We did not because we we did not want reproducibility in this case.
Consider a piece of code that, run on the same data, produces different results in different runs. Assume the random behavior is not an intentional outcome of the procedure/formulas that the code implements. Then the observed random behavior indicates a mistake in the procedure/formulas or a mistake in the code. I've seen both.
In my experience, indeterministic sorts are the second most likely cause of different-results-in-different-runs problems. The most likely cause is uninitialized variables. Besides the statistical goals I mentioned in the posting, another of our goals in writing -sort- so that it randomly ordered ties was to highlight indeterministic sorts that lead to irreproducible results. At StataCorp, we run certification scripts in a loop, over and over again, looking for just such problems.
Setting the random-number seed is a way of reproducing results from routines that are intended to produce different results in different runs. -sort- is not such a function; if it produces different results in different runs, and that matters, that is a bug.
