Earlier quoted context omitted.
I skimmed it and saw at least two attempts that were backed out for various reasons.
From a rudimentary reading. They introduced a state cache to fix conditions where applications poll continuously for certain states (mouse button press for example). This wasn't thread safe however, and multi-threaded applications could get a previous/incorrect state back from the cache and stop functioning as intended. This was highly visible for photoshop tools where you click and hold, as the separate mouse poller…
"eventually" is right...
This seems to be a failure on the part of the devs to prioritize. AFAIK code that causes a regression should be reverted and fixed properly, not left affecting the product for FIVE YEARS before finding a fix. The burden can not be on every use case to work around a bug that was introduced, the core "fix" is clearly a half-fix and should have been reverted on identification.