14743 2011-11-09 18:47:16 +0000 Indexing NodeList out of range should return undefined 2011-11-09 20:04:04 +0000 1 1 1 Unclassified WebAppsWG DOM unspecified All All RESOLVED INVALID P2 normal --- 1 ayg annevk mike www-dom public-webapps-bugzilla oldest_to_newest 59790 0 ayg 2011-11-09 18:47:16 +0000 Try this on Live DOM Viewer: <!DOCTYPE html> <script> try {w(document.childNodes[-1])} catch(e) {w(e)} try {w(document.childNodes.item(-1))} catch(e) {w(e)} try {w(document.childNodes[20])} catch(e) {w(e)} try {w(document.childNodes.item(20))} catch(e) {w(e)} </script> IE9: undefined, null, undefined, null Firefox 9.0a2: INDEX_SIZE_ERR, null, undefined, null Chrome 16 dev: undefined, INDEX_SIZE_ERR, undefined, null Opera 12.00: undefined, null, undefined, null Correct behavior is clearly IE/Opera, IMO. Throwing anywhere here is annoying because it means you can't use if (document.childNodes[i]) to test whether the child exists. The current spec says NodeList.prototype.item is a getter, and therefore as far as I can tell its behavior has to match actual indexing per WebIDL. The correct fix is probably to make item() not a getter, and define separate getting behavior that returns undefined. 59795 1 ayg 2011-11-09 20:04:04 +0000 Hixie pointed me to the correct part of WebIDL. I think it contradicts the parts I was looking at, so I'll file some WebIDL bugs, but DOM4 is correct, anyway.