Drawing PointCloud2 in the World Coordinate System by Rviz

edit

I'm drawing PointCloud2 with rviz on melodic. They are superimposed based on a long decay and a tf based on the exact sensor position (with about 200hz).

However, there was something wrong on this drawing after an update the other day. All points during the decay period are drawn based on the latest tf now, so sensor movement is not considered.

This only happens in PointCloud2 and LaserScan is correct as before the update. (It remains in the correct world coordinate system because it is drawn based on the tf of the corresponding timestamp.) This is happening in every PointCloud2 and is normal in every LaserScan.

Did any recent updates cause any bugs?