[Development] New library dependencies

[lars.knoll at nokia.com]

On 10/31/11 2:25 PM, "ext Giovanni Bajo" <rasky at develer.com> wrote:

>Il giorno 31/ott/2011, alle ore 13:17, <lars.knoll at nokia.com>
><lars.knoll at nokia.com> ha scritto:
>
>> On 10/31/11 10:53 AM, "ext Thiago Macieira" <thiago at kde.org> wrote:
>> 
>>> On Monday, 31 de October de 2011 10:44:16 Giovanni Bajo wrote:
>>>> What is the policy on adding new dependencies to the Qt project?
>>> 
>>> I'd guess that the maintainer for that module approves. For qtbase.git,
>>> that's 
>>> Lars.
>>> 
>>> Importing third-party source code requires approval under the CLA too.
>>> 
>>>> To me, it seems a bad idea to add a dependency on any library unless
>>>> there
>>>> is a specific use case were it is really necessary. So adding Boost
>>>>just
>>>> because it's "cool" (for some definition is cool) doesn't look like a
>>>> deal.
>>>> It doesn't help that I specifically dislike Boost, but that's not the
>>>> subject of this comment.
>>> 
>>> Agreed. But I know João wouldn't do that: his complaint was that we
>>> reinvent 
>>> the wheel just so we don't add the dependency. So I agree with him that
>>> if 
>>> there is an implemented solution with no ill side effects, there's no
>>> reason 
>>> not to use it.
>> 
>> Compile/build time complexity is one reason to be careful. The other one
>> is the size of the full stack. We have to be careful and do decent
>> judgement calls here whether the benefits are worth the additional
>> dependency.
>
>Agreed. 
>
>> Supporting some C++11 features on old compilers is e.g. a case where I
>> wonder whether adding a dependency for everybody is worth it. Most
>> compilers (we care about) already support a decent subset of C++11, if
>>you
>> need other compilers simply don't use the feature.
>
>
>I'm not sure who is the "you" in the above sentence. A maintainer?  Or a
>patch submitter?

The Qt users in that case.

For Qt itself I don't believe we can currently introduce a hard dependency
onto C++11 features as that would exclude a large range of users that
still need to work with older compilers.

>
>The choice of the (subset of the) language is a different beast from
>adopting new libraries. I believe that choices like "C++11 allowed in
>source code" or "Feature X,Y,Z of C++11 allowed in source code" should be
>taken by maintainers/Lars, at least for the essentials modules, and
>published in the wiki. I can't see how else things should be done -- it's
>a guidance for any random contributor on the way code must be written,
>and it saves everybody from wasting time using a language feature that is
>not allowed because of constraints that might not be apparent.

Code written for Qt must (for the moment) still compile on C++98 compliant
compilers. But we can use C++11 to optimize things when available.

Cheers,
Lars