New iteration behavior via FileIter and TreeWalker

Instead of returning a channel of files, Tree.Files() now returns a
FileIter with these qualities:

* It returns files in the original order of the repository (relying on a
* new Tree.OrderedNames property)
* It can return errors encountered when retrieving files and trees from
* underlying storage
* It can be Closed without having to drain the entire channel
* It defers the heavy lifting to a new TreeWalker type
* Its behavior is a little more consistent with other Iter types
* It's a little less prone to memory leaks

This update includes a new TreeWalker type that will iterate through all
of the entries of a tree and its descendant subtrees. It does the dirty
work that Tree.walkEntries() used to do, but with a public API.

A new TreeIter type is also included that just walks through subtrees.
This could be useful for performing a directory search while ignoring
files/blobs altogether.

1 files changed, 31 insertions, 0 deletions

diff --git a/file.go b/file.go
index b0b0141..9ef0e56 100644
--- a/file.go
+++ b/file.go
@@ -33,3 +33,34 @@ func (f *File) Lines() []string {
 	}
 	return splits
 }
+
+type FileIter struct {
+	w TreeWalker
+}
+
+func NewFileIter(r *Repository, t *Tree) *FileIter {
+	return &FileIter{w: *NewTreeWalker(r, t)}
+}
+
+func (iter *FileIter) Next() (*File, error) {
+	for {
+		name, entry, obj, err := iter.w.Next()
+		if err != nil {
+			return nil, err
+		}
+
+		if obj.Type() != core.BlobObject {
+			// Skip non-blob objects
+			continue
+		}
+
+		blob := &Blob{}
+		blob.Decode(obj)
+
+		return &File{Name: name, Reader: blob.Reader(), Hash: entry.Hash}, nil
+	}
+}
+
+func (iter *FileIter) Close() {
+	iter.w.Close()
+}