Category: Work

The following error on my current project at work really gave me lots of headaches today:

SQLSTATE[HY000]: General error: 2014 Cannot execute queries
while other unbuffered queries are active. Consider using
PDOStatement::fetchAll().
Alternatively, if your code is only ever going to run against
mysql, you may enable query buffering by setting the
PDO::MYSQL_ATTR_USE_BUFFERED_QUERY attribute.

So, yes, I already have PDO::MYSQL_ATTR_USE_BUFFERED_QUERY set to TRUE, so why is PDO still complaining? And especially why it is complaining now, because the same code which triggered the error today ran without problems for the past 9 months?

After struggling a lot I found the cause: I missed to close a statement which was reused in a loop!

So take care that you always call PDOStatement->closeCursor() in use cases like this:

$stmt = $con->prepare(“SELECT * FROM doodle WHERE id = ?”);

foreach (range(1,10) as $id)
{
$stmt->execute(array($id));
$row = $stmt->fetch();
$stmt->closeCursor();
}

If you’ve worked with external repository definitions and branches before, you probably know the problem: If you create a new branch off an existing one or merge one branch into another, subversion is not smart enough to update svn:externals definitions which point to the same repository, but rather keep them pointing to the old (wrong) branch. (I read they fixed that with SVN 1.5 by supporting relative URLs, but still, a couple of people might not be able to upgrade and I want to keep rather explicit with external naming anyways.)

Anyways, today at work I was so sick of the problem that I decided should hack something together. Here is the result:

#!/bin/bash
export LANG=C
if [ $# -ne 2 ]
then
    echo "Usage:" $(basename $0) "<old> <new>"
    exit 1
fi

old=$1
new=$2
repo_root=`svn info | grep "Repository Root" | cut -f3 -d" "`;

if [ -n "$(svn info $repo_root/$new 2>&1 | grep "Not a valid URL")" ]
then
    echo "$repo_root/$new is not a valid URL"
    exit 1
fi

for ext in  $(svn st | grep -e "^X" | cut -c 8- | xargs -L1 dirname | uniq)
do
    externals=$(svn propget svn:externals $ext)
    if [[ "$externals" == *$repo_root/$old* ]]
    then
        externals=${externals//$repo_root\/$old/$repo_root\/$new}
        svn propset svn:externals "$externals" $ext
    fi
done

Save this into a file, make it executable and you’re good to go! The script is smart enough to check if the target URL (based on the repositories’s root and the given <new> path) actually exists and also only changes those external definitions which actually match the repository root.

Fun!

I encountered a small, but ugly problem in our Symfony-driven project today: Unauthenticated AJAX requests, which f.e. may happen when the session timed out on the server, but the user hasn’t reloaded the page in the meantime, are also forwarded to the globally defined login module / action. This of course leaves the HTML page, which is constructed upon single HTML components, in a total mess. Ouch!

So yeah, rendering the complete login mask HTML as partial to the client is stupid, but also relatively easy to fix:

public function executeLogin($request)
{
    if ($request->isXmlHttpRequest())
    {
        // renderJSON is a custom function which json_encode's
        // the argument and sets an X-JSON header on response
        return $this->renderJSON(array("error" =>
                                    "Your session expired"));
    }
    ...
}

This was of course only half of the fix. I still had to handle this (and other) special JSON response on the browser’s side:

new Ajax.Request("mymodule/myaction", {
     onSuccess: function(response, json) {
         if (json.error)
         {
             // display the error
             alert(json.error);
             return;
         }
         // the actual callback functionality
     }
}

Uh, anyone screams “spaghetti code”? Yeah, you’re right. I quickly headed for a more general implementation, also since we can’t do that for a couple of symfony-specific prototype helpers anyways, like update_element_function, whose Javascript code gets generated by Symfony dynamically. So how can this be generalized?

Ajax.Responders to the rescue

Actually, prototype already contains some kind of “global hook-in” functionality for all Ajax requests triggered by the library: Ajax.Responders.

While this seemed to support all common callbacks (among them onCreate, onSuccess, onFailure and onComplete), some testing showed though that f.e. the onComplete callback was always called after the specific AJAX request’s onComplete callback, so this was pretty useless for me. After all, I also wanted to prevent that the specific callback gets executed when I encountered an error…

After diving through prototype’s code for some hours I found a solution. Particularily helpful here is that prototype signals every created Ajax request to the onCreate handler and gives the request and response object handling this request as arguments to it. Time to overwrite prototype’s responder code! Here is it:

Ajax.Responders.register({
    onCreate: function(request) {
        var oldRespondToReadyState = request.respondToReadyState;
        request.respondToReadyState = function(readyState) {
            var state = Ajax.Request.Events[readyState];
            var response = new Ajax.Response(this);
            var json = response.headerJSON;
            
            if (state == 'Complete' && json && json.error)
            {
                alert(json.error);
                return;
            }
            oldRespondToReadyState.call(response.request, 
                                           readyState);
        }
    }
});

Another particularily useful piece of knowledge I gathered today to let this work is how Function.prototype.call and Function.prototype.apply work (both are available since Javascript 1.3).
Basically they allow the execution of a function in scope of the object given as first parameter (there is a nice introduction available here).

If you’ve ever wanted to “send an event to some object to make its listener fire” because the listener’s code depended on the fact that the this reference points to the object the event was fired upon, you should now have a viable alternative:

Event.observe(myObj, 'click', myHandler);
// is call-wise equivalent to
myHandler.call(myObj);

No need to create custom mouse events and throw them around any longer… 😉