Trigger hook for inner nodes (#4)

* trigger hook for inner nodes

* ensure correct line and column numbers in hooks

* add support for Go 1.20 Unwrap() []error

* update documentation

decode.go

@@ -38,7 +38,7 @@ type parser struct {
 	textless bool
 
 	hook     DecoderHook
-	hookPath []string
+	hookPath []*Node
 }
 
 func newParser(b []byte) *parser {
@@ -72,7 +72,7 @@ func (p *parser) init() {
 	}
 	p.anchors = make(map[string]*Node)
 	p.expect(yaml_STREAM_START_EVENT)
-	p.hookPath = make([]string, 0)
+	p.hookPath = make([]*Node, 0)
 	p.doneInit = true
 }
 
@@ -157,13 +157,10 @@ func (p *parser) parse(triggerHook bool) *Node {
 	case yaml_ALIAS_EVENT:
 		n = p.alias()
 	case yaml_MAPPING_START_EVENT:
-		triggerHook = false // maps are not leaf nodes, skip hook
 		n = p.mapping()
 	case yaml_SEQUENCE_START_EVENT:
-		triggerHook = false // sequences are not leaf nodes, skip hook
 		n = p.sequence()
 	case yaml_DOCUMENT_START_EVENT:
-		triggerHook = false // documents are not leaf nodes, skip hook
 		n =  p.document()
 	case yaml_STREAM_END_EVENT:
 		// Happens when attempting to decode an empty buffer.
@@ -175,11 +172,49 @@ func (p *parser) parse(triggerHook bool) *Node {
 	}
 
 	if triggerHook && p.hook != nil {
-		p.hook(p.hookPath, n)
+		p.triggerHook(n)
 	}
 	return n
 }
 
+func (p *parser) triggerHook(node *Node) {
+	path := make([]string, 0, len(p.hookPath))
+	for _, n := range p.hookPath {
+		if n.Kind == SequenceNode {
+			// sequence nodes have no value, we use the index of the element instead
+			path = append(path, strconv.Itoa(len(n.Content)))
+		} else {
+			path = append(path, n.Value)
+		}
+	}
+
+	// we want to give the hook its own pointer
+	hookNode := *node
+	if len(p.hookPath) > 0 {
+		if parent := p.hookPath[len(p.hookPath)-1]; parent.Kind != SequenceNode {
+			// Go-yaml represents keys in a map as scalar nodes, which are not
+			// related to the map values in any way. In the decoder hook we
+			// simplify this by exposing the path to the key node and supplying
+			// the corresponding value node that contains the data for that key.
+			// The line and column in the value node point to where the value
+			// starts, which can cause some confusion, as it does not match the
+			// line and column of the key referenced in the path. That's why we
+			// adjust the line and column to match the key node's line and
+			// column. After the hook returns we copy the node back into the
+			// original node so we need to adjust the line and column again to
+			// preserve the original values.
+			oldLine, oldColumn := node.Line, node.Column                // store original line and column
+			hookNode.Line, hookNode.Column = parent.Line, parent.Column // adjust line and column
+			defer func() {
+				node.Line, node.Column = oldLine, oldColumn // restore original line and column
+			}()
+		}
+	}
+
+	p.hook(path, &hookNode)
+	*node = hookNode // overwrite node in case the hook changed it
+}
+
 func (p *parser) node(kind Kind, defaultTag, tag, value string) *Node {
 	var style Style
 	if tag != "" && tag != "!" {
@@ -271,14 +306,14 @@ func (p *parser) sequence() *Node {
 	}
 	p.anchor(n, p.event.anchor)
 	p.expect(yaml_SEQUENCE_START_EVENT)
+	p.hookPath = append(p.hookPath, n)
 	for p.peek() != yaml_SEQUENCE_END_EVENT {
-		p.hookPath = append(p.hookPath, strconv.Itoa(len(n.Content)))
 		p.parseChild(n, true)
-		p.hookPath = p.hookPath[:len(p.hookPath)-1]
 	}
 	n.LineComment = string(p.event.line_comment)
 	n.FootComment = string(p.event.foot_comment)
 	p.expect(yaml_SEQUENCE_END_EVENT)
+	p.hookPath = p.hookPath[:len(p.hookPath)-1]
 	return n
 }
 
@@ -293,7 +328,7 @@ func (p *parser) mapping() *Node {
 	p.expect(yaml_MAPPING_START_EVENT)
 	for p.peek() != yaml_MAPPING_END_EVENT {
 		k := p.parseChild(n, false)
-		p.hookPath = append(p.hookPath, k.Value)
+		p.hookPath = append(p.hookPath, k)
 		if block && k.FootComment != "" {
 			// Must be a foot comment for the prior value when being dedented.
 			if len(n.Content) > 2 {

decode_test.go

@@ -1725,13 +1725,12 @@ a:
 f:
   g:
     h:
-      i:
-        - 1
-        - 2
+      i: [1,2]
 j: {"k": true, "l": false}
 
 x:
-  - z: yes
+  - z:
+      zz: yes
     y: no
   - z: up
     w: down
@@ -1749,24 +1748,39 @@ x:
 	err := dec.Decode(map[string]interface{}{})
 	c.Assert(err, IsNil)
 
-	want := map[string]string{
-		"a.b": "foo",
-		"a.c.d.e": "bar",
-		"f.g.h.i.0": "1",
-		"f.g.h.i.1": "2",
-		"j.k": "true",
-		"j.l": "false",
-		"x.0.z": "yes",
-		"x.0.y": "no",
-		"x.1.z": "up",
-		"x.1.w": "down",
+	// values represent a slice that contains the expected line, column and value
+	want := map[string][]interface{}{
+		"a":         {2, 1, ""},
+		"a.b":       {3, 3, "foo"},
+		"a.c":       {4, 3, ""},
+		"a.c.d":     {5, 5, ""},
+		"a.c.d.e":   {6, 7, "bar"},
+		"f":         {7, 1, ""},
+		"f.g":       {8, 3, ""},
+		"f.g.h":     {9, 5, ""},
+		"f.g.h.i":   {10, 7, ""},
+		"f.g.h.i.0": {10, 11, "1"},
+		"f.g.h.i.1": {10, 13, "2"},
+		"j":         {11, 1, ""},
+		"j.k":       {11, 5, "true"},
+		"j.l":       {11, 16, "false"},
+		"x":         {13, 1, ""},
+		"x.0":       {14, 5, ""},
+		"x.0.z":     {14, 5, ""},
+		"x.0.z.zz":  {15, 7, "yes"},
+		"x.0.y":     {16, 5, "no"},
+		"x.1":       {17, 5, ""},
+		"x.1.z":     {17, 5, "up"},
+		"x.1.w":     {18, 5, "down"},
 	}
 	c.Assert(len(got), Equals, len(want))
-	for k,v := range want {
+	for k, v := range want {
 		comment := Commentf("key: %s", k)
 		node, ok := got[k]
-		c.Assert(ok, Equals, true, comment)
-		c.Assert(node.Value, Equals, v, comment)
+		c.Check(ok, Equals, true, comment)
+		c.Check(node.Line, Equals, v[0], comment)
+		c.Check(node.Column, Equals, v[1], comment)
+		c.Check(node.Value, Equals, v[2], comment)
 	}
 }
 

yaml.go

@@ -127,11 +127,11 @@ func (dec *Decoder) KnownFields(enable bool) {
 	dec.knownFields = enable
 }
 
-// DecoderHook is called for every leaf node and receives the node itself and
-// the path leading up to the node.
+// DecoderHook is called for every node and receives the node itself and the
+// path leading up to the node.
 type DecoderHook func(path []string, node *Node)
 
-// WithHook adds a DecoderHook that gets called for every leaf node.
+// WithHook adds a DecoderHook that gets called for every node.
 func (dec *Decoder) WithHook(h DecoderHook) {
 	dec.parser.withHook(h)
 }
@@ -495,6 +495,17 @@ func (e *TypeError) Error() string {
 	return b.String()
 }
 
+func (e *TypeError) Unwrap() []error {
+	if len(e.Errors) == 0 {
+		return nil
+	}
+	errs := make([]error, len(e.Errors))
+	for i,err := range e.Errors {
+		errs[i] = err
+	}
+	return errs
+}
+
 type Kind uint32
 
 const (