#:drain? #t does not wait until all fibers have completed.

[wingo]

#:drain? #t does not wait until all fibers have completed.

[emixa-d]

My suspicion:

When creating a scheduler with parallelism (make-scheduler #:parallelism N), it created a few other schedulers internally, which do not share a workqueue.

For testing whether the fibers have finished (see scheduler-work-pending & run-fibers), scheduler-work-pending? is used. But that procedure only looks at the workqueues of only one scheduler, and all other schedulers (co-schedulers?) are ignored!

Seems like we need to call scheduler-work-pending? on all schedulers in run-fibers.

[amirouche]

Oh but it does not rely on POSIX threads? Then I do not know.

[emixa-d]

This is for version 1.0.0 btw. Didn't test on master.

[emixa-d]

I can reproduce on master (44d17e6) (GNU Guile 2.2.7) as well.