Re: [boost] [qvm] deduce_xx traits wouldn't introduce ODR issues.

Subject: Re: [boost] [qvm] deduce_xx traits wouldn't introduce ODR issues.
From: Emil Dotchevski (emildotchevski_at_[hidden])
Date: 2015-12-23 18:10:13

On Wed, Dec 23, 2015 at 2:44 PM, Adam Wulkiewicz <adam.wulkiewicz_at_[hidden]>
wrote:

> Emil Dotchevski wrote:
>
>> On Wed, Dec 23, 2015 at 6:24 AM, Adam Wulkiewicz <
>> adam.wulkiewicz_at_[hidden]>
>> wrote:
>>
>> Vicente J. Botet Escriba wrote:
>>>
>>> Le 10/12/2015 08:30, Emil Dotchevski a écrit :
>>>>
>>>> On Wed, Dec 9, 2015 at 10:36 PM, Vicente J. Botet Escriba <
>>>>> vicente.botet_at_[hidden]> wrote:
>>>>>
>>>>> Le 10/12/2015 00:15, Emil Dotchevski a écrit :
>>>>>
>>>>>> On Wed, Dec 9, 2015 at 2:51 PM, Vicente J. Botet Escriba <
>>>>>>
>>>>>>> vicente.botet_at_[hidden]> wrote:
>>>>>>>
>>>>>>> The problem appear when two such programmers do it (think of 3pp).
>>>>>>>
>>>>>> If within one program two programmers introduce two different
>>>>>> definitions
>>>>>>
>>>>>> under the same name, yes you have ODR but that is hardly a library
>>>>> problem.
>>>>> :)
>>>>>
>>>>>
>>>>> Hmm. This is a library issue, because the library provides an interface
>>>>>
>>>> that allows to configure types that don't know each other by two
>>>> different
>>>> parts.
>>>>
>>>> TBoost.Conversion was rejected exactly by this reason.
>>>>
>>>> This indeed is an important point, unfortunately I cannot take it into
>>> account because it wasn't mentioned in any review.
>>> Vincente, do you plan to write one?
>>>
>>> If the problem is that QVM allows you to specialize a type template, and
>> that it is possible for two programmers to introduce two different
>> definitions for the same specialization, how is that a library issue? If
>> that's a problem then we can't allow any templates in any library
>> interface, which seems odd.
>>
>
> Basically I think we should clarify this. AFAIU the problem might occur if
> there were e.g. two libraries (abc and def) providing their own matrix
> representations (e.g. abc::mat4x4 and def::mat4x4). Then we'd have two
> other libraries (X and Y) both using those two matrix types together with
> QVM. So both X and Y libraries would have to specialize QVM traits for
> abc::mat4x4 and def::mat4x4.

They don't *have* to, the deduce_m2 template does have a default definition.

> Then, if both X and Y libraries were included in a program there would be
> problems since the specializations of QVM traits would be duplicated.

Yes, however that is true for any specialization of any template any
library defines. It makes no sense to say that there is a problem with a
library because the *user* might violate ODR.

Emil