[dhardy]

This is ~500 lines of code that we don't need in the main Rand crate IMO. (The reason WeightedIndex is there is that it is used by the seq module; but the alias_method variant is not since it's initialisation cost makes it completely inappropriate for this single-use API.)

Question: should we deprecate the weighted pub module in rand, since this module doesn't need to be public now? (I guess this might require some re-exports or something internally, but we can do it.)

Similarly, the weighted pub module and alias_method sub-module in rand_distr seem somewhat unnecessary.

[vks]

Question: should we deprecate the weighted pub module in rand, since this module doesn't need to be public now? (I guess this might require some re-exports or something internally, but we can do it.)

Probably, but arguably this can be decided after merging this. We should probably have some release where it is just deprecated anyway. For rand_distr, maybe flattening the module structure from weighted::alias_method to weighted_alias_method makes sense?

[dhardy]

Well, it also makes sense to tie up loose ends while we're on the topic. I'll make a change to deprecate the rand::distributions::weighted module then.

Regarding rand_distr, I think not, (1) because it's a breaking change with little motivation, and (2) because alias_method does use the error type from weighted.

[dhardy]

Right, could you still review this @vks so I don't have to "force merge"?

[vks]

Sorry, I forgot about this. The failure seemed unrelated, so I went ahead and merged.