Add RDataFrame equality comparision operator

[ynikitenko]

Explain what you would like to see improved and how.

I use RDataFrame from Python. This is very convenient and efficient, since calculations go in C++. However, a high-level object in Python (the same holds for C++) may use an equality comparison operator.

How it could be implemented:

• since there is a constructor of RDataFrame(tree name, file path), one may simply save the strings for the tree name and file path and compare those. This would give correct comparison right after the initialization.
• If RDataFrame is initialized from a class without a comparison operator, it is compared unequal. This would be similar to a Python lambda, which are not equal even with exactly same code.

The benefit: this may save computational power, since if there are two same RDataFrames, one can use only one of them (provided that the operations are indeed efficient, as the class promises).

A potential caveat would be if we define a column in one instance of the "same" RDataFrame and do not define that in another. In that case it would be easier to change the "equality state" of the RDataFrame after we make some operations with that.

ROOT version

any

Installation method

any

Operating system

any

Additional context

It would be useful to add equality comparison operators (normal in object-oriented world) to other ROOT classes as well (if those are missing). A remotely related issue was posted on the ROOT Forum.

[ferdymercury]

It would be useful to add equality comparison operators (normal in object-oriented world) to other ROOT classes as well (if those are missing). A remotely related issue was posted on the ROOT Forum.

Related discussion: #19038

[vepadulano]

HI @ynikitenko ,

Thank you for reaching out. RDataFrame is a complex piece of technology (as any other data frame implementation). It involves a dataset (which can be arbitrarily complex in terms of dataset schemas and relations) and a computation graph (with a myriad of different types of operations). As such, it is not clear (at least to me) how to uniquely identify an RDataFrame object in a programmatic way. Thus, it is at the same time not easy to identify what constitutes the "equivalence operator" for two RDataFrame objects. The two scenarios you list merely scratch the surface, and I can imagine many other scenarios where if we applied the prescriptions you describe the equality will be simply wrong.

I believe something like this would involve a lot of design thinking, discussion with users (you at the front of course since it's your request, but also many others) and careful implementation. Given the magnitude of this effort, I cannot say for sure if and when this could be implemented. I believe a prerequisite for all of the above is understanding what is the actual relevance and impact of this for your own work. Could I ask you to reach out to vincenzo.eduardo.padulano@cern.ch so we could set up a meeting and discuss this further?

[ynikitenko]

Dear @vepadulano , I wrote an email to you. I'm not sure I can arrange a meeting in the near future, but I wrote the basic idea and put links. Let us stay in touch (you can remove your email from the comment above) - I can always reply to an email.