feat: Add ClockWaiter and ClockConditionalVariable #2691
Conversation
|
This pull request has been mentioned on ROS Discourse. There might be relevant details there: https://discourse.ros.org/t/next-client-library-wg-meeting-friday-6th-december-2024-8am-pt/40954/2 |
|
Assigned @wjwwood, and recommend Client Library Working group review 🧇 |
| * Note, this class does not handle shutdowns, if you want to | ||
| * haven them handles as well, use ClockConditionalVariable. | ||
| */ |
There was a problem hiding this comment.
Is there a use-case for a conditional variable that does not wake up when ros shuts down?
There was a problem hiding this comment.
Yes, if you need to handle the shutdown and wakeup explicit. Executors are an example of this.
| * Note, this class does not handle shutdowns, if you want to | ||
| * haven them handles as well, use ClockConditionalVariable. | ||
| */ | ||
| class ClockWaiter |
There was a problem hiding this comment.
The naming is a bit strange. This one is called ClockWaiter, ther other one is called ClockConditionalVariable.
According to your comments they are both conditional variables and their function is nearly identical. Why such different names?
| const rclcpp::Time& abs_time, const std::function<bool ()> &pred ) | ||
| { | ||
| // Install jump handler for any amount of time change, for two purposes: | ||
| // - if ROS time is active, check if time reached on each new clock sample |
There was a problem hiding this comment.
What do you mean by "if ROS time is active". How can the ROS time not be active?
There was a problem hiding this comment.
"ROS time is active" means the node runs on the external time input from /clock. Also know as sim_time. If it is not active, this means the system just runs on system time.
| rcl_jump_threshold_t threshold; | ||
| threshold.on_clock_change = true; | ||
| // 0 is disable, so -1 and 1 are smallest possible time changes | ||
| threshold.min_backward.nanoseconds = -1; | ||
| threshold.min_forward.nanoseconds = 1; |
There was a problem hiding this comment.
Use named struct initializer to ensure that all members of the struct are initialized.
That way the compiler will tell you if anyone changes rcl_jump_threshold_t
There was a problem hiding this comment.
sadly, rclcpp is using c++17 and this is a c++20 feature.
rclcpp/src/rclcpp/clock.cpp
Outdated
|
|
||
|
|
||
| bool | ||
| wait_until_ros_time(std::unique_lock<std::mutex>& lock, |
There was a problem hiding this comment.
Add documentation.
Especially document that system time will be used if ros time is not active
jmachowinski
force-pushed
the
clock_cond_rolling
branch
from
7cd8101 to
90acb47
Compare
jmachowinski
force-pushed
the
clock_cond_rolling
branch
2 times, most recently
from
9eaf8f5 to
f4070bc
Compare
Added two new synchronization primitives to perform waits using an rclcpp::Clock. Signed-off-by: Janosch Machowinski <[email protected]>
Signed-off-by: Janosch Machowinski <[email protected]>
jmachowinski
force-pushed
the
clock_cond_rolling
branch
from
f4070bc to
e16b8f9
Compare
This replaces the Clock::cancel_sleep_or_wait with two new synchronization primitives, that can be instantiated separate from the Clock object.
The old API turned out the be a bad design, as one can not simply copy or create new clock objects.
If one copys a clock object, the same internal pimpl is copied, which is undesirable in combination with this API,
as they then all share the same conditional. Therefore one can't do sleeps with multiple threads without waking up
other threads.
If one tries to create a new clock, then there is no clockprovider registered to it. Leading to an unexpected
behavior in simulation mode. We should perhaps add a BIG warning to the constructor to not do this.
This merge request is a first step in reworking the clock code, and I want to get some feedback on the API / class names, before refactoring further.
@arneboe Can you have a look the documentation and API ?