IME abstractions generally begin with the latter case. Perhaps there's one or two corner cases that the abstraction also covers, but it seems justifiable at the

IME abstractions generally begin with the latter case. Perhaps there's one or two corner cases that the abstraction also covers, but it seems justifiable at the time because the core of the code really ought to work the same across all cases. Then you slowly start adding corner cases, or you change your new feature slightly so that the abstraction has to change. Little by little you end up with this insanely complicated (and unclean) abstraction which is arguably significantly worse than simply duplicating the code would have been in the first place. I make this remark because I think that "the code needs to be the same in a few places" may not be a sufficiently strong reason for writing an abstraction right away. Even in these cases, sometimes it's ok to leave duplication in your codebase for a little while while you let the feature you're working on shake itself out, then come back and see if the abstraction is worth writing. That's the only way to break the cycle.