Do scrum have a method for handling tasks which can not be estimated in any meaningful way? I'm going to give an example we faced recently. We sell some hardware devices and recently found out that after we upgraded our compiler the new compiler introduced a (subtle) bug on some of our older devices.
It is still unclear if the bug is in the old chip or the compiler but it doesn't really matter to us since we need to fix the problem - and the only way we can fix it is to locate the part of our code which is causing the problem and write it in a way so it compiles into a binary which runs as expected.
From our point of view it's a very black box problem. My initial gut feeling was that it would take 3 man weeks, but it ended up taking closer to 3 man months. If the developer had been lucky and started looking at the part of the code which turned our to be problematic from the start he could have solved it in 2-3 days.
We don't really know how to handle these problems. It is completely impossible to break down, and while working on it there is very little progression.
I like to compare it to making a mathematical proof. When you have the right idea it is fast and straight forward (or can be), but until then it can be extremely hard/impossible to say if you are close to solving it. You can sit for hours or days, and when the right idea suddenly pops into your head it's solved in 20 minutes.