Support SO_REUSEADDR

[ry]

Because of the lack of this, many networking tests resort to using different ports.

[ry]

Feb 24, 2020: Still an issue that we want to solve.

[jsouto18]

@ry as soon as I'm done with UDS I'll take care of this one!

[bartlomieju]

To implement this feature, net2 crate should be added to dependencies and reuse_address method should be called before creating a socket.

[attila-lin]

Is it still be an issue? Can I take it?

[bartlomieju]

Marking as good first issue. For example how to implement this see #13103

[Trolloldem]

@bartlomieju I tried to add the requested feature. I decided to not use net2 as suggested, since the library is now deprecated.
In order to achieve the same functionality I used the socket2 crate, which is already a dependency listed in ext/net/Cargo.toml.

For now I have a working unit test for op_net_listen, now I have to adapt the changes for the TLS version.
In order to achieve retro compatibility, the proposed method is composed as follows:

const listener1 = Deno.listen({ hostname: "127.0.0.1", port: 3500, reuseAddress: true});
const listener2 = Deno.listen({ hostname: "192.168.1.7", port: 3500, reuseAddress: true});

where the reuseAddress field is optional and defaults to false.

I should be able to open a PR in the following days. But first I have a question:

• in this is issue is mentioned that support for SO_REUSEADDR is needed
• this options allows the use case in which the same host has two different ip (127.0.0.1 and 192.168.1.7 for example)
• the exposed use case shows that without this feature it is necessary to use different ports
• I think that it in order to fully achieve the wanted result it is necessary to support the SO_REUSEPORT option
• in this way it is possible to support the following code:

const listener1 = Deno.listen({ hostname: "127.0.0.1", port: 3500, reusePort: true});
const listener2 = Deno.listen({ hostname: "127.0.0.1", port: 3500, reusePort: true});

where the reusePort field is optional and defaults to false.

Let me know if you think that these changes could be interesting and if it is the case to open a PR to add them.

[bartlomieju]

@Trolloldem thanks for looking into this. What you described sounds good to me and the proposed API looks good. socket2 is indeed preferable in this case.

It might be hard to test these properly, but please open a PR first and let's work from there.

[lucacasonato]

We specify SO_REUSEADDR on all sockets by default, just like Node.js and Go.