Paradoxically, bubblesort can actually be pretty fast on small inputs (i.e. short arrays).

I once experienced this first hand in C codebase. I had to sort an array of structs where the array never contained more than ~50 members, and the comparison was very simple; in that case, bubblesort actually ran faster than the qsort(3) supplied by the compiler.

Then again, at such small scale, the performance win is so tiny it usually is not worth the effort.