dockerfile: handle comments correctly

I consulted the official Dockerfile parser, and this seems more correct.
Initial whitespace is ignored when parsing comments, which is done first
on every line. Dockerfile comments (as opposed to comments in any
embedded language) can only start a line, and are logically deleted from
the token stream. Except they're not because actually ignored because
they can contain parser directives.

The handling of leading spaces continues to be a divergence from the
official parser, because its behavior is bonkers.

Signed-off-by: Hank Donnay <hdonnay@redhat.com>

rhel/dockerfile/dockerfile.go

@@ -95,12 +95,15 @@ func (p *labelParser) Run() error {
 					return err
 				}
 				for _, kv := range pairs {
-					i := strings.IndexByte(kv, '=')
-					k, _, err := transform.String(p.unquote, kv[:i])
+					idx := strings.IndexByte(kv, '=')
+					if idx == -1 {
+						return fmt.Errorf(`invalid syntax: %+#q`, i.val)
+					}
+					k, _, err := transform.String(p.unquote, kv[:idx])
 					if err != nil {
 						return err
 					}
-					v, _, err := transform.String(transform.Chain(p.unquote, p.vars), kv[i+1:])
+					v, _, err := transform.String(transform.Chain(p.unquote, p.vars), kv[idx+1:])
 					if err != nil {
 						return err
 					}
@@ -127,12 +130,15 @@ func (p *labelParser) Run() error {
 				return err
 			}
 			for _, kv := range pairs {
-				i := strings.IndexByte(kv, '=')
-				k, _, err := transform.String(p.unquote, kv[:i])
+				idx := strings.IndexByte(kv, '=')
+				if idx == -1 {
+					return fmt.Errorf(`invalid syntax: %+#q`, i.val)
+				}
+				k, _, err := transform.String(p.unquote, kv[:idx])
 				if err != nil {
 					return err
 				}
-				v, _, err := transform.String(transform.Chain(p.unquote, p.vars), kv[i+1:])
+				v, _, err := transform.String(transform.Chain(p.unquote, p.vars), kv[idx+1:])
 				if err != nil {
 					return err
 				}

rhel/dockerfile/lex.go

@@ -58,6 +58,8 @@ type item struct {
 
 type itemKind int
 
+//go:generate stringer -type itemKind $GOFILE
+
 const (
 	itemError itemKind = iota
 	itemComment
@@ -118,13 +120,18 @@ func (l *lexer) consumeWhitespace() (err error) {
 func (l *lexer) collectLine() (err error) {
 	var r rune
 	var sz int
-	var esc bool
+	var esc, inComment, started bool
 Read:
 	for r, sz, err = l.rd.ReadRune(); err == nil; r, sz, err = l.rd.ReadRune() {
 		switch {
+		case inComment && r == '\n':
+			inComment = false
+			started = false
+		case inComment: // Skip
 		case esc && r == '\r': // Lexer hack: why do some things have DOS line endings?
 		case esc && r == '\n':
 			esc = false
+			started = false
 		case esc:
 			// This little lexer only cares about constructing the lines
 			// correctly, so everything else gets passed through.
@@ -134,9 +141,17 @@ Read:
 			_, err = l.sb.WriteRune(r)
 		case r == l.escchar:
 			esc = true
+			started = true
 		case !esc && r == '\n':
 			err = l.rd.UnreadRune()
 			break Read
+		case !started && !esc && r == '#':
+			inComment = true
+		case !started:
+			if !unicode.IsSpace(r) {
+				started = true
+			}
+			fallthrough
 		default:
 			_, err = l.sb.WriteRune(r)
 		}

rhel/dockerfile/testdata/Dockerfile-Comments

@@ -0,0 +1,9 @@
+# Normal comments are fine
+FROM scratch
+
+LABEL \
+	# This style of comment is buck wild, but allowed.
+	A=B \
+	C=D
+	# This comment is stripped despite the apparent continuation
+

rhel/dockerfile/testdata/Dockerfile-Comments.want

@@ -0,0 +1,4 @@
+{
+  "A": "B",
+  "C": "D"
+}

rhel/dockerfile/testdata/Dockerfile-InvalidLabel

@@ -0,0 +1,2 @@
+LABEL A=B \
+	C=D # Comments not allowed in this position

rhel/dockerfile/testdata/Dockerfile-InvalidLabel.want

@@ -0,0 +1,4 @@
+{
+  "A": "B",
+  "C": "D"
+}

rhel/dockerfile/testdata/Dockerfile-InvalidLabel.want.err

@@ -0,0 +1 @@
+invalid syntax: `A=B 	C=D # Comments not allowed in this position`