One of my pet peeves I've had about dmd implementation since I worked on the array append code is how dmd reuses the array allocation routines to allocate structs on the heap.

I think Martin Nowak found a rather compelling case for why we shouldn't do it that way.  If you look at the phobos pull request https://github.com/D-Programming-Language/phobos/pull/556, he's actually suggesting working around it.  And I think I would rather see the compiler druntime fixed to do the "right thing" than have to tell people, "oh, don't use new X, it's inefficient.  Instead use these 3 lines of confusing code".

http://d.puremagic.com/issues/show_bug.cgi?id=7243

This isn't a regression/blocker, but it's still pretty early in the release cycle, so can we get this fixed in this cycle?  As I said in the report, I'll fix druntime (not today, though :).

-Steve