• Checked for duplicates

Describe the bug

Distributed RDataFrame has support for friend trees, but it seems something is missing. See the gist at https://gist.github.com/vepadulano/b42343bff7297958c46675577bce46a9 :

1. Two RDF are created and one column is filled
2. They are both snapshotted to disk and merged into a single file through hadd
3. A TChain is created with one column from the merged file and a friend TChain is attached to it with the second column from the merged file
4. A distributed RDataFrame with spark is created using the TChain as input, then one histogram per column is booked and drawn to a canvas
5. The operation fails with

   TypeError: Template method resolution failed:
     none of the 4 overloaded methods succeeded. Full details:
     ROOT::RDF::RResultPtr<TH1D> ROOT::RDF::RInterface<ROOT::Detail::RDF::RRange<ROOT::Detail::RDF::RLoopManager>,void>::Histo1D(experimental::basic_string_view<char,char_traits<char> > vName) =>
       runtime_error: Unknown column: myfriend.rnd
   

Expected behavior

The program should not fail, in fact substituting the distributed rdataframe object with a plain rdataframe gives the correct output image

To Reproduce

1. Source an environment with ROOT master
2. download the linked gist
3. python friendtrees_spark.py

Setup

Fedora 32
ROOT version: master
Built from source

Additional context

Thanks to @Zeguivert for originally reporting this issue