Test failures when building with clang

[dsobek]

Describe the bug
When building with clang on master, tests are failing with c++ exception:
C++ exception with description "Value outside numeric limits" thrown in the test body.

Any.Cast appears to be the most minimal testcase that is failing. It comes down to casts from any integral type to a double type.

How to Reproduce

Please provide a specific description of how to reproduce the issue or source code that can be compiled and executed. Please attach a file/project that is easy to compile, don't copy and paste code snippets!

From project root:

CC=clang CXX=clang++ colcon build
./build/behaviortree_cpp/tests/behaviortree_cpp_test

Root cause

I've found that the root cause is in checkTruncation

In particular if To has a larger max value than From (say for example To is double and From is int), static_cast<From>(std::numeric_limits<To>::max()) does not always evaluate to the max value of From. In clang, with any optimization (I see behaviortree.cpp is built with -O3), this evaluates to 0. I am not sure if this is meant to be undefined behavior but it sure is acting like it.

[dsobek]

Potential solution in #940

[Aglargil]

This might help too #932

[facontidavide]

I am aware of the issue.

[facontidavide]

the change in the code was introduced mased on #919 , as suggested by @cktii

but clearly that is not a solution that works on all platforms.

@Aglargil how is #932 related?

@dsobek your PR #940 doesn't pass the tests in wondows

[Aglargil]

@Aglargil how is #932 related?

@facontidavide
I found this error when I was adding the _onStart precondition function to the script, so I fixed it, you can take a look at this commit: d8ca1f1

[facontidavide]

fixed with #941 .Thanks @Aglargil you were right!

[dsobek]

Thanks for fixing @facontidavide!

@dsobek your PR #940 doesn't pass the tests in wondows

For the record, I think you were looking at Picknik's fork in PickNikRobotics#6 where we haven't gotten around to fixing the windows build because we don't support it. I don't think #940 would've shown the same failures.