Make Batch a helper type/function

batch.go

+package format
+
+import (
+	cid "github.com/ipfs/go-cid"
+)
+
+func Batching(ds DAGService) *Batch {
+	return &Batch{
+		ds:      ds,
+		MaxSize: 8 << 20,
+
+		// By default, only batch up to 128 nodes at a time.
+		// The current implementation of flatfs opens this many file
+		// descriptors at the same time for the optimized batch write.
+		MaxBlocks: 128,
+	}
+}
+
+type Batch struct {
+	ds DAGService
+
+	// TODO: try to re-use memory.
+	nodes     []Node
+	size      int
+	MaxSize   int
+	MaxBlocks int
+}
+
+func (t *Batch) Add(nd Node) (*cid.Cid, error) {
+	t.nodes = append(t.nodes, nd)
+	t.size += len(nd.RawData())
+	if t.size > t.MaxSize || len(t.nodes) > t.MaxBlocks {
+		return nd.Cid(), t.Commit()
+	}
+	return nd.Cid(), nil
+}
+
+func (t *Batch) Commit() error {
+	_, err := t.ds.AddMany(t.nodes)
+	t.nodes = nil
+	t.size = 0
+	return err
+}

merkledag.go

@@ -36,25 +36,11 @@ type DAGService interface {
 	// nodes of the passed in node.
 	GetMany(context.Context, []*cid.Cid) <-chan *NodeOption
 
-	Batch() Batch
+	AddMany([]Node) ([]*cid.Cid, error)
 
 	LinkService
 }
 
-// An interface for batch-adding nodes to a DAG.
-
-// TODO: Is this really the *right* level to do this at?
-// Why not just `DAGService.AddMany` + a concrete helper type?
-//
-// This will be a breaking change *regardless* of what we do as `Batch` *used*
-// to be a plain struct (passed around by pointer). I had to change this to
-// avoid requiring a `BlockService` (which would introduce the concept of
-// exchanges and I really don't want to go down that rabbit hole).
-type Batch interface {
-	Add(nd Node) (*cid.Cid, error)
-	Commit() error
-}
-
 // TODO: Replace this? I'm really not convinced this interface pulls its weight.
 //
 // Instead, we could add an `Offline()` function to `NodeGetter` that returns an