Use ipld-prime based go-merkledag

[hannahhoward]

This is a build of IPFS that uses a copy of go-merkledag that is mostly built on go-ipld-prime.

Note that this does not:

• attempt to modify any of Core API (there are methods that expect oldschool IPLD-format nodes)
• address the way nodes are parsed in for IPFS DAG Put (which is heavily tied to go-ipld-format)
• attempt to do anything other than incorporate the branch of go-merkledag and get the tests to pass

License: MIT
Signed-off-by: hannahhoward hannah@hannahhoward.net

Notes

• I have disabled most sharness test of the git IPLD plugin, as we need to build an ipld-prime base decoder for this to work
• There is one remaining sharness test that is failing. I believe this is because our current Decode function in go-ipld-legacy uses Link.Load, because we don't have access to the underlying multicodec table, meaning this is the only way to load links. However, I believe that Link.Load's current implementation requires a rehash of the bytes against the CID. The specific sharness test in question actually modifies the on disk bytes while leaving datastore.HashOnRead off. This means the bytes in this case do not match the CID, I think, causing link.load to fail. I think. In any case, it's an open question of whether loading should always rehash the bytes, given the performance penalty particularly if they are from a trusted source like the local datastore. This question is best left open for when LinkSystem is introduced (merging in go-ipld-prime soon, I think).

[willscott]

should we keep these failing until we fix the underlying issues?

[hannahhoward]

closed in favor of #7976