SetBranchAddress does not work with np.int16 and uint16 in PyROOT

ROOT
1

Hello,

This bug exists in 6.34.04 and master (today) but not in 6.32.02 and 6.30.06.

It seems SetBranchAddress in PyROOT stopped accepting numpy’s short: uint16 and int16:

import ROOT
import numpy as np

a = np.zeros(3, np.uint16)
t = ROOT.TTree("t", "t")
t.Branch("b", a, "b/s")
t.Print()
t.SetBranchAddress("b", a)

In the working versions, there is no error, in the newer ROOTs the output is:

******************************************************************************
*Tree    :t         : t                                                      *
*Entries :        0 : Total =             754 bytes  File  Size =          0 *
*        :          : Tree compression factor =   1.00                       *
******************************************************************************
*Br    0 :b         : b/s                                                    *
*Entries :        0 : Total  Size=        465 bytes  One basket in memory    *
*Baskets :        0 : Basket Size=      32000 bytes  Compression=   1.00     *
*............................................................................*
Traceback (most recent call last):
  File "/tmp/./root_branch_test.py", line 10, in <module>
    t.SetBranchAddress("b", a)
  File "/home/lewhoo/software/root/root-6.34.04/lib/ROOT/_pythonization/_ttree.py", line 236, in _SetBranchAddress
    data_type = _determine_data_type(addr)
                ^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/home/lewhoo/software/root/root-6.34.04/lib/ROOT/_pythonization/_ttree.py", line 204, in _determine_data_type
    return _get_cpp_type_from_numpy_type(addr.__array_interface__["typestr"][1:])
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/home/lewhoo/software/root/root-6.34.04/lib/ROOT/_pythonization/_rvec.py", line 90, in _get_cpp_type_from_numpy_type
    raise RuntimeError("Object not convertible: Python object has unknown data-type '" + dtype + "'.")
RuntimeError: Object not convertible: Python object has unknown data-type 'u2'.

Platform: Fedora 40

2

Hi @LeWhoo,

Thank you for your question. I think maybe @jonas can help.

Cheers,
Dev

3

I guess this might be related to issue PyROOT multidimenstional int16_t arrays stopped working in 6.34 · Issue #17841 · root-project/root · GitHub

4

Could be, but not in a very obvious (to me) manner. In my case it occurs also for 1D arrays, goes from numpy to C++, not from C++ to numpy, and results in an error message.

5

The linked issue is a different one. However, we are also working in fixing the problem that you reported:

6

Thanks. I can see that the pull request in the cited issue has been merged. But it does not fix the problem discussed here…

7

To not forget about this, I’d suggest creating a specific issue for these two data types on Github, (or feel free to submit a new Pull Request if you know what change is needed in the map)

8

Created an issue:

9

This topic was automatically closed 14 days after the last reply. New replies are no longer allowed.