We use make test as the measure of correctness.
This laxity has let bugs creep into several systems.
This commit changes our target to always run expensive
tests, unless one specifically runs `make test_short`
(we would do well to remove most if not all timing--
that's usually what makes tests take a long time.)
This is causing test failures because tests don't usually
have "/-/-" format. we can decide whether or not to allow
keys without validators, but for now removing.
cc @whyrusleeping
TestGetFailures may just be operating very slowly, instead
of completely failing. Right now it gets caught on travis
often. not sure if its actually wrong.
we shouldn't use an arbitrary timeout here. since Get
doesnt take in a context yet, we give a large upper bound.
think of an http request. we want it to go on as long as
the client requests it.
Make sharness tests' output helpful when verbose.
This means cating certain files, or running diagnostic
commands. I used a construction like:
test_expect_success ".go-ipfs/ has been created" '
test -d ".go-ipfs" &&
test -f ".go-ipfs/config" &&
test -d ".go-ipfs/datastore" ||
fsh ls -al .go-ipfs
'
The `|| ...` is a diagnostic run when the preceding command
fails. `fsh` is a trivial script that echoes the args, runs
the cmd, and then also fails, making sure the test case fails.
(wouldnt want the diagnostic accidentally returning true and
making it _seem_ like the test case succeeded).
This commit makes all network tests use ZeroLocalTCPAddress
as the initial peer address, and then relies on
net.ListenAddresses()
This should get rid of the tcp addr clash problems.
The separation of work in the p2p pkg is as follows:
- net implements the Swarm and connectivity
- protocol has muxer and header protocols
- host implements protocol muxing + services
- identify took over handshake completely! yay.
- p2p package works as a whole
