From a000caf094d6a7157932af1ddeb5bd5f3355f139 Mon Sep 17 00:00:00 2001
From: Jeromy <why@ipfs.io>
Date: Thu, 29 Sep 2016 13:19:07 -0700
Subject: [PATCH] add comment detailing the algorithm and fix
License: MIT
Signed-off-by: Jeromy <why@ipfs.io>
---
set.go | 13 +++++++++++++
1 file changed, 13 insertions(+)
diff --git a/set.go b/set.go
index d93ccd1..e2ac757 100644
--- a/set.go
+++ b/set.go
@@ -139,6 +139,19 @@ func storeItems(ctx context.Context, dag merkledag.DAGService, estimatedLen uint
}
hashed := make(map[uint32][]item)
for {
+ // This loop essentially enumerates every single item in the set
+ // and maps them all into a set of buckets. Each bucket will be recursively
+ // turned into its own sub-set, and so on down the chain. Each sub-set
+ // gets added to the dagservice, and put into its place in a set nodes
+ // links array.
+ //
+ // Previously, the bucket was selected by taking an int32 from the hash of
+ // the input key + seed. This was erroneous as we would later be assigning
+ // the created sub-sets into an array of length 256 by the modulus of the
+ // int32 hash value with 256. This resulted in overwriting existing sub-sets
+ // and losing pins. The fix (a few lines down from this comment), is to
+ // map the hash value down to the 8 bit keyspace here while creating the
+ // buckets. This way, we avoid any overlapping later on.
k, data, ok := iter()
if !ok {
break
--
GitLab