json_decode() destroys important information

The current implementation of DrupalApacheSolrService uses solr's JSON Response Writer to receive the response as json.

Unfortunately the json response does not contain all information compared to the other response writers available. That's a documented behavior (see section "JSON specific parameters"):

Using a JSON object (essentially a map or hash) for a NamedList results in the loss of some information.

In Apache Solr Multilingual we support multiple spell checkers in different languages. If you use the standard response writer the corresponding part in the XML looks like this:

 <lst name="spellcheck">
  <lst name="suggestions">
    <lst name="intolerans">
      <int name="numFound">1</int>
      <int name="startOffset">0</int>
      <int name="endOffset">10</int>
      <arr name="suggestion">
        <str>Intoleranz</str>
      </arr>
    </lst>
  </lst>
</lst>
<lst name="spellcheck">
  <lst name="suggestions">
    <lst name="intolerans">
      <int name="numFound">1</int>
      <int name="startOffset">0</int>
      <int name="endOffset">10</int>
      <arr name="suggestion">
        <str>intolerance</str>
      </arr>
    </lst>
  </lst>
</lst>

The example shows a German and an English correction of "intolrans". Using XML it's no problem to access both suggestions.

If you use JSON, all suggestions are present in the result, but json_decode() collapses them and the latest element wins:

"spellcheck":{
    "suggestions":{
      "intolerans":{
        "numFound":1,
        "startOffset":0,
        "endOffset":10,
        "suggestion":["Intoleranz"]}}},
"spellcheck":{
    "suggestions":{
      "intolerans":{
        "numFound":1,
        "startOffset":0,
        "endOffset":10,
        "suggestion":["intolerance"]}}},

An interesting point is, how the available PHP Response Writers deal with that. Both, the php and phps responses automatically add counters to identically named elements:

'spellcheck'=>array(
    'suggestions'=>array(
      'intolerans'=>array(
        'numFound'=>1,
        'startOffset'=>0,
        'endOffset'=>10,
        'suggestion'=>array('Intoleranz')))),
'spellcheck 1'=>array(
    'suggestions'=>array(
      'intolerans'=>array(
        'numFound'=>1,
        'startOffset'=>0,
        'endOffset'=>10,
        'suggestion'=>array('intolerance')))),

a:1:{s:11:"suggestions";a:1:{s:10:"intolerans";a:4:{s:8:"numFound";i:1;s:11:"startOffset";i:0;s:9:"endOffset";i:10;s:10:"suggestion";a:1:{i:0;s:10:"Intoleranz";}}}}s:12:"spellcheck 1";a:1:{s:11:"suggestions";a:1:{s:10:"intolerans";a:4:{s:8:"numFound";i:1;s:11:"startOffset";i:0;s:9:"endOffset";i:10;s:10:"suggestion";a:1:{i:0;s:11:"intolerance";}}}}}

So I see different ways to solve this issue:
1. Use a better response writer than json in general. (Heavy rewrite and API changes)
2. Create patches for solr itself to let the json response writer behave like the php response writers. (Might be the best solution in the future but might not solve the issue for older solr 1.x and 3.x versions)
3. Fake a valid json result on our side.

I see two ways to implement solution 3 until solution 2 will be available. I will post a patch soon.