A bit of an explanation of what happens here. Let's take @jpgr87 's compiler output from #343 :

...
/home/rich/code/ros2-beta2/src/ros2/rclcpp/rclcpp/include/rclcpp/function_traits.hpp: In instantiation of 
‘struct rclcpp::function_traits::function_traits<std::_Bind<
int (ObjectMember::*(ObjectMember*, std::_Placeholder<1>, 
std::_Placeholder<2>))(bool, float)> >’:
...

this error happens because the std::bind signature does not match any of the signatures in function_traits.

function_traits has the following template arguments for std::bind objects (https://github.com/ros2/rclcpp/blob/master/rclcpp/include/rclcpp/function_traits.hpp#L82):

• ClassT: since this is an std::bind for an object method, ClassT is the type of such object, in our case ObjectMember
• ReturnTypeT: whatever type the functor returns, in this case int
• Args: the types of the arguments of the functor, namely bool and float
• FArgs: any other arguments whose types are not know at compile time (e.g. placeholders), but which we're not really interested in. They are matched only to avoid compiler errors, these are usually implementation-specific, in our example, std::_Placeholder<1> and std::_Placeholder<2>.

so having all that, we can construct a signature for function_traits that can match all that, and thus:

struct function_traits<std::_Bind<
int (ObjectMember::*(ObjectMember*,
std::_Placeholder<1>, std::_Placeholder<2>))
(bool, float)>

becomes:

struct function_traits<std::_Bind<
ReturnTypeT (ClassT::*(ClassT*,
FArgs ...))
(Args ...)>>

I hope that helped. @clalancette @allenh1 I often called function_traits write-only code, it's unfortunately a bit hard to read, but given the requirements that we had of being able to deduce argument types at compile time transparently (callbacks with unique_ptr vs shared_ptr, service callbacks with/without a request header), I couldn't figure out a more readable way. It'd be great to replace it with something less esoteric while attaining the same results, though.

@esteve thank you for the fix, and thank you for the super detailed explanation! I appreciate it, and definitely learned a thing or two.

Cheers!

@allenh1 no problem :-) Glad to hear it helped, it's unfortunate that the function_traits uses a fair bit of black magic and is not more readable, though. Let me know if this PR solves the issue with GCC 8.0, I've only tested it with 7.1

@esteve this definitely compiles with GCC 8.0 as well. Does this still work with 5.4?

• Linux
• Linux-aarch64
• macOS
• Windows

@esteve Looks like the linter is unhappy.

--- include/rclcpp/function_traits.hpp
+++ include/rclcpp/function_traits.hpp.uncrustify
@@ -86 +86 @@
-struct function_traits<std::_Bind<ReturnTypeT (ClassT::*(ClassT*, FArgs ...))(Args ...)>>
+struct function_traits<std::_Bind<ReturnTypeT(ClassT::*(ClassT *, FArgs ...))(Args ...)>>

rclcpp.cpplint.whitespace/comments [2] (/home/rosbuild/ci_scripts/ws/src/ros2/rclcpp/rclcpp/include/rclcpp/function_traits.hpp:85)

Other than that, it appears GCC 5.4 is ok with these changes (that's quite relieving!).

given that _GLIBCXX_RELEASE appeared in GCC 7.1 I don't expect any behavior change on CI machines. I pushed a linter fix and dont think that running CI on all platforms is necessary. Here is a linux CI to confirm it passes the linters now:

• Linux

Shout out to @esteve for the quick fix and the great explanation!

CI is green, I;m going to squash and merge this if there is no objection

@allenh1 @mikaelarguedas thanks for fixing the linter errors, I forgot to run uncrustify before submitting the pull request 🙁

It happens. Thanks for the fix!