Issue Type: Bug Created: 2009-04-13T07:13:01.000+0000 Last Updated: 2009-09-18T08:03:57.000+0000 Status: Resolved Fix version(s): - 1.9.3 (22/Sep/09)
Reporter: Ivo Jansch (ijansch) Assignee: Pádraic Brady (padraic) Tags: - Zend_Service_Twitter
Related issues: Attachments: - Zend_Service_Twitter.diff
I have 2 Zend_Service_Twitter objects, both with their own account credentials. After sending an update with one object, an update on the next object will use the wrong useraccount.
$a = new Zend_Service_Twitter("userA", "passwordA"); $b = new Zend_Service_Twitter("userB", "passwordB");
$a->status->update("This is a test"); $b->status->update("This is a test too");
// result, both messages get posted in userA's timeline.
Posted by Ivo Jansch (ijansch) on 2009-04-13T08:53:34.000+0000
I investigated a bit further, and the problem probably is is that Zend_Service_Abstract, which Zend_Service_Twitter is indirectly derived from, uses a protected static $_httpClient member to handle rest calls. This is probably a design flaw, since this could lead to a number of problems where various different objects could inadvertently get unexpected results, since they share a single common http client object.
Posted by Jon Whitcraft (sidhighwind) on 2009-04-30T17:30:38.000+0000
Actually Zend_Service_Twitter does not extend Zend_Service_Abstract. It extends Zend_Rest_Client as it's using their REST API. This might be a design flaw in the Rest Client but with no maintianer of the Rest Service I'm not sure how/if this will ever get fixed. I think the twitter client should get a rewrite and us the Zend_Service_Abstract method instead but that would push it off to 2.0.. I'll send this off to Zend to see what they say.
Posted by Ivo Jansch (ijansch) on 2009-05-01T02:56:39.000+0000
Actually it is; which is why I said 'indirectly'.
Zend_Service_Twitter extends Zend_Rest_Client, but Zend_Rest_Client in turn inherits from Zend_Service_Abstract. (I think composition would be a better design for the service api than inheritance by the way).
So the problem is still with Zend_Service_Abstract and its static $_httpClient member. Does Zend_Service_Abstract have a maintainer?
Posted by Jon Whitcraft (sidhighwind) on 2009-05-01T03:19:11.000+0000
Ahh see if i would have followed the tree all the way i found i found that. My mistake. I have a feeling that it doesn't reset the variables like it should. I ran into this problem with the Ec2 Component where if you did more than one request it would fail as the $_httpClient was tainted with the previous requests items.
I'll look into this further.
Posted by Ivo Jansch (ijansch) on 2009-05-01T03:32:23.000+0000
Yes, that's indeed another symptom of the same issue. I think the cleanest solution is to replace the static $_httpClient with one that is not shared between instances, in other words a protected member variable.
Posted by Ben Scholzen (dasprid) on 2009-07-22T07:22:34.000+0000
I just ran into the same problem and after some looking around, I found a workaround which worked for me. Before every statusUpdate() call, I did:
<pre class="highlight"> $twitter->setPassword($twitter->getPassword());
That did reset the _authInitialised, so it updated the HTTP client when doiung the next statusUpdate().
Posted by Ivo Jansch (ijansch) on 2009-07-26T14:28:54.000+0000
Thanks Ben. I can confirm that this indeed is a working workaround.
The disadvantage is that this causes a re-authentication upon every update, even if they are from the same account. While this is acceptable in my situation, this is something to take into account if you do high volumes of updates.
Posted by Pádraic Brady (padraic) on 2009-09-17T13:48:55.000+0000
This patch implements some changes in an effort to ensure each Zend_Service_Twitter instance uses a distinct local HTTP client object to prevent the conflcts noted above. The summary of changes:
Note: This allows the setting of a custom client statically, so long as you remember they are cloned once a new object is created (i.e. you cannot change the clients once the object is instantiated which does differ from previously)
Posted by Jon Whitcraft (sidhighwind) on 2009-09-17T16:48:08.000+0000
It looks good to me. Let me konw when this is in so I can update the block addition for this and get it committed.
Posted by Ivo Jansch (ijansch) on 2009-09-18T00:10:10.000+0000
Thanks for the patch, as soon as I have a chance I'll test it in a real world scenario.
Posted by Pádraic Brady (padraic) on 2009-09-18T04:11:28.000+0000
I can commit the patch this afternoon if it looks fine.
Posted by Pádraic Brady (padraic) on 2009-09-18T08:03:57.000+0000
Issue fixed by applying patch (with some changes) in r18243
Have you found an issue?
See the Overview section for more details.