gpm_epoll_wait() revisions

[frozencemetery]

Always service timerfd in gpm_epoll_wait()

This may prevent a tight loop around epoll_wait() when both fire.

[cipherboy]

While we're here, let's initialize ret to something (EINVAL?). I think it'll help make sure we prevent errors based on uninitialized ret values in the future since this is a huge method.

[simo5]

@frozencemetery can you add a description that explains what this PR is trying to do ?

[frozencemetery]

@simo5, applied description from the second commit. First commit is a refactor.

[cipherboy]

(Per IRC), this addresses the feedback I had last time. I won't have time to re-review this as much as I did earlier.

[simo5]

LGTM

[simo5]

@frozencemetery are you planning to add anything else?
If not feel free to merge.

[frozencemetery]

I think we were waiting to hear back if this fixed the issue before merging. (I haven't heard back yet.)

[simo5]

@frozencemetery have you heard back yet ?

[frozencemetery]

Yes, thanks for the reminder. This did not fix the issue in question - I don't know whether it's better to merge this anyway or to drop it.

[simo5]

Do you think this improves the code in other ways?
If it does and it is not likely to cause any side issues I would merge, otherwise I would drop.

[frozencemetery]

I think so - if nothing else, there are more comments explaining what's going on.