Properly dispatch exceptions thrown from hpx_main to be rethrown from hpx::init/hpx::stop

[hkaiser]

No description provided.

[msimberg]

Very nice, thanks!

[msimberg]

Hmm, this is throwing more/in a different place than some of our tests expect. We do have some tests that catch an exception thrown in hpx_main (error_callback is the simplest). What sort of exceptions were not propagated up through init/main?

[msimberg]

report_error is mean to store exceptions thrown, which the runtime then rethrows after everything has been shut down. Could it be that report_error is not getting called, or encounters an error itself (did the application hang?)?

[hkaiser]

Hmm, this is throwing more/in a different place than some of our tests expect. We do have some tests that catch an exception thrown in hpx_main (error_callback is the simplest). What sort of exceptions were not propagated up through init/main?

I'll investigate.

[msimberg]

@hkaiser I've attempted to clean up the thread number handling. local_num is allowed to be -1, for example when creating a thread from a non-worker thread (like spawning hpx_main/run_helper as was the case here). @biddisco could you have a look to see if the changes to the shared_priority_queue_scheduler make sense?

[hkaiser]

@hkaiser I've attempted to clean up the thread number handling. local_num is allowed to be -1, for example when creating a thread from a non-worker thread (like spawning hpx_main/run_helper as was the case here).

Thanks @msimberg. I did suspect that much, but was not sure what the correct handling in the scheduler should be. Let's see how the tests fare.