CI up to rmw_implementation, test_rmw_implementation, and rcl:

• Linux
• Linux-aarch64
• macOS
• Windows

Catching the exception seems reasonable to me,
but I think to use const string& symbol_name instead of const char * symbol_name for lookup_symbol is not good, which causes an extra calling constructor of std::string.
I know SharedLibrary of rcpputils has a method

  RCPPUTILS_PUBLIC
  void *
  get_symbol(const std::string & symbol_name);

that the parameter type is std::string, but I think it's better to ask rcpputils to provide an extra method for SharedLibrary, such as

  RCPPUTILS_PUBLIC
  void *
  get_symbol(const char * symbol_name);

, rather than updating the parameter type for lookup_symbol in rmw_implementation.

What do you think about that?

@iuhilnehc-ynos that's reasonable, but it doesn't preclude the need for this patch. Unless we make those methods noexcept, there are other reasons (besides a failing std::string construction) for them to throw. We have to handle them all.

Running CI again, this time computing test coverage:

• Linux
• Linux-aarch64
• macOS
• Windows

Also, for reference, a Linux build with test coverage but w/o this patch:

• Linux

Alright, from what I can observe in above's partial coverage computations, this patch does not appreciably degrade coverage. I'll go ahead and merge. Let's see what nightlies have to say.