In YAML parser, use a placeholder struct for empty sequences

[bartgol]

Motivation

In EAMxx, we were forced to add custom YAML tags (like !!ints) to specify that a YAML sequence was meant to be a sequence of a particular type. The need comes from allowing strong type safety when dealing with empty lists, so that parsing my_double_vec: [] would indeed yield a param list storing a std::vector<double> of length 0. However, this added some clunkyness to our yaml file construction in EAMxx.

This PR changes approach: when parsing a YAML file and encountering an empty sequence, we store a EmptySeq object. In ParameterList, we then allow to call get<T>("my_seq") with any std::vector<S>: if the param list is non-const, we will replace the EmptySeq entry with a std::vector<S>, while if the param list is const we return a const ref to a static local var.

Testing

Added testing for this feature

Additional Information

There is a drawback in this approach: if the user accidentally does something like

ekat::ParameterList pl = ...;
auto& my_doubles = pl.get<std::vector<int>>("my_doubles");

and my_doubles was indeed still set to an instance of EmptySeq, it will replace it with a std::vector<int>, which was not intended. However, I consider this a minor corner case. Getting rid of custom-made YAML tags is probably outweighing the "risk".

[jeff-cohere]

Hey Luca. I see this in the testing log:

Here's the content of /home/runner/_work/EKAT/EKAT/ctest-build/debug/Testing/Temporary/LastTestsFailed_20251008-1701.log:
110:yaml_parser

The LastTestsFailed log file from CTest only lists the test numbers/names that fail. Is the corresponding LastTests_xyz.log too long to include in the log file? I bet it probably is. :-)

[bartgol]

Hey Luca. I see this in the testing log:

Here's the content of /home/runner/_work/EKAT/EKAT/ctest-build/debug/Testing/Temporary/LastTestsFailed_20251008-1701.log:
110:yaml_parser

The LastTestsFailed log file from CTest only lists the test numbers/names that fail. Is the corresponding LastTests_xyz.log too long to include in the log file? I bet it probably is. :-)

Yeah, I forgot to retest the last commit before pushing, a classic :)

The log is uploaded as an artifact in the following step, to keep the terminal output more "scrollable".

[jeff-cohere]

Nice solution, given the problem. I agree with your assessment of the risks.

[jgfouca]

Looks good. Those ! tags were causing issues