@flixr since you have contributed this code part recently can you please comment on this patch.

Hm.. this actually seems to be a "bug" in boost, it doesn't handle the case that pthread_condattr_setclock is apparently not available under OSX...
This is a real shame, as the SteadyTimer then won't work (or rather use the system time instead of monotonic time).
But in the absence of a better solution, this seems ok as it will at least make it compile again...

@ffurrer what boost version are you using?

Falling back to system clock when a steady clock is requested on OSX doesn't sounds like a good behavior. If there is no "simple" fix it might require a custom implementation for OSX to achieve feature parity.

FWIW, I just did some research on this topic for ROS2: ros2/rcutils#43 (comment) . We could use one of the solutions I pointed out there.

I will merge this to fix the regression introduced by the switch to the steady time. Please create a follow up PR to support steady time on OS X (#1143).

Actually it would make sense to also #ifdef out the specialization in steady_timer.cpp, so it uses the same implementation as for WallTime (with the time jump detection) on OSX.

Actually it would make sense to also #ifdef out the specialization in steady_timer.cpp, so it uses the same implementation as for WallTime (with the time jump detection) on OSX.

I would like to see a PR to implement steady for OS X rather sooner than later. Therefore I don't think we need a patch for the hopefully very short time period in between.

@flixr I'm using 1.64:

/usr/local/Cellar/boost/1.64.0_1 (12,628 files, 395.7MB) *
Poured from bottle on 2017-04-30 at 08:38:38