FeedAPI reports "1 existing item(s) were updated" when consuming a brand-new feed

I just retested this:
- emptied database from existing D6 install, reinstall drupal
- current feedapi CVS
- parser_simplepie/simplepie.inc 1.1.3
- enable feedapi, feedapi_node, parser_simplepie
- create feed node with http://www.php.net/releases.atom and default options (see first attachment)
- click "refresh", issue is observed as reported! (second attachment)
- open /admin/content/node and observe two items of content (one is feed, one is content)

I just tried this yet again, with an utterly empty site (from scratch, empty filesystem and database). Same results as above. Attached is a complete log of my drupal installation, and new screenshots (same as above).

Notice, in particular, that only one of the two nodes has the "new" tag.

Next, I installed devel module (fourth attachment), so I can look at those created nodes more carefully. The new feed item looks fine (fifth attachment). Time to add some debugging to feedapi to figure out why it thinks it's updating an item...

Adding some debug output to _feedapi_invoke_refresh() shows that the feed looks basically okay, and it's processing both nodes, per first attachment. It's calling _feedapi_node_unique() to determine whether the items are distinct, and it thinks that they're the same! This appears to be because of:

  // Feed item is duplicate, if URL or GUID are duplicate or if they are both missing.
  if (isset($feed_item->options->original_url)) {
    $nid = db_result(db_query("SELECT fni.nid FROM {feedapi_node_item} fni JOIN {feedapi_node_item_feed} ff ON ff.feed_item_nid = fni.nid WHERE fni.url = '%s' AND ff.feed_nid = %d", $feed_item->options->original_url, $feed_nid));
    if ($nid !== FALSE) {
      return $nid;
    }
  }

When $feed_item->options->original_url is an empty string (as opposed to unset), like it is in this case, it will keep thinking nodes are duplicates.

So... the problem is really that the feed item original_url should not be set, rather than being empty. Or, this check should treat the two the same.

It seems this was broken by this change which added html_entity_decode() to the output of simplepie. However html_entity_decode(NULL) ==> '', which is what's causing the problem downstream.

Suggest we change this to check for NULL first, see attached patch.

(FWIW, my original report was confused by the two nodes - one of which was the _feed_ and one of which was the created+updated feed item)

That's what I mean by '' - an empty string (which is what's breaking _feedapi_node_unique)

From #6: "When $feed_item->options->original_url is an empty string (as opposed to unset), like it is in this case, it will keep thinking nodes are duplicates."