hash: allow Hash to implement encoding.BinaryMarshaler, encoding.BinaryUnmarshaler

[Capstan]

There are some use cases that aren't met by current hashing functions

• Serialize state of a partial byte stream processing for later work; recalculating the hash from the beginning is inefficient and, in our case, infeasible due to the size of the data and legacy compatibility requirements that inhibit use of, say, a Merkle tree.
• Sharing calculation of common prefix hashes; this only requires in-memory copying rather than serialization.
• Conditional append based on hash matches, which is a corollary of the previous.

Previous thread on golang-dev.

In my specific case, I'm looking for MD5 [de]serialization.

/cc @jlhawn

[bradfitz]

Seems like we could make the concrete types implement https://golang.org/pkg/encoding/#BinaryMarshaler and https://golang.org/pkg/encoding/#BinaryUnmarshaler at least.

We can't change the signature of e.g. https://golang.org/pkg/crypto/md5/#New but we could document in an Example that you can type-assert that the returned hash.Hash is also an encoding.BinaryUnmarshaler etc.

[rsc]

Any objections to hashes implementing BinaryMarshaler and BinaryUnmarshaler? Will leave this open for a week.

[rsc]

Proposal accepted. CLs welcome for the Go 1.10 cycle for

• Updating hash.Hash docs to document the convention.
• Updating hash/* and crypto/* hashes to implement it.

[gopherbot]

Change https://golang.org/cl/66710 mentions this issue: crypto, hash: implement BinaryMarshaler, BinaryUnmarshaler in hash implementations