Didier Remy

> * I would prefer if the `[@@unboxed]` attribute was per-constructor rather than global. (This was also hinted at by @garrigue.) The meaning of each `[@@unboxed]` is that this constructor...

On 11/11/2019 22:39, Jacques Garrigue wrote: > Thanks for the suggestion. > Another idea is |[@@logical]|, to emphasize that quantifiers follow the > laws of logic. But we now also...

> Discussion > > * The need for a special handling of ground types makes me think that the > propagation approach is only a hack: having strong type variables...

> * As far as references are concerned, the distinctive property of strong/safe type variables is that values at their types cannot "escape" to the ambient execution context through the...

> Your description is simpler, but: I should have said somethinge even simpler: - A type variable is weak when it is used in the type if a subexpression whose...

Yes, indeed, I thought about nonexpansive as well, but it's a bit longer with a negative connotation and (I think, but I am not sure) this originally refered to expressions...

Why not---that may be the best of the proposals so far. "generalize 'a" itself is not quite the intuition, since 'a is generalized anyway as it is the type of...

I am not against changing the name: this may be a good idea---or not. And it is a good time to question this and choose the best name. I am...

> Scala uses _path-dependent type_ for an analogous feature. I think a similar name would work well in OCaml, too. That sounds a good name, perhaps even clearer (or less...

The counter example ``` module F (type a) = struct let r : a option ref = ref None end ``` is indeed a counter example if the compilation would...