go-path integrated #8028
There are no files selected for viewing
| Original file line number | Diff line number | Diff line change |
|---|---|---|
|
|
@@ -6,20 +6,21 @@ package readonly | |
| import ( | ||
| "context" | ||
| "fmt" | ||
| "github.com/ipfs/go-cid" | ||
|
There was a problem hiding this comment. let's group the dependencies here and have separate groupings for builtins, fuse, ipfs+ipld |
||
| "io" | ||
| "os" | ||
| "syscall" | ||
|
|
||
| fuse "bazil.org/fuse" | ||
| fs "bazil.org/fuse/fs" | ||
| core "github.com/ipfs/go-ipfs/core" | ||
| ipld "github.com/ipfs/go-ipld-format" | ||
| logging "github.com/ipfs/go-log" | ||
| mdag "github.com/ipfs/go-merkledag" | ||
| path "github.com/ipfs/go-path" | ||
| ft "github.com/ipfs/go-unixfs" | ||
| uio "github.com/ipfs/go-unixfs/io" | ||
| cidlink "github.com/ipld/go-ipld-prime/linking/cid" | ||
| ) | ||
|
|
||
| var log = logging.Logger("fuse/ipfs") | ||
|
|
@@ -65,20 +66,41 @@ func (s *Root) Lookup(ctx context.Context, name string) (fs.Node, error) { | |
| return nil, fuse.ENOENT | ||
| } | ||
|
|
||
| nd, ndLnk, err := s.Ipfs.Resolver.ResolvePath(ctx, p) | ||
| if err != nil { | ||
| // todo: make this error more versatile. | ||
| return nil, fuse.ENOENT | ||
| } | ||
|
|
||
| cidLnk, ok := ndLnk.(cidlink.Link) | ||
| if !ok { | ||
| log.Debugf("non-cidlink returned from ResolvePath: %v", ndLnk) | ||
| return nil, fuse.ENOENT | ||
|
There was a problem hiding this comment.
|
||
| } | ||
|
|
||
| // convert ipld-prime node to universal node | ||
| blk, err := s.Ipfs.Blockstore.Get(cidLnk.Cid) | ||
| if err != nil { | ||
| log.Debugf("fuse failed to retrieve block: %v: %s", cidLnk, err) | ||
| return nil, fuse.ENOENT | ||
| } | ||
|
|
||
| var fnd ipld.Node | ||
| switch cidLnk.Cid.Prefix().Codec { | ||
| case cid.DagProtobuf: | ||
| fnd, err = mdag.ProtoNodeConverter(blk, nd) | ||
| case cid.Raw: | ||
| fnd, err = mdag.RawNodeConverter(blk, nd) | ||
|
Comment on lines
+81
to
+93
There was a problem hiding this comment. This seems confusing/bad in that it'll result in multiple blockstore fetches for the same data. Perhaps if the underlying datastore has caching this doesn't end up being so bad since we likely just fetched the data, but if this is a recurring pattern in the associated set of PRs here then we might need to make some changes. I added some comments in the go-merkledag PR ipfs/go-merkledag#67 There was a problem hiding this comment. @aschmahmann I don't actually see those :( Can you point me to them? There was a problem hiding this comment. or maybe you need to submit review or something? There was a problem hiding this comment. In terms of the more general question -- it's a bit tricky. In IPLD Prime land, we don't always have a block to go with a node, cause nodes don't always match block boundaries. In a future pure IPLD Prime world, this isn't a problem -- we wouldn't load the block here, cause we wouldn't try to produce a legacy merkledag node, cause UnixFS would already work with prime nodes natively. However to convert back to legacy node, we have to do this. Hopefully, it's just a temporary step till we make more progress on go-ipld-prime conversion There was a problem hiding this comment. Some of my comments seem to have disappeared, but this is the one I was thinking of ipfs/go-merkledag#67 (comment) |
||
| default: | ||
| log.Error("fuse node was not a protobuf node") | ||
| return nil, fuse.ENOENT | ||
| } | ||
| if err != nil { | ||
| log.Error("could not convert protobuf node") | ||
| return nil, fuse.ENOENT | ||
|
Comment on lines
+96
to
+100
There was a problem hiding this comment. Why did we switch these from There was a problem hiding this comment. that's a great question -- I didn't actually write this bit of code here (Alex C did but he's not working on PL stuff any more). Seems odd. I can change it back. |
||
| } | ||
|
|
||
| return &Node{Ipfs: s.Ipfs, Nd: fnd}, nil | ||
| } | ||
|
|
||
| // ReadDirAll reads a particular directory. Disallowed for root. | ||
|
|
||
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
minor: It would be nice if there were some
constvalues for the segment indexes. If those do not already exist, that probably better in a separate future PR.