Hi, Polykarpos, Thank you for your comment. Currently, schedulers do not manage `ABT_self_xxx` (specifically `lp_ABTI_local->p_thread`); they are only for non-scheduler ULTs. It includes not only the type you mentioned, but also the descriptor (i.e., `ABT_thread_self`) for example. This behavior change is introduced by the following commit: https://github.com/pmodels/argobots/commit/6b07ca2918e6bacbb7febb1706452701d...) I note that they are not managed correctly even in the previous version. This change was for performance; it can reduce the overheads of the `lp_ABTI_local` management. Are you using this functionality? If you think this functionality is useful, we will implement it. Even if not, we still need to clarify this specification. We will appreciate it if you make an GitHub issue about it. Thank you, Shintaro Iwasaki On Thu, Dec 6, 2018 at 4:16 PM POLYKARPOS THOMADAKIS via discuss <[email protected]<mailto:[email protected]>> wrote: Hello, I have observed the following behavior when I have a custom scheduler that runs a Tasklet: I retrieve the type of the scheduler before I run the Tasklet unit popped from the argobots pool and it is ULT which is what expected. The scheduler runs the Tasklet, and after it's done I retrieve its type again and it has changed to Tasklet which is unexpected. This seems a bit weird for me and I would like to know whether this shall be the functionality. I also checked against an older version of argobots and I found out that this is not happening there. Specifically, in version 1.0a1 the function ABTI_xstream_schedule_task<https://press3.mcs.anl.gov//wp-content/blogs.dir/231/files/argobots/doxygen/latest/d8/d4d/stream_8c.html#a0e222490fec0b8b0c5513a17ed5774b6> saves the state of current thread/task and recovers it after it's done executing the Tasklet. This is removed in current version. Thank you for your time, Polykarpos _______________________________________________ discuss mailing list [email protected]<mailto:[email protected]> https://lists.argobots.org/mailman/listinfo/discuss