Add SO_LINGER option

[sateffen]

Hey,

because I didn't submit a pull-request (hopefully yet, up till now I'm missing time) I want to add this issue as reminder. Maybe someone else wants to implement this, but I think it has to be implemented at some point (and yes, you can implement it as well, I don't mind)

According to linux and windows documentation, this is just passing a struct looking like

struct linger {
    int l_onoff;    /* linger active */
    int l_linger;   /* how many seconds to linger for */
};

as optval for SO_LINGER. This should work cross plattform as far as I know.

I think the API should take a duration and convert that in an on-off version of the struct. What do you guys think?

[alexcrichton]

Yeah sounds like this should be:

fn linger(&self) -> io::Result<Option<Duration>>;
fn set_linger(&self, dur: Option<Duration>) -> io::Result<()>;

Other than that looks good to me!

[sateffen]

That API looks like the one I imagined, nice :)

Just one question: Why giving an option, and not the duration itself? A duration of 0 tells me to turn it of as well, just like giving None. Is there anything special or nice about this, that I don't see?

[alexcrichton]

We tend to try to distinguish with types rather than sentinel values, but I'm not sure if Some(Duration::new(0, 0)) is actually possible

[sateffen]

Just tried the Duration::new(0, 0) with rust 1.10, and it worked. Otherwise this would be a bug in the std lib in my opinion. Durations may reach the zero^^

[sateffen]

I've added a pull-request for the SO_LINGER option. I tested this on windows, and it works, even though it tells me lingering is off by default.

This option doesn't exist for udp sockets as far as I read, so it's only implemented for the tcp parts.

[alexcrichton]

Awesome, thanks!

[sateffen]

Hmm, I just realized, that the libc binding doesn't expose the linger struct. I filed a bug about it. The winapi crate exposes this struct, so on windows this works flawlessly.

Maybe I'll find some more time tomorrow to open up a pull-request to the libc bindings, or someone else has some time to fix this issue. Afterwards this PR would work.

Or do you prefer an own struct for this? This would solve the problem immediatly, without updating the dependencies, but introduces some inconvenience, and could lead to problems if it gets added later on.

Any opinions?

[alexcrichton]

Yeah let's try to add this to libc if we can, no reason it's not there other than someone just hasn't done it yet!

[sateffen]

The pullrequest is merged, and in my opinion this issue is resolved. Is there anything missing, that needs to be done?

[alexcrichton]

Aha yes, indeed!