Refactor and add comments to IOCP #run_once
#14380
Conversation
| unless next_event | ||
| Crystal::System.print_error "Warning: No runnables in scheduler. Exiting program.\n" | ||
| ::exit | ||
| end |
There was a problem hiding this comment.
For the record, this cannot actually happen because min_by instead of min_by? is used
There was a problem hiding this comment.
thought (non blocking): is that the job of the eventloop to decide to exit? As stated in the error message, this is a matter for the scheduler.
libevent merely returns an error (no events), and the case is expected in my implementation for RFC #2: the schedulers use it to decide to start spinning then park themselves for example.
There was a problem hiding this comment.
Yeah, this is definitely not a final solution. But it's just a stop gap for some odd edge case. So far it has never happend because the branch couldn't be reached and you'd get a different error if the queue was empty.
I think this was introduced before we even had an actual event loop on Windows and things were much more brittle.
| unless next_event | ||
| Crystal::System.print_error "Warning: No runnables in scheduler. Exiting program.\n" | ||
| ::exit | ||
| end |
While reading through this method 2 years after writing it, I found it hard to understand what its doing.
So this patch adds some comments to annotate the code.
And I also realized the control flow could be restructed to be more linear and easier to follow.
Thes code changes introduce no semantic difference, the just bring it to a slightly better readable form. The resulting LLVM IR is almost identical, just with a few branch lables swapped (as expected).