Problem with RDataFrame Alias when using `.` in the alias name.

[jminguezCern]

Check duplicate issues.

• Checked for duplicates

Description

When trying to use RDataFrame Alias using an alias that contains a . in the string the following error message is thrown:
RDataFrame::Alias: cannot define variation "reco.el_e_NOSYS_B". Not a valid C++ variable name.

This is because the function CheckValidCppVarName() checks if the new alias characters are either a letter, an underscore or a number. Since the . is not identified as a neither of these cases the error is thrown.

This was found when matching 2 trees and trying to alias a systematic variation column of the matched tree.
For example:
Original column in the matched tree: truth.variation_customName
Desired alias: truth.variation_commonName

Reproducer

The minimal reproducer can be found here in the following cern box link: https://cernbox.cern.ch/s/c12lOxGVx0vahWy

ROOT version

ROOT Version: 6.40.00
Built for linuxx8664gcc on May 26 2026, 15:56:51
From tags/v6-40-00@v6-40-00

Installation method

Build from source

Operating system

Ubuntu 25.10

Additional context

No response

[TomasDado]

Maybe to add that this is also relevant in the context of RNTuples, where we understood that we always have to add these prefixes to the columns for dataset joints.

[vepadulano]

Hi @jminguezCern ,

Thanks for the report! I am hesitant on this one. Column names defined by using the RDataFrame API (i.e. via Define or Alias) have always only supported valid C++/Python identifiers. Consider that also RNTuple explicitly forbids the character . as part of the field name. Is there any way we can imagine to create a path towards migrating away from the usage of column names (whether via Define or Alias) which contain .?

[TomasDado]

Hi @jminguezCern ,

Thanks for the report! I am hesitant on this one. Column names defined by using the RDataFrame API (i.e. via Define or Alias) have always only supported valid C++/Python identifiers. Consider that also RNTuple explicitly forbids the character . as part of the field name. Is there any way we can imagine to create a path towards migrating away from the usage of column names (whether via Define or Alias) which contain .?

Hmm but I thought for RNTuple we do explicitely need to pass RNTupleName.branchName in case of joint datasets.

[vepadulano]

Hmm but I thought for RNTuple we do explicitely need to pass RNTupleName.branchName in case of joint datasets.

Apologies for the confusion. Yes, the implementation of the JOIN operation on RNTuple datasets requires the user to be explicit about which of the joined datasets they want to access. What I meant is that the . character is disallowed in any field name within a single RNTuple dataset schema. What you are instead trying to do is to call Alias on an existing column name coming from a joined dataset. Still, the rule that in-memory columns must have names which are valid C++ identifier holds. So currently there is no way to have neither Define nor Alias accept a name with .. I find it would be a bad design decision to lift this restriction, although I can see potentially some inconvenience in having to remember this for this particular scenario. Would it be a problem to also use the _ character in this case?

[TomasDado]

Thanks for the explanation and sorry for my misunderstanding. If Alias allows to pass columns to be rename with a full stop, but requires the new alias name to not have a full stop in there, this is perfectly fine for us. Maybe slightly confusing if the fill stops are required in some case (joint datasets) and not allowed in another case (alias names) but as long as this is well documented it should be fine.