fix: traverser should skip over identity CIDs #4
Conversation
Codecov Report
Additional details and impacted files@@ Coverage Diff @@
## main #4 +/- ##
=======================================
Coverage ? 64.67%
=======================================
Files ? 69
Lines ? 7519
Branches ? 0
=======================================
Hits ? 4863
Misses ? 2512
Partials ? 144 |
LexLuthr
left a comment
LexLuthr
left a comment
There was a problem hiding this comment.
Apart from chain data and toor identity CIDs, do we have any other use case where we will find identity CIDs in the DAG?
|
I wish I could be confident in suggesting how common they are but I just don't know. I try to avoid them but some people see them as an efficiency hack so it wouldn't surprise me if they were not uncommon and made up some portion of our "could not load link" errors (there's no clue in error messages currently that would give us a hint, we'd have to rewind the DAG to see). 🤷 So urgency is unknown, I'm just treating this as a TODO I finally got around to in Lassie and found that all 3 protocols had some form of problem with them! In terms of risk, this PR is very low risk. Doesn't change the happy path and only fixes the identity CID path. |
Fix is in ipldutil/traverser.go; without this, an identity CID in a DAG that contains a link to more blocks won't be properly traversed. Unsure how common this is in the real world, most identity CIDs are for chunks that are small enough that they can't fit a CID in but there are instances where people feel it worthwhile to inline chunks much larger than this (e.g.. The Filecoin chain itself has such inline CIDs too.