gh-79698: selector.EpollSelector: add new parameter to support extra events

[Zheaoli]

bpo-35517: selector.EpollSelector: add new parameter to support EPOLLEXCLUSIVE

https://bugs.python.org/issue35517

https://bugs.python.org/issue35517

• Issue: selector.EpollSelector: add new parameter to support EPOLLEXCLUSIVE #79698

[asvetlov]

A documentation update is needed

[asvetlov]

extra_events=0 maybe?

[zhangyangyu]

I prefer None, but no False plz.

[eamanu]

None?

[asvetlov]

extra_events is a bitmask, why None?

[Zheaoli]

extra_events is a bitmask, why None?

In the most circumstance, people do not need the extra_events , so I give it a default value

@asvetlov

[asvetlov]

With 0 default value people still don't need to pass extra_events.
But we don't need to extra check the value for None, 0 should work fine as no flags for all possible selectors: poll, epoll, kqueue etc

[Zheaoli]

that sounds a good idea. I will fix it! Thx!

[asvetlov]

Should modify() method be updated as well?
Should new argument be mentioned in the base abstract class?

[zhangyangyu]

After this change, we could pass more eventmask in, even wrong ones (not before since only read/write allowed). What to do with them?

[Zheaoli]

@zhangyangyu

In select module, people use it by passing event mask directly.
So, I think it's necessary to check the extra_events.

[asvetlov]

Regarding to check for invalid event mask: the underlying call to epoll (or another poller API) should raise an exception for invalid mask bits.
Why check them again in selectors?

[giampaolo]

I commented on the bug tracker.

[zhangyangyu]

Why check them again in selectors?

@asvetlov It's the current status of register and modify, see their docs:

This returns a new SelectorKey instance, or raises a ValueError in case of invalid event mask or file descriptor, or KeyError if the file object is already registered.

And IMHO, a ValueError is easier to identify the root cause of the error than an OSError of poll. And as my test and understanding the manual, both poll and epoll_wait won't return error for invalid bitmask, it seems just pick the bitmask it thinks valid.

But it seems it's somewhat messy to implement the check. Relying on the users to always use constants in select module is also acceptable.

[erlend-aasland]

@Zheaoli, would you mind to resolve the conflicts and pull in main?

[github-actions]

This PR is stale because it has been open for 30 days with no activity.