On Thu, 16 May 2019 at 23:28, Jonathan Wakely wrote:
> Here's what I've tested and am about to commit.
Looks good to me.
On 16/05/19 12:43 +0100, Jonathan Wakely wrote:
On 16/05/19 12:29 +0100, Jonathan Wakely wrote:
These two changes both result in smaller code for std::variant.
The first one means smaller tables of function pointers, because we
don't generate an instantiation for the valueless state. Instead we
On 16/05/19 12:29 +0100, Jonathan Wakely wrote:
These two changes both result in smaller code for std::variant.
The first one means smaller tables of function pointers, because we
don't generate an instantiation for the valueless state. Instead we do
a runtime branch, marked [[unlikely]] to make
These two changes both result in smaller code for std::variant.
The first one means smaller tables of function pointers, because we
don't generate an instantiation for the valueless state. Instead we do
a runtime branch, marked [[unlikely]] to make _M_reset() a no-op if
it's already valueless. In
