The GADT debate

[Richard Eisenberg]

On May 7, 2016, at 11:05 PM, Gershom B <gershomb at gmail.com> wrote:
> 
> an attempt (orthogonal to the prime committee at first) to specify an algorithm for inference that is easier to describe and implement than OutsideIn, and which is strictly less powerful. (And indeed whose specification can be given in a page or two of the report).

Stephanie Weirich and I indeed considered doing such a thing, which conversation was inspired by my joining the Haskell Prime committee. We concluded that this would indeed be a research question that is, as yet, unanswered in the literature. The best we could come up with based on current knowledge would be to require type signatures on:

1. The scrutinee
2. Every case branch
3. The case as a whole

With all of these type signatures in place, GADT type inference is indeed straightforward. As a strawman, I would be open to standardizing GADTs with these requirements in place and the caveat that implementations are welcome to require fewer signatures. Even better would be to do this research and come up with a good answer. I may very well be open to doing such a research project, but not for at least a year. (My research agenda for the next year seems fairly solid at this point.) If someone else wants to spearhead and wants advice / a sounding board / a less active co-author, I may be willing to join.

Richard