If the datastore has a value for the key, we already have
the block. We should not write it again. This will make redundant
writes much faster.

At the moment, a datastore.Has on leveldb is a GetBackedHas.
Track https://github.com/jbenet/go-datastore/issues/6

perf(cmd/ipfs) write mprof file in debug mode

TODO add memory profiling to ipfs2

discovered by @mappum and applied at https://github.com/jbenet/go-ipfs/commit/414e20b71e50b65e40883f3ac85f91aa81c1e486

fixing here to address #233

https://github.com/jbenet/go-ipfs/issues/233

refactor(blockservice, merkledag, namesys) deprecate u.ErrNotFound

refactor(routing) replace u.ErrNotFound with routing.ErrNotFound

fix(bitswap/message) duplicate entries

to emphasize idempotence

comes at the cost of O(n) Blocks() method.

so there's one less name to think about

A nice invariant for bitswap sessions:

        Senders and receivers can trust that messages do not contain
        duplicate blocks or duplicate keys. Backing the message with a
        map enforces this invariant.

        This comes at the cost of O(n) getters.

implementation will be patched to ensure bitswap messages cannot contain
duplicate blocks or keys

@whyrusleeping noticed this a couple days ago

potential long-term fix: prevent duplicate entries in the wantlist by
using a map/set and iterating over this data structure on export

Revert "perf(mux) use RWMutex"

perf(mux) use RWMutex

ipfs tour