On Fri, Jul 4, 2014 at 3:09 PM, David Moon <[email protected]> wrote:
> Maybe someone should do a PhD thesis on it. This is essentially what Jeff's PhD thesis is on. You're absolutely right that it's currently pretty much up to what the implementation does right now. The trouble with making up rules for such things before anyone has really used them is that rules that people make up without any experience tend be unusable. Instead, we've taken the approach of winging it until we've got enough experience to make up well-informed rules based on real-world experience. At some point (soon) we'll have formal rules for such things and that will be a big part of Julia 1.0. But it won't massively break from what we're doing now because what we're doing seems to mostly work even if we can't explain it other than what the code implements.
