Remove usage of merkledag.Link.Node pointer outside of merkledag

[mildred]

This prepares for inclusion of IPLD where the Node pointer won't be there. In any case, we should probably move away from using the struct attributes and use functions to access them instead. This will make it easier to plug in IPLD in place of merkledag.

[whyrusleeping]

timeout should be specified on the context using context.WithTimeout on the caller.

[mildred]

Note: We have two ways to go now:

• Make go-ipld gold a pointer to the Node object. It won't be as straightforward, and that's why we must limit usage of that pointer to the merkledag package (and have other packages uses functions instead)

• Remove the notion that a merkledag.Node can hold a pointer to its linked nodes, and refactor the API to reflect that.

  If we decide to refactor the API, the following functions will need to be modified:

  • DAGService.AddRecursive(Node)
  • DAGService.Remove(Node)
  • Node.AddNodeLink(string, Node)
  • Node.AddRawLink(string, Link)
  • Node.UpdateNodeLink(string, Link): Only the codehttp PUT handler uses it, and this is being refactored in PR fixing putHandler #1886
  • Link.GetNode(...) : has the ability to query the DAGService every time if the Node pointer is not there (may be slower, how much?)
  • Node.GetLinkedNode(..., string) : same as above

  A simple approach could be to introduce a NodeTree type that will hold a root node and a collection of Node dependencies. When constructing a DAG, we will use the NodeTree type instead of Node. AddRecursive(Node) will have to be renames AddTree(NodeTree).

[jbenet]

let's do this on top of dev0.4.0 branch

[jbenet]

see also discussion in ipld/go-ipld-deprecated#14