When you imagine an invention, you probably think of something that could be described in a few words, such as "a flying machine with fixed, curved wings" or "an electrical communicator with a microphone and a speaker". But most patents cover complex detailed processes that have no simple descriptions--often they are speedups or variants of well-known processes that are themselves complex.
Most of these patents are neither obvious nor brilliant; they are obscure. A capable software designer will "invent" several such improvements in the course of a project. However, there are many avenues for improving a technique, so no single project is likely to find any given one.
For example, IBM has several patents (including patent number 4,656,583) on workmanlike, albeit complex, speedups for well-known computations performed by optimizing compilers, such as register coloring and computing the available expressions.
Patents are also granted on combinations of techniques that are already widely used. One example is IBM patent 4,742,450, which covers "shared copy-on-write segments." This technique allows several programs to share the same piece of memory that represents information in a file; if any program writes a page in the file, that page is replaced by a copy in all of the programs, which continue to share that page with each other but no longer share with the file.
Shared segments and copy-on-write have been used since the 1960's; this particular combination may be new as a specific feature, but is hardly an invention. Nevertheless, the Patent Office thought that it merited a patent, which must now be taken into account by the developer of any new operating system.
Obscure patents are like land mines: other developers are more likely to reinvent these techniques than to find out about the patents, and then they will be sued. The chance of running into any one of these patents is small, but they are so numerous that you cannot go far without hitting one. Every basic technique has many variations, and a small set of basic techniques can be combined in many ways. The patent office has now granted at least 2000 software patents--no less than 700 in 1989 alone, according to a list compiled by EDS. We can expect the pace to accelerate. In ten years, programmers will have no choice but to march on blindly and hope they are lucky.