net: wrap connect in nextTick

[evanlucas]

Fixes an edge case regression introduced in
1bef717.

With the lookup being skipped, an error could be emitted before and
error listener has been added.

An example of this was presented by changing the server’s IP address
and then immediately making a request to the old address.

Related: #1823

[evanlucas]

The test that is removed in this PR was added specifically to test the functionality introduced in the commit that caused the issue.

EDIT: No longer applicable.

[trevnorris]

This is more costly than simply putting the connect in a nextTick.

[evanlucas]

@trevnorris yea, good point. Updated to just wrap connect in process.nextTick

[evanlucas]

CI: https://jenkins-iojs.nodesource.com/job/iojs+any-pr+multi/81/

[cjihrig]

LGTM

[trevnorris]

LGTM

For reference it could have also been done as

process.nextTick(connect, self, host, port, addressType, localAddress, localPort);

But I'm fairly certain it would be slower this way. So don't change how you currently have it.

[trevnorris]

Oh, is there a test we could add for the edge case regression?

[evanlucas]

I was trying to think of a way to test that... The only case I've seen has been changing the server's IP address and then continuing to attempt to make a request to the old one.

I was able to reproduce fairly consistently using https://gist.github.com/m0ppers/3709f7d87b73be9aca7b and changing my IP

[evanlucas]

I haven't been able to write a test that exhibits the exact same behavior each time for my local machine. It happens when a EHOSTDOWN or EHOSTUNREACH error is emitted and it takes a few requests to actually make the situation happen...

[trevnorris]

@evanlucas I've tried myself. Can't figure out a way to do it from a simple test script. So we won't worry about it.

[evanlucas]

Thanks, landed in af249fa