RFC: Stable Sort

[dubiousjim]

This RFC proposes introducing a stable sorting algorithm to the table library. The particular algorithm proposed is "Block Merge Sort."

Rendered

[ccuser44]

A potential problem though is that the semantics of table.sort are unclear and the validity of table.stablesort existing depends on that. I agree though that the sorting mechanisms of table.sort can be somewhat problematic in some scenarios

[dubiousjim]

Sorry, I didn't notice this review/comment when it was posted. I think I agree: I think the best (at least eventual) target would be to have a single sort function that was stable and still had pretty good performance. From my reading, this algorithm seems to be a good way to achieve that. The main alternative would be something in the timsort family. I went with blocksort because it's configurable to take minimal (or a fixed amount) of auxiliary memory, which timsort implementations don't do. (The design of timsort would make doing so challenging, but maybe there are tricks one could employ that aren't immediately obvious.)

The downside of both of those options is the sorting code becomes substantially more complex. It wouldn't be reasonable to include it inline in the table library file; you'd definitely want to separate the source into its own file.

Where do we go from here? Do you want me to post my implementation (here, or somewhere else)? Discuss the design further? Wait for more reviews?