My operating system is a huge and complex beast. But a lot it, if not most of, is modularized in a way that extending it never requires my touching any of its source files. It assumes a mission statement to the user (I'll make one up and say it's "I will organize your computer's resources and let you use them easily") and code contracts that allow multiple components to serve this goal. Internally, the components are abstracted into statements of a very simple logic-flow.
This paradigm persists in many components themselves, from the bootloader to the end-user. When I want to update something in user space, it involves, from my perspective, no interaction with the operating system (even though it is really the one doing everything). Conversely if I want to do something as advanced as completely rewrite my filesystem implementation (something my operating system is very dependent on), I can, and without ever needing to touch my operating system's source. I could come up with any ridiculous way for a filesystem to work, and my operating system would be cool with it, and would operate the same way because if its modularized and abstracted design.
I would say these are incredibly good design features and that you would have to be extremely hard pressed to find a system that inherently cannot be designed in such a way as to mitigate maintainability and trivialize extensibility. Is your top-level project inherently so complex that adding smaller pieces requires that much screen real estate? If so, you have my sympathies.
> I had meant to ask if you analyse your profiling data on the command line. Do you?
Yes, my command-line editor has some powerful cvs support and manipulation features that make slicing the data up very easy.