NodeProcessor: Ignore changes to certain fields upon node update

Thanks Alex. I did notice an initial problem with my hack - specifically around checksums. At the top of the process function, it has this code:

if ($hash == $this->getHash($nid)) {
  continue;
}

I ran into a case where I did a large import of a few thousand items. The taxonomy terms weren't mapping correctly, so I wanted to re-import a few times while I cleaned them up. After that I wanted manual control over those fields as indicated above.

This re-import captured the new hash, wrote it to the database, and then did not update my fields based on my logic from the initial post of this thread. The future imports where I did want to have those fields updated did not work because the checksums in the database had not changed, so the entire process was aborted. I had to ignore checksums and re-import every item.

So to you or anyone else who happens to want to selectively ignore fields upon update, it is important to properly integrate this logic with the hashing function.

Thanks for the logic as well Alex - I will try to follow that approach and post back results of my findings.

Thanks mxwixel, you actually helped me track down a problem I am having. Here's what was happening with my taxonomy fields. This is specific to a CCK content taxonomy field that is also set to use core taxonomy. I'm using this field to facet on with apachesolr, so it's important that it be there.

1) Import feed items to create a node. 1 new node created in this case. One of the imported fields is a CCK content taxonomy field that also saves to core taxonomy.

2) The created node is 16837 with values in the term_node taxonomy table as follows. We have a taxonomy field that showed up and was correctly populated from the import file.

nid    vid     tid
16837  16838   241

4) I then manually edit the node and set one of the blank (non-imported) content taxonomy fields to a term (also saves to core taxonomy.) This is because this particular field did not import a term for that item since I have partial data for some fields that require manual entry when they come in empty. After I saved the node, a look at term_node shows my new term in there:

nid    vid     tid
16837  16838   229 - NEW term added
16837  16838   241

5) I manually re-imported the feed. Let's say some random other field changed which triggered a different checksum and a new import. This updated my node we are talking about. But look what happened to the term_node table:

nid    vid     tid
16837  16838   241

6) Notice that the tid 229 is now gone. That's the one I manually added in step #4. The really bad part of this for me is that since this is a content taxonomy field, Feeds killed the value from term_node (which breaks Apachesolr and its facets since it is now gone), but when a user views the node it still LOOKS like the term is there because the CCK content taxonomy field is still populated. Feeds did not overwrite that, it just cut off the term_node table from underneath it.

I don't have a solution for this yet, but I absolutely need to find a way to stop this from happening. I need manual control over specific fields post-import and can't have any future imports removing tids from the term_node table from manually edited nodes.

I'd welcome any thoughts as to how to accomplish this.

The issue in #4 happens regardless of whether of not the source field exists in the import. In this case, the source mapping field isn't even there yet it still overwrites the term_node items when saved. I verified this by dumping the $node object right before the node_save().

I am wondering if my comment in #4 is a bug? On update I would assume it isn't ideal to blow away what was already in the taxonomy field. Here's a little bit of code from the file:

          // If updating populate nid and vid avoiding an expensive node_load().
          $node->nid = $nid;
          $node->vid = db_result(db_query('SELECT vid FROM {node} WHERE nid = %d', $nid));

Now have a look at this:
#471074: Taxonomy synchronization caching bug
#594004: taxonomy_node_get_terms() needs a $reset param that node_load can call when *it* is reset

Would it would be appropriate to include something like that fix to load the original taxonomy terms to ensure they are saved back to the node with the node_save? Are there other things that may not be included based on the avoidance of a node_load?

Perhaps this is becoming a different issue from excluding content.

I have a fix that stops it from dropping taxonomy. In FeedsNodeProcessor.inc, inside public function process, I now have the following which is based on #59004 above.

          // If updating populate nid and vid avoiding an expensive node_load().
          $node->nid = $nid;
          $node->vid = db_result(db_query('SELECT vid FROM {node} WHERE nid = %d', $nid));

          // START CHANGE - Do not forget the taxonomy terms
          if($node->nid = $nid) {
            $result = db_query("SELECT * FROM {term_node} WHERE nid = %d AND vid = %d", $node->nid, $node->vid);
            while ($term = db_fetch_object($result)) {
              $existing_terms[] = $term;
            }
            $terms = array();
            foreach ($existing_terms as $term) {
              $terms[] = $term->tid;
            }
            $node->taxonomy = $terms;
            unset($result, $term, $existing_terms, $terms);
          }