To add a little more detail to this issue.

Currently the system_time_total is calculated only on the main thread: https://github.com/SanderMertens/flecs/blob/b5241cc76e50cba8cefe423c4478bf6fadb28462/src/addons/pipeline/pipeline.c#L632-L641

This means that the stats only knows about the amount of time the operation took to execute the main thread portion of the work, however, systems may have unbalanced work. Meaning a worker thread could have more to do than the main thread. If this happens it will cause the time interval on the main thread will not be equivalent to the amount of time the system actually took to execute.

To fix this issue, the following changes need to be made:

• ecs_system_t::time_spent should be changed to a vec<ecs_ftime_t> that has as one element for single threaded systems, and as many elements as threads in the application for multithreaded systems
• each thread should only increase its own element using the stage_index (thread index) to index the vector
• the stats API must aggregate values by taking the min/avg/max of the values in the time_spent array

The following scenarios have to be tested:

• verify time_spent is set correctly in single-threaded application
• verify time_spent is set correctly in multi-threaded application
• verify time_spent is set correctly after increasing the number of threads
• verify time_spent is set correctly after decreasing the number of threads
• verify that the stats API works with a single threaded scenario
• verify that the stats API works with a multi threaded scenario
• verify that single threaded systems still work the same as before
• visually confirm that values look correct in the explorer