Calling to_hist() and to_numpy() gives error when opening a CategoryAxis which has been saved to a root file

[jlabounty]

Hello,

I have been using uproot to save/load hist objects from root files. I have noticed that calling to_hist() on a THn object which contains a category axis yields a ValueError: could not broadcast input array from shape (N,...) into shape (N+1,...). This snippet of code reproduces the issue:

import hist
print(f'{hist.__version__=}')
import uproot
print(f'{uproot.__version__=}')
import numpy

cat_axis = hist.axis.IntCategory([1,2,3], label='Category')
reg_axis = hist.axis.Regular(100,0,100, label='Random')
h = hist.Hist(cat_axis)
h2 = hist.Hist(cat_axis, reg_axis)
h.fill(
    numpy.random.randint(1,4,1000)
)
h2.fill(
    numpy.random.randint(1,4,1000), numpy.random.normal(20,5,1000)
)

with uproot.recreate("bug_output.root") as f:
    f['h'] = h
    f['h2'] = h2

    print(f['h'])
    print(f['h2'])

with uproot.open("bug_output.root") as f:
    h_opened = f['h'] 
    h2_opened = f['h2']

print(h_opened)
print(h2_opened)

#this fails 
print(h_opened.to_hist())
print(h2_opened.to_hist())

#this also fails
print(h_opened.to_numpy()) #works, but the bin edges are wrong
print(h2_opened.to_numpy()) #fails

(gm2) jlab@SB3:~/github/g2_analysis/analysis$ python hist_uproot_category_bug.py 
hist.__version__='2.6.1'
uproot.__version__='4.3.3'
<TH1D (version 3) at 0x7fa5dc5650c0>
<TH2D (version 4) at 0x7fa5db88e1a0>
<TH1D (version 3) at 0x7fa5eb986710>
<TH2D (version 4) at 0x7fa5eb793400>
Traceback (most recent call last):
  File "/home/jlab/github/g2_analysis/analysis/hist_uproot_category_bug.py", line 39, in <module>
    print(h_opened.to_hist())
  File "/home/jlab/miniconda3/envs/gm2/lib/python3.10/site-packages/uproot/behaviors/TH1.py", line 206, in to_hist
    self.to_boost(metadata=boost_metadata, axis_metadata=boost_axis_metadata)
  File "/home/jlab/miniconda3/envs/gm2/lib/python3.10/site-packages/uproot/behaviors/TH1.py", line 322, in to_boost
    view.value = values
  File "/home/jlab/miniconda3/envs/gm2/lib/python3.10/site-packages/boost_histogram/_internal/view.py", line 59, in fset
    self[name] = value
  File "/home/jlab/miniconda3/envs/gm2/lib/python3.10/site-packages/boost_histogram/_internal/view.py", line 39, in __setitem__
    super().__setitem__(ind, value)  # type: ignore[no-untyped-call]
ValueError: could not broadcast input array from shape (4,) into shape (5,)

The 2D to_numpy() call fails with:

Traceback (most recent call last):
  File "/home/jlab/github/g2_analysis/analysis/hist_uproot_category_bug.py", line 40, in <module>
    print(h2_opened.to_numpy())
  File "/home/jlab/miniconda3/envs/gm2/lib/python3.10/site-packages/uproot/behaviors/TH2.py", line 96, in to_numpy
    values = self.values(flow=flow)
  File "/home/jlab/miniconda3/envs/gm2/lib/python3.10/site-packages/uproot/behaviors/TH2.py", line 52, in values
    values = numpy.transpose(values.reshape(yaxis_fNbins + 2, xaxis_fNbins + 2))
ValueError: cannot reshape array of size 408 into shape (102,5)

I have tested this on uproot version 4.3.3 and hist version 2.6.1. I think this is related to the handling of over/underflow bins discussed here. As far as I know ROOT doesn't have an exact analogue to the CategoryAxis, so should an error be raised when trying to write this object to a root file in the first place?

Thank you for the great deal of work which goes into maintaining this project!