[CMake] Linking problem

[Renaud Detry]

On 07 Nov 2007, at 11:35, Salvatore Iovene wrote:

> On 11/7/07, Renaud Detry <renaudjdetry at airpost.net> wrote:
>>
>> On 07 Nov 2007, at 11:04, Salvatore Iovene wrote:
>>
>>> On 11/7/07, Renaud Detry <renaudjdetry at airpost.net> wrote:
>>>>
>>>> On 07 Nov 2007, at 09:59, Salvatore Iovene wrote:
>>>>
>>>>> On 11/7/07, Brandon Van Every <bvanevery at gmail.com> wrote:
>>>>>> On Nov 7, 2007 2:12 AM, Salvatore Iovene
>>>>>> <salvatore.iovene+cmake at googlemail.com> wrote:
>>>>>>> Hi,
>>>>>>> I'm not 100% sure this is really a CMake related question, but
>>>>>>> I'll
>>>>>>> fire it up anyway:
>>>>>>>
>>>>>>> I'm building a series of static libraries, name them liba.a,
>>>>>>> libb.a
>>>>>>> and libc.a, and linking them into a shared library libfoo.so.
>>>>>>>
>>>>>>> Then I'm building libx.a liby.a and libz.a and linking them
>>>>>>> into the
>>>>>>> shared libbar.so.
>>>>>>>
>>>>>>> Then I have an executable whatever.exe that's linked to to
>>>>>>> libfoo.so
>>>>>>> and libbar.so. The linking of the executable fails  
>>>>>>> complaining of
>>>>>>> certain missing simbols. Some symbols from liba.a are missing in
>>>>>>> libbar.so.
>>>>
>>>> Let's say that there's a symbol S in liba.a, that S is the only
>>>> symbol
>>>> in its object, and that S is not used by libfoo. In Darwin, S will
>>>> not
>>>> be included in libfoo, and the result you get is expectable. I  
>>>> don't
>>>> know if this simple explanation translates to Linux.
>>>
>>> I don't know either, but that seems to be the case. Any ideas on
>>> how to fix it?
>>
>> IMHO, you shouldn't use symbols from liba in libbar if libbar doesn't
>> link against liba.
>>
>> I think you could do either of
>>
>> 1.- make all your static libs dynamic, and link libbar against  
>> liba if
>>      libbar uses symbols from liba.
>>
>> 2.- link neither libfoo nor libbar to liba, but link whatever.exe to
>>      liba. This will require an extra flag about undefined symbols
>>      under Darwin, but I don't think ld will complain under Linux.
>>
>> Option (1) seems far nicer to me.
>>
>> Hope this helps.
>> Renaud.
>
> Thanks. Your options seem very reasonable. The thing is, though, that
> the various liba, libb, libc were meant to be only used to build a
> bigger library called libfoo.

Then why don't you build all objects from foo, a, b, c to be PIC and
pack them all together in foo? Or if you really want intermediate
archives, maybe you can build the objects of a, b, c to be PIC,
archive them in .a archives, and use ld to put them together. But this
is not the same thing as linking foo to a, b, c using gcc and -la -lb
-lc, which means to only fetch needed symbols.

The bottom line is, what you mean to do is probably not very common,
and I don't think there will be a high-level CMake primitive that will
do that for you.

This is just my opinion though :-)

Renaud.