Python getattr() function in newer ROOT versions

ROOT
1

I need to find a replacement for the last line (check if the object is iterable / a dict).
It seems the “if” misbehaves in newer ROOT versions (it prints “???”).

import ROOT
o = ROOT.TH1F("h", "h", 100, 0., 1.)
if getattr(o, '__iter__', False): print("???")
2

Dear @Wile_E_Coyote ,

The snippet you present is not really misbehaving: the histogram can be iterated over the bins, see the following example.

import ROOT
import numpy as np

# Create a histogram
h = ROOT.TH1D("h", "h", 10, -10, 10)
# Create sample data
data = np.random.normal(0, 2, 10000)
# Fill histogram with data
h.Fill(data)
# Reduce bin content to make the following print statements more evident
h *= 0.01
for bin in h:
    print("*" * int(bin))

So indeed the __iter__ method is well defined. Note that this functionality is currently very rudimentary, we have plans to enhance this interoperability of histograms with common Python patterns via the GitHub - scikit-hep/uhi: Universal Histogram Interface

Cheers,
Vincenzo

1 Like
3

How can I get the old behavior?

4

Dear @Wile_E_Coyote ,

I am not sure I understand. If you mean “how can I have a ROOT histogram object that does not define the __iter__ method”, then I believe this is not possible, it’s just how the class functionality works. If the question is “how can I get a falsey value when I’m trying to access the __iter__ method of a ROOT histogram class?”, then I would approach it differently and would have a series of classes that you want to return a falsey value for in that if branch. Something like

if isinstance(o, myclass): # do stuff
if getattr(o, "__iter__"): # do stuff

Can be generalised by having a Python dictionary where the key is the class type itself.

Let me know if I missed something from your question.

Cheers,
Vincenzo

5

No, the actual code (which recursively processes what it gets as an input) breaks because the new ROOT says histograms (and probably some other objects now) have an “__iter__” (it then immediately dies because it tries to call “items()” on them, to process them recursively).
This problem seems to have been introduced in ROOT 6.32 (older versions behave well).

6

Dear @Wile_E_Coyote ,

__dict__ is even more common than __iter__. Almost every Python object has a dict (by default).

>>> class MyClass: ...
... 
>>> a = MyClass()
>>> a.__dict__
{}

ROOT objects do not escape this rule. In fact even on older ROOT versions e.g. 6.28 you will see that ROOT Python objects have this method, following Python standards:

[vpadulan@lxplus932 ~]$source /cvmfs/sft.cern.ch/lcg/views/LCG_104/x86_64-el9-gcc13-opt/setup.sh
[vpadulan@lxplus932 ~]$python
Python 3.9.12 (main, Jul 11 2023, 14:40:15) 
[GCC 13.1.0] on linux
Type "help", "copyright", "credits" or "license" for more information.
>>> import ROOT
>>> ROOT.__version__
'6.28/04'
>>> h = ROOT.TH1D("h", "h", 10, 0, 10)
>>> h.__dict__
{}
7

I misspelled it (see my previous post).

8

Dear @Wile_E_Coyote ,

ROOT 6.32 has seen a major update of the underlying machinery of our dynamic python bindings (cppyy). Namely, ROOT now is up-to-date with the latest functionalities of cppyy. This is by and large a huge upgrade of functionality. Personally, I believe the nature of Python objects is so dynamic that I find it a bit fragile to rely on the presence of __iter__ for any particular class, ROOT or not (see also this SO answer). From the little information provided so far, I would still suggest you to apply a separate logic for ROOT histograms, if you need to treat them differently, as I showed in a previous message. If you want you can share more information on your use case and maybe we can find an even nicer solution.

I will also invite @danilo or @jonas to share their thought on this.

Cheers,
Vincenzo

9

I think I need a robust test to determine whether the object “o” is a genuine “dict” and not some strange ROOT object that pretends to be one.

I could try to use the following:

if hasattr(o, '__iter__') and hasattr(o, 'items'):

But … can you guarantee that you will NOT add the “items()” method to histograms (and possibly other relevant objects, e.g. graphs and so on)?

10

Dear @Wile_E_Coyote ,

No, there is no such guarantee, as there is no guarantee that any specific class will not get improved/extended in the future, I believe this is true for any software package. Of course, removing methods from a class would be a different story. If you really just need to call the items method on a certain class, you can just call hasattr(o, "items"). Again, if ROOT histograms specifically need special treatment, then they should be treated separately.

Cheers,
Vincenzo