On Wed, May 4, 2022 at 12:53 PM 'Axel Wagner' via golang-nuts
<golang-nuts@googlegroups.com> wrote:
>
> On Wed, May 4, 2022 at 9:46 PM Ian Lance Taylor <i...@golang.org> wrote:
>>
>> Zero sized values are useful in Go, because they can have methods and
>> they can be stored in interfaces. But if the addresses of zero-sized
>> values must be distinct, then although zero-sized values appear to
>> have zero size they must in fact be implemented as taking up one byte.
>> For example, given `struct { a, b struct{} }`, the addresses of `a`
>> and `b` must be distinct, so that struct is actually two bytes. So,
>> sure, we could change it. But there are surprising results either
>> way.
>>
>> (Historically this was introduced for https://go.dev/issue/2620.)
>
>
> The other side of the question is "why not define them to always be the
> same"? It seems a fairly simple optimization.
> I guess it can be argued that `a[0:0:0]` points to a zero-sized object, so
> should always be the same, but that's expensive to do, as every slicing
> operation has to check for 0? But that seems a weak argument.
> Is there a better one?
Using the reflect package we can walk through the fields of a struct
and get the address based on the size and alignment of earlier fields
in the struct. But if the address of all zero-sized values is the
same, then that doesn't work.
It's not a great argument, but it's not nothing.
Also it's a little weird that given `var v1 v2 struct { f1 int; f2
struct{}; f3 int }` then &v1.f2 == &v2.f2, but maybe that's not much
weirder than how it works today.
Ian
--
You received this message because you are subscribed to the Google Groups
"golang-nuts" group.
To unsubscribe from this group and stop receiving emails from it, send an email
to golang-nuts+unsubscr...@googlegroups.com.
To view this discussion on the web visit
https://groups.google.com/d/msgid/golang-nuts/CAOyqgcXBQ14JMzAJbN9NZ%2BJeMYPRH9Za0G7iJDQgQ4eUGqLvBQ%40mail.gmail.com.
