Re: [boost] [test] dynamic linking on Windows

From: Gennadiy Rozental (gennadiy.rozental_at_[hidden])
Date: 2006-11-10 16:23:55

"Vladimir Prus" <ghost_at_[hidden]> wrote in message 
news:ej2f2f$p32$1_at_sea.gmane.org...

>>> Yes, but I mean that it's possible to have:
>>>
>>>    - UTF, that is build either as static lib or DLL and does *not*
>>>      have any main function.
>>>    - Small static library that contains stock definition of main.
>>
>> No. This is not going to be acceptable IMO.
>
> Why? Is there some problem I've overlooked?

It's unacceptable for Boost.Test users outside of Boost.

>>> So, when using UTF DLL, user must "just" link to an extra library, which
>>> might be simpler than defining macroses.
>>
>> Why? How is it easier?
>
> Adding one library to link link is easier that modifying sources for a 
> every
> test program.

There still a change that needs to be done in either Jamfile or source file

>>> And importantly, Boost.Build V2
>>> can make this transparent for all Boost.Test users inside Boost.
>>
>> Not everybody are using Boost.Build.
>
> I was talking about Boost.Test users inside Boost; of them all use
> Boost.Build.

Yes. But Boost.Test is used outside of Boost either

>> IMO the issue is not that significant. In a long term Boost.Test users
>> should switch to auto generated test tree and this case is easily
>> supported in both builds. Conversion in 95% of the cases is
>> straightforward and simple to do.
>
> Sure, but I if nobody does nothing, then in a year from now most tests in
> Boost will still be linking to Boost.Test statically.

 My point is that we couldn't make this change in general. That doesn't mean 
that we couldn't introduce something boost internals specific.

Gennadiy