On Tue Apr 25 17:11:34 2023 +0000, Jacek Caban wrote:
Making our forced collection more aggressive seems reasonable to me, but I'd hope for a better interface for it. Depending on some side effect of otherwise unrelated logging does not look solid. If Gecko doesn't expose needed interface, it shouldn't be too hard to extend Gecko. Message pumping in destructor would still be a problem. I guess it's a problem only in some specific tests. We could perhaps just pump window messages before destroying the object in problematic tests themselves.
I'm not sure if message pumping in the tests is reliable for this, unfortunately. It doesn't happen immediately, since it has some complex heuristics to do the CC. In tests it's not uncommon for the last 3 or 4 tests to leak. It even seems different on each run…