html: refactor the blacklist for the "render and re-parse" test.

R=andybalholm CC=golang-dev, mikesamuel https://golang.org/cl/5331056

html: refactor the blacklist for the "render and re-parse" test.
R=andybalholm CC=golang-dev, mikesamuel https://golang.org/cl/5331056
90b76c0f · Nigel Tao · f753e3fa · 90b76c0f · 90b76c0f
Commit 90b76c0f authored Nov 01, 2011 by Nigel Tao
Hide whitespace changes
Inline Side-by-side

Showing with 35 additions and 16 deletions

parse_test.go src/pkg/html/parse_test.go +14 -6

render.go src/pkg/html/render.go +21 -10

No files found.
--- a/src/pkg/html/parse_test.go
+++ b/src/pkg/html/parse_test.go
@@ -160,14 +160,10 @@ func TestParser(t *testing.T) {
 				t.Errorf("%s test #%d %q, got vs want:\n----\n%s----\n%s----", filename, i, text, got, want)
 				continue
 			}
-			// Check that rendering and re-parsing results in an identical tree.
-			if filename == "tests1.dat" && (i == 30 || i == 77) {
-				// Some tests in tests1.dat have such messed-up markup that a correct parse
-				// results in a non-conforming tree (one <a> element nested inside another).
-				// Therefore when it is rendered and re-parsed, it isn't the same.
-				// So we skip rendering on that test.
+			if renderTestBlacklist[text] {
 				continue
 			}
+			// Check that rendering and re-parsing results in an identical tree.
 			pr, pw := io.Pipe()
 			go func() {
 				pw.CloseWithError(Render(pw, doc))
@@ -187,3 +183,15 @@ func TestParser(t *testing.T) {
 		}
 	}
 }
+
+// Some test input result in parse trees are not 'well-formed' despite
+// following the HTML5 recovery algorithms. Rendering and re-parsing such a
+// tree will not result in an exact clone of that tree. We blacklist such
+// inputs from the render test.
+var renderTestBlacklist = map[string]bool{
+	// The second <a> will be reparented to the first <table>'s parent. This
+	// results in an <a> whose parent is an <a>, which is not 'well-formed'.
+	`<a><table><td><a><table></table><a></tr><a></table><b>X</b>C<a>Y`: true,
+	// The second <a> will be reparented, similar to the case above.
+	`<a href="blah">aba<table><a href="foo">br<tr><td></td></tr>x</table>aoe`: true,
+}
--- a/src/pkg/html/render.go
+++ b/src/pkg/html/render.go
@@ -19,17 +19,28 @@ type writer interface {

 // Render renders the parse tree n to the given writer.
 //
-// For 'well-formed' parse trees, calling Parse on the output of Render will
-// result in a clone of the original tree.
+// Rendering is done on a 'best effort' basis: calling Parse on the output of
+// Render will always result in something similar to the original tree, but it
+// is not necessarily an exact clone unless the original tree was 'well-formed'.
+// 'Well-formed' is not easily specified; the HTML5 specification is
+// complicated.
 //
-// 'Well-formed' is not formally specified, but calling Parse on arbitrary
-// input results in a 'well-formed' parse tree if Parse does not return an
-// error. Programmatically constructed trees are typically also 'well-formed',
-// but it is possible to construct a tree that, when rendered and re-parsed,
-// results in a different tree. A simple example is that a solitary text node
-// would become a tree containing <html>, <head> and <body> elements. Another
-// example is that the programmatic equivalent of "a<head>b</head>c" becomes
-// "<html><head><head/><body>abc</body></html>".
+// Calling Parse on arbitrary input typically results in a 'well-formed' parse
+// tree. However, it is possible for Parse to yield a 'badly-formed' parse tree.
+// For example, in a 'well-formed' parse tree, no <a> element is a child of
+// another <a> element: parsing "<a><a>" results in two sibling elements.
+// Similarly, in a 'well-formed' parse tree, no <a> element is a child of a
+// <table> element: parsing "<p><table><a>" results in a <p> with two sibling
+// children; the <a> is reparented to the <table>'s parent. However, calling
+// Parse on "<a><table><a>" does not return an error, but the result has an <a>
+// element with an <a> child, and is therefore not 'well-formed'.
+// 
+// Programmatically constructed trees are typically also 'well-formed', but it
+// is possible to construct a tree that looks innocuous but, when rendered and
+// re-parsed, results in a different tree. A simple example is that a solitary
+// text node would become a tree containing <html>, <head> and <body> elements.
+// Another example is that the programmatic equivalent of "a<head>b</head>c"
+// becomes "<html><head><head/><body>abc</body></html>".
 func Render(w io.Writer, n *Node) os.Error {
 	if x, ok := w.(writer); ok {
 		return render(x, n)