Bert JW Regeer (畢傑龍)

Interesting bug

I was starting to see something interesting in the output from my Google AppEngine based project. I was automatically creating a link based on a config setting, named SETTINGS["contact"], and the code would work perfectly the first time the page was created and loaded, however multiple refreshes of the same page would show that the link was gaining slashes even when it shouldn't have.

The logic was as follows:

if contact is "mailto:" then do nothing
        if contact does not start with "/" or not with "http"
                contact = "/" + contact

This code would work perfectly the first time around, however if I had properly tested by setting contact = "/contact" as well, I may have caught this bug faster.

However since that piece of code was only being called once per page view, it worked perfectly as far as I was aware, the thing is I started seeing more and more slashes being prepended to contact each time I refreshed the page. I looked at the code above, and did not see that my logic should have contained an "and" instead of an "or". Even so if the code is being loaded once per page view the variable "contact" that started off containing the word "contactme" should only turn into "/contactme" and not "////contactme" on the fourth page load.

Each time I added in a logging or debugging statement and refreshed the page it would go back to "/contactme", I would change another thing, or add another debug statement and "/contactme" was there staring me in the face. It took me a little while before I hit refresh a couple of times and noticed that my debug output now contained what I had seen beforehand. That is when it hit me, Google AppEngine caches imports that included my global variables. As long as I did not modify any of the files in my project it would happily use the cached variable, which after the first reload would contain "/contactme", then because of my use of or instead of and, would become "//contactme" on the second reload.

Google AppEngine has some very aggressive caching, which may well be required to have it run at the scale Google wants it to run. This is just another warning for myself that I need to watch out when I modify global variables and that they will stay the same across different sessions.

This is the logic I ended up with:

if contact is "mailto:" or contact starts with "http" then do nothing
        if contact does not start with "/"
                contact = "/" + contact

Works exactly how I had expected it to work.