fix meshblock pack issues coming from template shadowing and add a test

[Yurlungur]

PR Summary

Downstream, @brryan noticed that calls to MeshData::PackVariables with a PackIndexMap, for example

// MeshData *rc defined above
PackIndexMap imap;
rc->PackVariables(names, imap);

cause compile-time errors. The issue was introduced in the sparse interface update, #535 , where this machinery changed significantly. In short, in that update, explicit overloads were replaced with variatic templates. Unfortunately, C++ does not resolve these templates appropriately, and they shadow each other, causing bugs. We didn't catch this until now because templated functions are optimized out if they're not called.

This MR resolves #570 . I replace the variatics with explicit overloads again, with a few changes to reduce line number as best I can. I also add a call to this machinery in the calculate pi example, so this issue should be caught by regression tests in the future.

Note that the full regression test suite is currently failing. (See #571 ). So we should either merge this in without triggering the full suite on the CI, or we should wait for #571 to be resolved. I lean towards the former. Thoughts?

PR Checklist

• Code passes cpplint
• New features are documented.
• Adds a test for any bugs fixed. Adds tests for new features.
• Code is formatted
• Changes are summarized in CHANGELOG.md
• CI has been triggered on Darwin for performance regression tests.
• (@lanl.gov employees) Update copyright on changed files

[pgrete]

Pending the minor question, looks good! Thanks.
I'm also fine with merging this in while #571 is not fixed yet, but we should still run the full test suite here manually and see if nothing else new fails.

[pgrete]

Are these actually required?
If I'm not missing something they're matching the

  const auto &PackVariablesAndFluxes(const std::vector<Elem> &names_or_flags,
                                     PackIndexMap &map, vpack_types::StringPair &key)

above, aren't they?

[Yurlungur]

These exclude either the PackIndexMap or the key, or both. Whereas the one you quoted has both of them. They're just syntactic sugar overloads.

[Yurlungur]

Thanks @pgrete . Agreed, I'll run the full test suite

[brryan]

Just one question for my own edification. Looks good, thanks for adding a typical use case to the calculate_pi example.

[Yurlungur]

Pending the minor question, looks good! Thanks.
I'm also fine with merging this in while #571 is not fixed yet, but we should still run the full test suite here manually and see if nothing else new fails.

Looks like the only error is the Input metadata diff