On 3/30/2012 12:11 PM, Steven Schveighoffer wrote:
On Fri, 30 Mar 2012 14:27:43 -0400, Walter Bright <[email protected]>
wrote:
I would argue that:
3. An extension method for an argument of type template parameter T will be
looked up only in the instantiation scope.
I don't think you looked at my counter case in detail. Your idea leads to two
different instantiations of tmpl!Foo having two different implementations,
depending on which extension methods you include. In fact, in one place, the
instantiation might succeed, but in another, the instantiation might fail.
Yes, you're right. I missed that nuance. I don't really know how to fix it.
Given that the compiler simply trims out identically defined symbols, assuming
they are the same, this could have disastrous consequences. You could end up
with code being different based on the link order, or possibly arbitrary
decisions made by the linker!
My argument is that the one thing in common is that the module that defines the
type has to have been included. I think you would agree that the one truth we
must maintain is that X!Y must be implemented exactly the same no matter what
module instantiates it.
I realize that the compiler likely doesn't keep enough information to know what
functions would fall under the rules I specified, so this is probably a
difficult change to effect. But it could be an incremental change later.
-Steve
