156: footprint types that are not serializable

[cz4rs]

fixes #156

current pain points in vt:

• Open MPI types

• std::list iterators

note:

• general solution for any non-serializable type in a container would be best

[PhilMiller]

I'd really like to see that countBytes logic applied to any non-serializable type in a container. Did you see my suggestion for a potentially appropriate type trait? It would go on the enable_if for the footprinting-specific container serialize functions

[cz4rs]

I'd really like to see that countBytes logic applied to any non-serializable type in a container. Did you see my suggestion for a potentially appropriate type trait? It would go on the enable_if for the footprinting-specific container serialize functions

Yes, I have seen it, but I was unable to get it to work so far (various errors at the dispatch level). Let me push a cleaned up example for easier discussion.

[cz4rs]

ed726eb and b41585c are two main lines that I've tried, but they are both not working - resulting in the same error as with the serializer completely missing

[PhilMiller]

Ok, I've gotten the type trait stuff working for the serializeArray case. However, I realize that more sensible thing to do is maybe solve this at a slightly lower level, in the Traverse machinery or below, so that it works for containers other than std::vector more automatically.

[PhilMiller]

We need to decide what to do here.

[cz4rs]

hard to say how much additional value a separate warning would provide, maybe expanding the first (unconditional) print on the entrance would be sufficient (e.g. adding typeid)?

[PhilMiller]

My thinking is that we may want a printed warning here in much broader circumstances than when we have full debug printing turned on. Perhaps aligned with CMAKE_BUILD_TYPE=Debug or such.

[PhilMiller]

Regardless, both this debug print and the ones below should probably have a name from typeid incorporated.

[lifflander]

I agree that printing typeid would be good here.

[PhilMiller]

I just tested that this passes all the footprinting tests in vt.

[cz4rs]

@PhilMiller this looks good to me, I can see that the PR check is complaining about signatures missing (in your commits) - feel free to force push the amendments

@lifflander this probably requires you review as a third party

edit: after giving it a second thought, this probably needs an explicit mention in the documentation, because now user will always be able to footprint stuff, but might be unaware about a silent fallback to this approach

[lifflander]

Overall this looks good. A few minor points about formatting fixes

[lifflander]

I agree that printing typeid would be good here.