This is difficult, to say the least, especially because I hate about any language, for various reasons. (even though python is working hard on shaking that statement).
For example, I hate to have nontrivial production code in C. Imagine building a large compiler in C. Yes, it works, and its fast, but in general, you are going to slam your head into a wall, because implementing simple optimizations result in programming more and more and more and more and then even more, and then, there is a pointer mistake and you can spend 3 days debugging, just to learn that someone of your coworkers did a mistake? bleh.
On the other hand, I don't want C to disappear. To some extent, C is a good compiler platform. It just works everywhere (Ha. Platforms that have no C-Compiler on them are no real platforms, just toys. :P). Furthermore, C has amazing compilers lying around, so compiling into C will result in a cruel speedup for your application.
On the other hand, consider Haskell. One fellow hardcore-functional programmer (he even is a professor researching parallel processing techniques using functional languages.) He discarded haskell, because "haskell is too academic". Heh. My thoughts. Building a simple thing that reads numbers from stdin and dumps them on stdout is a chore in Haskell, at least in my opinion, no matter how nice and fun things grow later on.
On the other hand, looking at Haskell taught me a few very interesting concepts (think the theory of monads). It would be ungrateful to want a language that taught you very interesting things to die / vanish / whatever.
Furthermore, I hate all languages, that try to restrict me in order to makes things "easier". Great. Because C++ f****d it's implementation of multiple inheritance up, Java has no MI anymore. (From what I know, in C++, you have to resolve problems in MI hierachies by hand. This implies that, given 3 projects and 2 programmers, you will end up with at least 6 implementations of MI an each will have 3 subtle bugs that will fall on your face if you use the MI implementation there and you are sure that everything works). However, No one considered for example Common Lisps implementation of Inheritance (or, pythons. in general, Linearizing MI. (And dont start with Interface Inheritance. Yes, you can inherit multiple interfaces so the interface is consistent, but in the current project, I still would have to implement the same code about 23 times (I counted))). Here, MI is well defined and -- even though it still is a bit tricky to use properly -- well usable. But ok, someone might not understand whats going on and thus, we need to remove that feature. Bleh.
Similar rants can be held for Makros, great in Lisp, hirrible in C, even worse with Templaces in C++. Thus, no one has them anymore, creating a syntactic barrier at refactoring.