Fix resolving links in sharded directories on gateway #5271
Conversation
License: MIT Signed-off-by: Łukasz Magiera <magik6k@gmail.com>
ghost
assigned 
magik6k
changed the title
License: MIT Signed-off-by: Łukasz Magiera <magik6k@gmail.com>
plugin/loader/load_linux.go
Outdated
| @@ -1,3 +1,5 @@ | |||
| // +build !noplugin | |||
License: MIT Signed-off-by: Łukasz Magiera <magik6k@gmail.com>
|
should be ready |
magik6k
changed the title
path/resolver/resolver.go
Outdated
| continue | ||
| } | ||
|
|
||
| val, rest, err := nd.Resolve(p) |
There was a problem hiding this comment.
This seems right (although rest should probably be _). This last resolve step is just to make sure that the rest of the path can be resolved. However, the function is only supposed to resolve up to the last node (it returns that last node and the rest of the path through this object).
Although, unless I'm mistaken, rest should be empty. It might make sense to check that.
path/resolver/resolver.go
Outdated
| return nil, nil, err | ||
| } | ||
| lnk, rest, err := r.ResolveOnce(ctx, r.DAG, nd, p) | ||
| if lnk != nil { |
There was a problem hiding this comment.
maybe we could clean this loop up a little bit by pulling everything below this if statement out of the for loop, then reversing the logic to 'break' if lnk == nil
License: MIT Signed-off-by: Łukasz Magiera <magik6k@gmail.com>
path/resolver/resolver.go
Outdated
| @@ -69,25 +69,42 @@ func (r *Resolver) ResolveToLastNode(ctx context.Context, fpath path.Path) (ipld | |||
| } | |||
|
|
|||
| for len(p) > 0 { | |||
| val, rest, err := nd.Resolve(p) | |||
| lnk, rest, err := r.ResolveOnce(ctx, r.DAG, nd, p) | |||
| if lnk == nil { | |||
There was a problem hiding this comment.
This'll drop the error, as far as I can tell.
There was a problem hiding this comment.
We actually have to drop the error here as ResolveOnce will try to either resolve to a full node or to a link (so for echo '{"foo":123}' | ipfs dag put, trying to resolve /foo it would return an error)
There was a problem hiding this comment.
Ah, interesting. Can we make sure to document that explicitly? At a glance, it looks wrong, so let's make sure nobody comes along and 'fixes' it
There was a problem hiding this comment.
We should probably remove the error check below in that case.
There was a problem hiding this comment.
It can technically still be set at that point, but that would be a rather weird edge case
License: MIT Signed-off-by: Łukasz Magiera <magik6k@gmail.com>
Fixes #5270