Mark Thomas

Adding a dependency to the Servlet API is a big deal that needs a wider discussion and broad consensus. Personally, I'm not (yet) convinced of the need for this.

Having looked at the various `SessionCookieConfig` tests in the TCK, none of them test the `cookie-config` element in `web.xml`.

The current `BeanELResolver` cache is at the instance level, not the class level. That was changed in #171. The entire cache should be GC eligible once the web application is...

I'll note that EL 6.0 was released at the start of this month so any changes being discussed would be for 6.1.

Tomcat has a similar approach to caching the factory implementation. The key difference is that is uses `WeakReference`s rather than requiring a `clear()` method. I think I prefer the `WeakReference`...

That behaviour of `com.sun.faces.el.ELUtils` is a memory leak waiting to happen. Whether that is an EL API issue or a mojarra implementation issue is debatable. We can probably protect the...

Replaced by #248

I have two concerns with this PR. 1. Given the existing `synchronized` keyword on both `registerCleaner()` and `unregisterCleaner()` I don't see how execution could reach line 1537 with `poolCleanTimer` both...

> please notice that both methods registerCleaner() and unregisterCleaner() having the synchronized keyword does not prevent that one thread executes registerCleaner() while another thread simulteneously executes unregisterCleaner() . Yes, it...

That the threads have different context class loaders doesn't explain the behaviour you are seeing. As per section 8.4.3.6 of the JLS, there is a single monitor for both methods....