On Thu, Aug 20, 2020 at 07:41:54PM +0000, Tamar Christina wrote:
> While I would agree that it's fundamentally more restrictive than an
> object based one I wouldn't say it's useless. At the very least it gives
> us something to build on later.
It is IMHO useless and has also very undesirable direction on the
implementation.
Because it is all or nothing for some type, there is very strong desire that it
does at
least something and thus deal with whatever prevents the optimization on the
testcases it is tried with, which results e.g. in the sizeof proposal etc.
While when the optimization is not all or nothing, but object based,
optimize what you can, it is much easier to do the optimization
conservatively, don't try to push it too much at the risks of misoptimizing
valid code. Start by optimizing only easy cases and over time extend it to
handle perhaps more, but always punt if there is something where it can't be
proven to be safe or encounters something the code isn't prepared to handle.
When it is per object, the price of giving up on a few objects isn't that
high.
Jakub
