Use ipld.ErrNotFound

[hsanjuan]

Rationale: ipfs/kubo#7074

[hsanjuan]

Note blockstore depends on raw-multihash-enabled datastore already, therefore this cannot be merged until that change is unlocked, or needs to be merged into a pre v1.0.0 branch.

[hsanjuan]

Pending ipfs/go-ipld-format#68

[Jorropo]

LGTM

[orpheuslummis]

Note that this breaks code using blockstore.ErrNotFound.

[Jorropo]

@orpheuslummis yes, that why this code has been released with a new minor version (not patch). 🙂
1.2.0 aa05ad6

If you see ipfs/kubo#7074 (comment) you can find most of the related PRs that updates various go-ipfs modules.

[orpheuslummis]

Thanks for the reference to the PRs. :)
That said semver indicates that minor version bumps should be backwards compatible. I'm assuming that go-ipfs-blockstore follows semver as this is what go-ipfs is doing as indicated https://github.com/ipfs/go-ipfs/blob/master/docs/releases.md.

[Jorropo]

That said semver indicates that minor version bumps should be backwards compatible.

@orpheuslummis

We usually use 0.x versions which use patch for minors and minors for breaking instead.

Under 0.x bumping the minor communicate a breaking change.

This modules somehow, have a 1.x release which then you are right should be a major bump (I didn't realised, I thought it was 0.2.0 not 1.2.0, I guess @hsanjuan had the same thing when he made the release).

I'll raise this issue with the maintainer team, but my personal opinion is that we do a semver like where we do <major breaking change, complete API rethink>.<breaking change>.<fix / patch> and not semver.

[Jorropo]

@orpheuslummis a really good point has been raised by someone else.

Golang's way of importing V2 of packages require adding /v2 to the import path.

Having to bump all imports paths and not just go.mod files for such trivial change sounds like a useless suffering.

[orpheuslummis]

I agree. I suggest considering switching to using v0.x.x versioning to avoid future confusions and to be semver-compatible.

[Jorropo]

@orpheuslummis I don't like that idea, golang also plays very badly with that, every time people would go get it would attempt to download the 1.x release.

Btw this package got a 1.x release because at one point they needed two branches because of two projects that were doing things differently, so they maintained a 0.x branch and a 1.x branch.

[orpheuslummis]

Ah yes. Then I suggest indicating in the README the versioning detail of the repo because it diverges from semver and other go-ipfs projects.

[Jorropo]

@orpheuslummis thx for your report, this has been openned:

ipfs/kubo#8832

[willscott]

could we in the short term put in an alias of

type ErrNotFound = ipld.ErrNotFound

[aschmahmann]

The new version takes a parameter (the cid that wasn't found) so an alias is insufficient.

[willscott]

var ErrNotFound = ipld.ErrNotFound{Cid: cid.Undef}

[Jorropo]

@willscott

No because the main issue is that code that use blockstore.ErrNotFound usually intent to do:

err == blockstore.ErrNotFound

However that code is broken now because even if that the same error (what the checks want to do), your proposition would only match both identicaly errors and CIDs. (so an ipld.ErrNotFound with not cid.Undef wouldn't work).

[hsanjuan]

var ErrNotFound = ipld.ErrNotFound{Cid: cid.Undef}

iirc there is intent here, in that people should switch to ipld.IsNotFound() checkers etc.

[willscott]

Yeah, realized as posting that the blockstore approach doesn't work.

I think the unfortunate thing i'm hitting is that there are blockstore interactions with go-ipld-prime that had been built i think with an intention not to depend on ipld-format. I'm struggling to figure out what error to return now because i'm hesitant to introduce the ipld-format dependency in ipld-prime.

[Jorropo]

I'm struggling to figure out what error to return now because i'm hesitant to introduce the ipld-format dependency in ipld-prime.

I think there is a good argument to make about moving ErrNotFound in go-ipld-prime.

Then we can deprecate and add aliases in go-ipld-format.