The software is a web forum and the action could for example be a „mark this thread as read” action. This would send a
POST AJAX request to a URL with a slug, e.g.
/self/2015/oct/2/twitter-sponsored-links-ausblenden-mit-eigenem-user-punkt-punkt-punkt-css/mark_read.json. I could not find the problem, I wasn't even able to reproduce the problem, neither on production nor on the development system.
After a lot of hence and forth with the user and some carefully placed debugging code I was able to get the exception text:
InvalidAccessError. I had a look at the standard and it stated:
This was weird. It is an async request, no CORS, no username/password.
withCredentials was false,
responseType was an empty string. For me there was no reason that this exception should be thrown. I even checked if the CSRF token was valid (it was). So I ignored this issue for some time, since I could not find a solution.
This evening I was finally able to reproduce the problem. While reading in the web forum suddenly I got the said error message. I fired up the inspector and checked what was going on. Totally weird: there was no request taking place, no network activity at all.
I fetched a database dump from production and restored it in development. This time I could reproduce the problem in development as well. So I spend a lot of time debugging, I even changed the URL the request was sent to – and bum, the request got through.
I was totally confused, what the heck was going on? I went step by step through the code for the gazillions time and this time I even stepped into
XMLHTTPRequest.open(). And this time I saw a function called
function block(m, u) in the call stack. And slowly it dawned to me: the request failed only on some URLs. This time the URL contained the keyword
Quickly I disabled the ad blocker (I'm using uBlock) and re-tried – and this time the request went through without any problem. Wow! The ad blocker replaces
XMLHttpRequest.open() by its own code and checks the request URLs for keywords on a block list. If the URL matches it would throw an
This was really surprising for me. What a journey – it took me round about three months to find this problem.
Update: Having this tested with Chrome would have saved me a lot of time. Chrome explicitly states:
Failed to load resource: net::ERR_BLOCKED_BY_CLIENT
This error message is much more useful. Firefox, on the other hand, is as uninformative as Safari is: it just states
error (even more generic than