That was certainly one of the counterarguments (& global variables also use
a type rather than a size). I've not really settled on which way to
go/haven't given it lots of thought. I may loop back around to the original
thread when it comes to that.
On Fri, Jun 10, 2016 at 2:18 PM, James Y Knight
Yes, it was intended to -- at least for bitcode produced by clang.
I do think it would be a good idea to continue to pass the value type to
byval, though...Either that or get rid of the type in the "alloca"
instruction. They're basically doing the same thing, and having them
specified completely
Excuse the necromancy, but do you know if this change (or other work you
did in this area) completely eclipsed LLVM's use of inferred alignment via
the llvm struct's alignment for byval arguments?
I ask because this was something I was going to need to fix for the
typeless pointer work & I have
