Joe Groff

@jckarter @schwa @cocoawithlove "rewrite" is a bit harsh, I think refactoring with an eye toward simplifying core concepts is more like it

@jckarter @schwa @cocoawithlove Once some representational issues are fixed it will be easier to add heuristics for more exponential cases

@slava_pestov @jckarter @schwa @cocoawithlove I'm skeptical of a rewrite of that type checker. But there are algorithmic wins to be had

@dgregor79 @slava_pestov @jckarter Maybe I missed this conversation earlier, but do y’all think @cocoawithlove’s idea would work?

@jtbandes @dgregor79 @slava_pestov @cocoawithlove so it's hard to make changes with confidence about what it will or won't break.

@jckarter @jtbandes @dgregor79 @cocoawithlove As I understand it there are improvements that can be made without changing behavior too

@jckarter @schwa @cocoawithlove Is there going to eventually be a proper design for global associated type witness inference?

@austinzheng @jckarter @schwa @cocoawithlove global inference of type witnesses requires architectural changes to work properly.

@dgregor79 @austinzheng Would that fall under finishing the iterative decl checker or is there more to it than that?

@slava_pestov @austinzheng that's a prerequisite. But you actually need to solve for (potentially) all of the type witnesses simultaneously

@dgregor79 @austinzheng Across multiple conformances or just one?