move vectors instantiated in wait_for_work to MemoryStrategy #116
Conversation
|
Seems reasonable to me, but I don't have time to try it right now. I'd recommend CI before merging. |
|
Is this necessary for the demo at ROSCon? If not I would suggest postponing it until Monday. |
jacquelinekay
added
in review
|
CI: Windows is already done and looks good besides the connext_dynamic tests, so that's a good sign. |
|
The |
| @@ -361,14 +361,14 @@ class Executor | |||
| for (auto & weak_subscription : group->subscription_ptrs_) { | |||
| auto subscription = weak_subscription.lock(); | |||
| if (subscription) { | |||
| subs.push_back(subscription); | |||
| memory_strategy_->subs.push_back(subscription); | |||
There was a problem hiding this comment.
I think the MemoryStrategy API would look clear if, instead of exposing the std::vector directly, provide add_subscription, add_service, etc. and hide the underlying storage.
jacquelinekay
force-pushed
the
fix_uninitialized_vectors
branch
from
3f4a2b9
to
af9d641
Compare
move vectors instantiated in wait_for_work to MemoryStrategy
* fix cmake for test - remove unused params * set env variable for rmw_implementation
…ackage (ros2#48) * ros2GH-111 Add package for converter plugins * ros2GH-111 Add CDR converter plugin * ros2GH-111 Add test for more primitives types * ros2GH-116 Fix cdr converter after rebase on new converters interface * ros2GH-116 Use rmw_serialize/rmw_deserialize directly in converter and link against rmw_fastrtps_cpp * Fix converter package.xml * Fix clang warnings * ros2GH-30 Change interface to the same convention as rmw_(de)serialize * comply to new rcutils error handling API * use poco to load fastrtps * Update rosbag2_converter_default_plugins/src/rosbag2_converter_default_plugins/cdr/cdr_converter.cpp Co-Authored-By: Karsten1987 <[email protected]>
* ignore cast function type warning Signed-off-by: Karsten Knese <[email protected]> * use pragma to ignore warning on specific lines of code Signed-off-by: Karsten Knese <[email protected]> * correctly branch the pragma Signed-off-by: Karsten Knese <[email protected]>
This was done in my attempt to reduce pagefaults in Executor::spin. Several vectors that were being instantiated on the stack in Executor::spin have been moved into the MemoryStrategy, so that they can be preallocated (with
vector::reserve) on construction.