make z_bytes_to_slice and z_bytes_to_string avoid making copy for non-fragmented non-shm data #822
Conversation
DenisBiryukov91
changed the title
|
The iRobot benchmark shows (roughly) a 40% decrease in the latency of the I used an Intel® NUC 12 Pro NUC12WSHi5 (12th Gen Intel(R) Core(TM) i5-1240P, 16GiB RAM) running Ubuntu 24.04.1 LTS and ROS2 rolling. |
| std::borrow::Cow::Borrowed(s) => { | ||
| // force a copy if zbytes is a single shared memory slice | ||
| #[cfg(all(feature = "unstable", feature = "shared-memory"))] | ||
| if payload.as_shm().is_some() { |
There was a problem hiding this comment.
There is a problem: if compiled as not "unstable" and with "shared-memory" zenoh will still be able to receive SHM buffers, but this check won't be performed. Thus we will have further code possibly executed on SHM buffer.....
There was a problem hiding this comment.
which is probably fine today, since there is no way to mutate shm buffers without unstable - so getting an increased ref count on shm-slice should not be an issue in this context, right ?
make z_bytes_to_slice and z_bytes_to_string avoid making copy for non-fragmented non-sim data