As a general meta-observation, the whole point of the "let's throw all
the pieces in the air" discussions is that often, when you break some
existing assumptions, you can reassemble the pieces in a lower energy
state -- but you usually can't get there in one move. So usually in the
middle of that process, you find yourself transiting through states
which may be more mathematically attractive but less syntactically
attractive; the key is to sit on your aesthetic reaction and realize
that this may well be an intermediate state.
There was a lot of pushback to the "User model stacking" thread
(including in the -comments postings), on the basis of "these names
suck" or "there are too many knobs". But its unlikely we get to the
right stacking without first going through a less attractive, but more
general stacking.
The inversion below might not be part of the final answer, but let's let
the process play out, I think we're making progress.
On 5/4/2022 3:36 PM, Brian Goetz wrote:
Pulling farther, there's a bucket-inversion we might be able to pull
here, just by moving some terminology around:
class B1 { } // ref only
value class B3 { } // ref and val projections
value-based class B2 { } // ref only
