following argument(s):
- argument 0: cannot determine Numba type of <class 'h5py._hl.dataset.Dataset'>

Additionally, I've noticed that disabling numba jit (setting the environment varable DISABLE_JIT: 0 with a yaml file named .numba_config.yaml) it works, instead.

Could you help me understand how to solve the problem?
I am available for further details if necessary.
Thanks

Start with the nearest_neighbors call in umap/rp_tree.py shown in the traceback and reproduce the failure using the provided 8192-row backed AnnData example. Investigate the interaction between the JIT path and the h5py Dataset, then verify that the example completes with JIT enabled without regressing ordinary array inputs.