forward declare fromMsg to avoid missing symbols in downstream libraries

[jcmonteiro]

This PR fixes rst-tu-dortmund/teb_local_planner#334 and mitigates #242 while a proper solution addressing fromMsg and other template prototype functions (e.g., #427) is not provided.

Why

Without the forward declaration, the symbol will be undefined in a target that depends on tf2 and tf2_geometry_msgs and includes first tf2/utils.h and later tf2_geometry_msgs/tf2_geometry_msgs.hpp since there are inlined functions in tf2/impl/utils.h that make use of fromMsg with the quaternions as input types.

Disclaimer

This solution is avoiding the deeper issue where fromMsg and other template functions are not being specialized. If, for example, fromMsg had been properly specialized in tf2_geometry_msgs.hpp for the types geometry_msgs::msg::Quaternion and tf2::Quaternion, the forward declaration here would be

template<>
void fromMsg(const geometry_msgs::msg::Quaternion & in, tf2::Quaternion & out);

Example

One can validate this solution by building a simple library

// include/library/library.hpp
#pragma once
namespace library {
void function();
}

// src/library.cpp
#include "library/library.hpp"
#include <tf2_geometry_msgs/tf2_geometry_msgs.hpp>
#include <tf2/utils.h>

void library::function()
{
  geometry_msgs::msg::Quaternion quat;
  tf2::impl::toQuaternion(quat);
}

...
add_library(library SHARED src/library.cpp)
...

changing the order of the includes and checking the symbols in liblibrary.so with

nm --demangle <path-to-workspace>/build/library/liblibrary.so | grep fromMsg

When checking the symbol, one should notice that

• when the symbol is undefined it refers to the template function tf2::fromMsg<...>(...),
• otherwise, it refers to the non-template function tf2::fromMsg(...) which is not a specialization of the template function.

[aprotyas]

I think #427 will take a while to land, so this can be merged for now.

[aprotyas]

CI:
Repos file: https://gist.githubusercontent.com/aprotyas/c86851dae9847cc1bcf70351926c5a37/raw/94dd686808cad6b23a27280cb26629b261426080/ros2.repos
Build args: --packages-above-and-dependencies tf2
Test args: --packages-above tf2
Job: https://ci.ros2.org/job/ci_launcher/9535/

• Linux
• Linux-aarch64
• macOS
• Windows

Note: the failing uncrustify tests were addressed in #486.

[guilyx]

Hey @aprotyas, any update on this ?

[aprotyas]

It's been a while and the failing tests have been addressed, so I'll re-run CI:

Repos file: https://gist.githubusercontent.com/aprotyas/c86851dae9847cc1bcf70351926c5a37/raw/94dd686808cad6b23a27280cb26629b261426080/ros2.repos
Build args: --packages-above-and-dependencies tf2
Test args: --packages-above tf2
Job: https://ci.ros2.org/job/ci_launcher/9629/

• Linux
• Linux-aarch64
• Windows

[aprotyas]

@ahcorde @clalancette can either of you take a look please?

[aprotyas]

@jcmonteiro can you rebase your branch on the ros2 branch? That way de1753c will be tested as well, addressing the uncrustify failures.

[aprotyas]

Re-running CI following the rebase:

Repos file: https://gist.githubusercontent.com/aprotyas/c86851dae9847cc1bcf70351926c5a37/raw/94dd686808cad6b23a27280cb26629b261426080/ros2.repos
Build args: --packages-above-and-dependencies tf2
Test args: --packages-above tf2
Job: https://ci.ros2.org/job/ci_launcher/9662/

• Linux
• Linux-aarch64
• Windows

[guilyx]

@aprotyas hey, any ETA on this ?

[aprotyas]

I left this hanging last time, my bad. Okay, @jcmonteiro last rebase request and then I'll run CI.

[aprotyas]

Running CI:

Repos file: https://gist.githubusercontent.com/aprotyas/c86851dae9847cc1bcf70351926c5a37/raw/94dd686808cad6b23a27280cb26629b261426080/ros2.repos
Build args: --packages-above-and-dependencies tf2
Test args: --packages-above tf2
Job: https://ci.ros2.org/job/ci_launcher/10016/

• Linux
• Linux-aarch64
• Windows