Support conversion between vector<Any> and vector<typename T::value_type>

[dyackzan]

These changes make way for us to do general vector operations like iterate over, index, append, and remove using BT.CPP. Using vector<Any> objects on the blackboard allows us to create these operations as singletons, rather than requiring each operation to be templated for each vector<T>

The 2 primary components to this PR:

• In setOutput, convert vector<T> objects to vector<Any> before placing them on the blackboard
• In getInput, attempt to unwrap and convert vector<Any> to vector<T> if a vector<T> is expected

[nbbrooks]

Could you document the gcc and clang demangled output for the std::vector<std::string> for reference here in the PR if it isn't possible to add a unit test establishing the demangled output passes the regex max? Ideally we'd have a unit test for partial specialization as well.

[dyackzan]

I checked clang and gcc with the type demangler and get the the same output for the demangled typeid of a std::vector<string> for both:

std::vector<std::__cxx11::basic_string<char, std::char_traits<char>, std::allocator<char> >, std::allocator<std::__cxx11::basic_string<char, std::char_traits<char>, std::allocator<char> > > >

[nbbrooks]

Oops, when I hit approve I lost track of my tabs and thought I was approving a submodule update to point to this PR branch 🤦

[facontidavide]

looks like a good addition. Can you please:

• use pre-commit to fix the formatting?
• add a simple unit test too?

[dyackzan]

How does that look @facontidavide ?

[dyackzan]

Tagging @facontidavide , could you give me another review on this?

[dyackzan]

Friendly ping @facontidavide , please let me know if there's anything else you'd like me to change here. We want to keep our fork as up to date with your base as possible

[facontidavide]

Ok, I am finally looking into this.

  std::string xml_txt_bad = R"(
    <root BTCPP_format="4" >
      <BehaviorTree>
        <Sequence name="root_sequence">
          <OutputVectorStringNode/>
          <InputVectorDoubleNode double_vector="{string_vector}"/>
        </Sequence>
      </BehaviorTree>
    </root>)";
    
ASSERT_NO_THROW(tree = factory.createTreeFromText(xml_txt_bad));

In my opinion this is SUPPOSED to throw. You are basically bypassing the entire type safety mechanism, unless you rely on the fact that string is the usual "wildcard" that has conversion to numbers.

[facontidavide]

Another problem I see is that this will not play nicely with this:

https://github.com/BehaviorTree/BehaviorTree.CPP/blob/master/examples/t13_access_by_ref.cpp

If I try to access any vector by reference, I will find a vector<Any>. This will break other people code.

[facontidavide]

Summary:

• you basically bypass the entire type safety checks for any vector<T>.
• accessing the blackboard by reference will return an unexpected type.
• the unit test doesn't seem meaningful to me, in terms of desired functionality. What you want to show is a generic node that can access an existing port as vector<Any>

I understand your problem and the annoyance of templates, and I would like to find a solution, but I am not sure that this is the one I want to merge.

I was thinking about doing the conversion to vector<Any> lazily, for instance you have your input port AnyVector and it should be able to connect and read from vector<string>. The problem is when writing back into the blackboard, because we don't want to change the original type.

A potential solution might be to store a to/from vector<Any> "converters" and run them only when requested, to avoid altering the original type in the blackboard.

On the other hand, what the "right" type of the blackboard is is based on some "fuzzy" criteria, i.e. the order of the XMl parsing...

I am not sure how to proceed. Probably creating a better test/use case first and then discussing that.

[facontidavide]

If you want, we can arrange a meeting, and iterate faster to find alternatives. You can contact me at [email protected]