Specifying hash algorithms in files.add doesn't work

[JonKrone]

Subsystem:

files.add and/or IPLD

Type:

Bug

Severity:

Medium

Description:

jsipfs files add --hash=<alg> ... generates and stores an incorrect DAG.
The option to specify a different hash alg was added with #1308.

Go/JS ipfs hash differences:

The hashes are both keccak-512-encoded yet we have different hashes - our object content must be different:

object get <key> returns a serialized blob that has a hash different from the one requested. It seems that on unixfs import, IPLD is generating a blob where the hash is coerced from the keccak-512 format to something else. I understood dag-pb to just be a style of serialization and not a hashing function itself so I'm confused what the codec of a CID holding a keccak-512 multihash should be.

@vmx do you have some insight as to what's happening above?

I think the coercion is happening in IPLD/unixfs-engine as unixfs-engine uses IPLD to generate the DAG tree that unixfs later writes to the fs.

Could it be caused by something like new CID's assumption that non-CID Buffers are dag-pb (here). It may also be that IPLD can't yet handle these custom hash functions.

[JonKrone]

I don't know if this is strictly related but it strikes me as unexpected behavior of --hash as well:

sha2-256 is the default hashing algorithm yet specifying it produces different output than the standard output. For both Go and JS

[alanshaw]

I think @achingbrain is all over this with #1371 - @JonKrone can you verify if this PR fixes your issues?

[alanshaw]

I just realised that PR doesn't actually change any dependencies, we might need to wait for the dependent PRs to be merged.

[JonKrone]

@alanshaw cool, thanks for bringing that to my attention! That might resolve the bit about go/js having different hashes.

I think we will still be generating an incorrect DAG. object.get shows that a blob created via --hash has an incorrect encoding/format of hash and links.hash, which is different than just hashing different content.

[vmx]

IPLD doesn't support custom hash functions: https://github.com/ipld/js-ipld/blob/a6329ce30e4b21bb2d0852326cfd55753c33b353/src/index.js#L239

[JonKrone]

Thanks for the confirmation @vmx. Does this mean that we aren't currently able to support #1308? Something like: jsipfs files add --hash=sha3-512

[vmx]

@JonKrone I'd be surprised if it would work.

[JonKrone]

@alanshaw As this feature is broken and has a more involved fix with no timeline, I think it should be reverted or we should otherwise indicate that it's not ready for use.

[alanshaw]

@JonKrone I believe this is now resolved - #1371 was merged and IPLD now supports specifying the hash algorithm ipld/js-ipld#133

Are you able to double check?