Sometime Today, Siddhesh Poyarekar assembled some asciibets to say:
I have a doubt though, probably kernel hackers on list can help. Does the core+servers design of pure microkernels make it any more robust? Meaning, is it possible for memory management to fail without affecting the core and FS manager or something like that?
Ideally implemented, it would, but you'll often notice that most software (even opensource software) is written by humans who have little time or motivation to be thorough and complete in their work. They sometimes cut corners to get a feature out in time, but most often it's just that they didn't know better.
When you start to do things with software that the original author never expected (but fall into the realm of possible albeit improbable uses of the system), that's when you notice the programmer's limitation.
I am not a kernel hacker either, nor am I an expert in operating systems. The only thing I have is a reasonable knowledge of how most people program computers, and in my experience, it's far from perfect. It is also my experience that perfection is rarely practical if attempted all at once. If developed for a very long time, an application will evolve towards perfection, but how long do most apps continue in development?