specialized templates not working as expected

[andre maute]

On 2/19/20 10:15 PM, Jonathan Wakely wrote:
>> I understand that this is a template instantiation problem.
>> But I'm wondering why there is such a difference
>> in standard interpretation for such an old C++ feature.
>
> Either MSVC has a bug, or it chooses to use the end of the file as the
> point of instantiation for all templates. The latter is allowed by the
> standard.
>
>> I also understand that there are other possibilities
>> to achieve a similar effect, but I don't like to discuss
>> these possibilities here.
>
> So then what's the question?

The question was only
if this could be due to different standard interpretations.

> The code says "can this block be simplified or even removed?" and the
> answer's yes, but you said you don't want to discuss that.
This comment is a leftover from a discussion with my colleagues.
We wondered, if it would be possible to simplify this block
for given N foos to an O(1) sized block.
As shown, this was possible with MSVC.

I got suggestions to completely rewrite this minimal reproducer,
which comes from a customer project.
I don't want to discuss this here again.

> If you reorder the specializations so that A<int>::A() comes last then
> it compiles with GCC (and Clang and EDG). The problem is that
> A<int>::A() triggers the implicit instantiation of the other member
> functions, before the specializations have been seen. By reordering
> them the specializations have been defined before the instantiation
> happens.
Reordering might not be possible :-( (see above)

Kind Regards
Andre