Re: Implementing p0515 - spaceship operator

[David Brown <david at westcontrol dot com>]

On 11/01/18 12:32, Jonathan Wakely wrote:
> On 11 January 2018 at 11:28, David Brown wrote:
>> On 11/01/18 11:13, Jonathan Wakely wrote:
>>> t all cleaOn 11 January 2018 at 10:05, David Brown wrote:
>>>> Maybe it is easier to say "gcc supports <=> in C++2a, and as an
>>>> extension also supports it in C and C++ of any standard" ?  I don't
>>>> believe there is any way for it to conflict with existing valid code, so
>>>> it would do no harm as a gcc extension like that - and C users can then
>>>> use it too.
>>>
>>> It's not very useful in C because you need the comparison category
>>> types, which are classes defined in <compare> (see
>>> http://en.cppreference.com/w/cpp/header/compare)
>>>
>>> C doesn't have those types, and can't define anything close.
>>>
>>
>> Surely you can get very close by simply returning an int, value -1, 0 or
>> 1?  That is what other languages (like PHP) do for their <=> operator.
>> There are complications - such as for floating point when you have NaNs.
>>  But I think you could have a very successful operator if you defined "a
>> <=> b" to be the same as "(a > b) - (a < b)".
>>
>> Whether it would be particularly /useful/ or not is another matter.  I
>> was thinking mainly in terms of saving effort when making C++2a support
>> - rather than having to making the new operator conditional on a
>> particular version of the standards, it could be accepted in any version.
> 
> It seems like the wrong trade-off. We have dozens of features that
> depend on the standards mode, we know how to do that.
> 
> Simply slamming a new language feature in all modes (*and* in a
> different language with a different front-end!) to save a small amount
> of effort doesn't seem like a good idea to me. Especially if it then
> doesn't even work the same pre-C++2a, as know we need to test it in
> additional ways.
> 

Fair enough.  I was merely thinking it might make the development a
little easier by avoiding more feature testing in the front ends.  And
gcc has a history of making some features of one language or standard
available in others as gcc extensions.  (Sometimes I find these useful
in my coding - though I can't think off-hand for a use of <=> in my own
programs.)

But I see pre-C++2a would be a problem here.  You can't make it match
the C++2a version, because you don't have the types available - and you
wouldn't want to make it have different types that are incompatible with
the C++2a version.  It would be fine with a -1, 0, 1 int result in C -
but not in pre-C++2a.