=). I wish I had your conviction. Would have saved us a lot of trouble.
Sure, we don’t owe users anything. We’ve just been on the conservative side of changing behavior. Perhaps too conservative. Perhaps not.
I just timed building a matrix test (inv_wishart_log_test) within prim with the prim header and the rev header. Ran each 5 times: prim 8.9s (0.23s sd) vs rev 10.6s (0.46s sd). There’s clearly some difference, but not sure it matters enough. So I’m happy if we collapse that into rev.