[CMake] Problem with set_target_properties on mixed-source library

[Martin Apel]

Alan W. Irwin wrote:
> On 2008-09-29 16:00+0200 Martin Apel wrote:
>
>> Hi all,
>>
>> I have a library consisting of multiple sources from Fortran and C, e.g.
>> file1.f, file2.f, file3.c.
>> I want to generate this library in two different versions, where in one
>> of those versions the Fortran files are compiled with different options,
>> in my case with -ipo.
>> That means:
>>    libv1.a   = file1.o file2.o file3.o               compiled without
>> ipo-options
>>    libv2.a   = file1.o file2.o with ipo-option, file3.o without
>> ipo-option
>>
>> I tried to use set_target_properties on libv2 to set the ipo-flag, but
>> this also sets this option for the C files, which the compiler does not
>> understand.
>> I also cannot use set_source_properties, because the same source files
>> will be compiled with different options.
>>
>
>
Hi Alan,

> The scenario I am thinking about is the one where you set source
> properties
> with non-ipo flag, build your non-ipo library, then set the source
> properties _again_ using ipo flag, and build your ipo library. Setting
> a new
> compilation flag using set_source_properties should (in an ideal
> world) mean
> that the source gets recompiled, but that is just my opinion and the
> CMake
> developers may be up against practical constraints that don't allow
> this or
> they may have intended recompilation to take place whenever a compile
> flag
> is changed, but there may be a bug in the implementation for Fortran
> or in
> general.
I think this does not work. I have tried setting and resetting some
flags in the same CMakeLists.txt before
and CMake always uses the value of variables in effect, when the end of
file is reached.
>
> To further the discussion and be absolutely sure about the behaviour
> you are
> reporting, I suggest you make the simplest-possible self-contained
> example
> of the issue consisting of a CMakeLists.txt file and two "hello
> world1" and
> "hello-world2" source files where the first one (only) should have
> different
> compiler flags for two distinct libraries that you build. Also, to
> rule out
> a Fortran specific bug, the two different files should be C.  Then you
> should be able to show (using make VERBOSE=1) exactly what flags are used
> for each build and indeed whether the object is rebuilt (like I think it
> should be in an ideal world) because of the changed compile flags set
> with
> source properties.
The problem will not show up, if I only use C files. The problem is,
that you cannot distinguish
between Fortran flags and C flags when using
  SET_TARGET_PROPERTIES (v2 PROPERTIES COMPILE_FLAGS "-ipo")
Those flags will be applied to both C and Fortran sources, if the target
contains both.