add sharness test for pubsub

[keks]

finally!

[whyrusleeping]

Looks like it fails on sharness (at least on teamcity): http://ci.ipfs.team:8111/viewLog.html?buildId=8663&buildTypeId=GoIpfs_CiTests&tab=buildLog

cc @chriscool for review, this is a bit of strange shell hackery

[chriscool]

It's a bit confusing to call a fifo "done" as it is a keyword in shell.

[chriscool]

It's a bit strange to have this "exit" here.
It means that the while loop will in fact never loop, so perhaps it could be an if read line; then ... fi instead.

[keks]

Yeah that is a bit hackish I agree. For some reason this loop even needs two runs, I didn't get why and I was glad it worked.

[chriscool]

It should be explained why we need to fork a process like this, especially as there is a "sleep 1" below.

[keks]

Yeah I'll add a comment

[chriscool]

Isn't is possible to do something less specific by doing something like:

num_nodes="$1"
shift
other_args="$@"

...

if test -n "$other_args"; then
    iptb start --args $other_args
else
    iptb start
fi

[keks]

I tried iptb start $@ but now I see why that didn't work - its executed in a completely different context. I use your solution now.

[chriscool]

Even if the two above commands cannot fail it is a good practice to anyway chain them with the commands below by using "&&" after them.

[keks]

I'll put && after the sleep. Where else do you mean?

[chriscool]

test_cmp expected actual is more standard.

[chriscool]

Yeah, it LGTM now, thanks!

[keks]

@chriscool The tests pass for me locally but fail on Travis and Teamcity. Do you have an idea what the problem could be? Do they pass on your machine?

[whyrusleeping]

@keks the problem is with your previous commit. Under the ndpayload encoding, youre outputting an empty line when the 'null' hello message arrives, where instead you should be outputting nothing (since the goal there is to ignore that first message)

[whyrusleeping]

This line here: https://github.com/ipfs/go-ipfs/blob/master/core/commands/pubsub.go#L133

Should just return an empty string reader instead of a newline

[keks]

Of course! Thank you. "The code is right, it must be the test's fault"...

I will move the checks to getPsMarshaler later today because now we have the same code in all three Marshalers.

[whyrusleeping]

@keks Thanks a bunch! This is great to have :)