wine-gecko does actually support synchronous XMLHttpRequests, even if they are deprecated, but unfortunately it is very broken compared to IE or all the other major browsers (more details [here](https://bugzilla.mozilla.org/show_bug.cgi?id=697151)). Thankfully, it still provides us with the event message loop, which is enough to fix it on mshtml side and act like IE.
For sync XHRs, send() is supposed to block all script code, except for notifications sent to the sync XHR. This is because when send() blocks, no other piece of code in the script should execute other than the sync XHR handlers. Unfortunately, that's not the case in Gecko's broken implementation, which can execute handlers as if they were APCs while it's "blocked" in send().
Note that it doesn't actually block, though, because we still process the message loop (non-event related tasks such as navigation, paints, and other stuff), and that's normal. Gecko doesn't block everything related to script events, only some things on the document and timers, but that's far from enough and not even enough for our purposes since we use our own timer tasks. And not even message events are blocked, even though we *also* use our own message event dispatchers (but it doesn't block Gecko ones either).
So what we have to do is we need to track all the timers and events dispatched that result in script handlers being executed, while "blocking" inside of a sync XHR send(), and queue them up to defer them to be dispatched later, after send() unblocks. But this is easier said that done. There are many corner cases to consider here, for example:
* Events dispatched *synchronously* from within a sync XHR handler or other piece of code called from it. * Nested sync XHR send() calls (called during handler of another sync XHR). * Async XHRs having their events dispatched during a blocking sync XHR send() are complicated. `readyStateChange` for example needs to have the async XHR's readyState at the time it was sent, **not** at the time it was actually dispatched, since we're going to delay it and dispatch it later. It's similar with other XHR states, such as responseText (which can be partial). * Aborts of async XHRs during such handlers.
These patches hopefully should address all the issues and, on a high level, work like this:
* Track the `readyState` and `responseText` length manually, so we can override / force them to specific values. * "Snapshot" the async XHR at the time we encounter an event for it, and queue this event with such information. When later dispatching this event (after being deferred), we temporarily set the state of the async XHR to the snapshot. * To deal with nested event dispatches, keep track of a "dispatch depth" (everytime we dispatch an event we increase the depth, and when it returns we decrease it). * For sync XHRs, we note the depth when send() is called, and defer events at that depth, since they're dispatched during the "blocked" message loop and need to be delayed (except for sync XHR events, which is a special case since it must be dispatched synchronously). When it returns, we restore the previous blocking depth.