Keep track of fallthrough value to avoid false decrement

[raJeev-M]

Cause: #52

[ghost]

uSockets/src/loop.c

Line 23 in 9136f22

loop->data.sweep_timer = us_create_timer(loop, 1, 0);

uSockets/src/loop.c

Line 35 in 9136f22

loop->data.wakeup_async = us_internal_create_async(loop, 1, 0);

Fallthrough polls are used pretty much only for internal timers and internal "asyncs" when creating a loop. It makes sense to only keep track/support fallthrough polls that are asyncs or timers, since those are rarely constructed.

[raJeev-M]

@alexhultman, Thank you for the response. I've come across those two polls before making this change, but for some weird reason I thought like-- what if there might be more fallthrough polls added in the future? (looks like we won't be).

Off the top of my head, one solution which I am currently thinking that requires no change in the interface as well as any Type is: directly calling free() in both timer/async_close function, rather than doing it via us_poll_free.

[ghost]

You can probably just make all polls increment the poll counter but if the poll is a fallthrough then also increment the threshold at where the loop falls through?

[ghost]

So if you have 5 polls where 2 are fallthrough then loop has counters 5 and 2 and every free decrements the total but not the threshold

[raJeev-M]

@alexhultman, I am thinking like, do we really need any change (adding threshold/ if..else) for the 2 fallthrough polls we are currently having? how about the below approach?

In us_create_poll() we do have an if statement for incrementing the 'num_polls', so, if we have two fallthrough polls, but zero non-fallthrough polls then the num_polls going to be zero, and with this being the case

we can add an one-line change under us_poll_free() like loop->num_polls? loop->num_polls-- : 0

would that be an acceptable change?