[CMake] How is GetPreRequisites used?

[Nicolas Desprès]

On linux I ship all .so files (with few exception) listed by ldd in a
given directory. I also ship a wrapper script that add this directory
to LD_LIBRARY_PATH prior to execute the executable.

Here the script:

#!/bin/sh

# This script overload the dynamic library path.

# Resolve links referring to me.
if test -L "$0"; then
    exec $(readlink -f "$0")
fi

appname=$(basename $0 | sed s,\.sh$,,)
dirname=$(dirname $0)
case "$dirname" in
    /*)
	# nothing to do
	;;
    *)
	dirname="$PWD/$dirname"
	;;
esac
paths="$dirname/../@EXTRA_LIBRARIES_DIR@"
LD_LIBRARY_PATH="$paths"
export LD_LIBRARY_PATH
DYLD_LIBRARY_PATH="$paths"
export DYLD_LIBRARY_PATH
"$dirname/$appname" $*

and here the list of .so that I do not ship:

  set(DLDEP_UNIX_EXCLUSION_LIST
    "^libc.so"
    "^libdl.so"
    "^libpthread.*"     # Fedora 9
    )

Note that, when you ship binary package for Linux you have to compile
your project on a system with the oldest GLIBC you want to support.
For instance package built on Ubuntu (glibc 2.4) won't work on Debian
stable (glibc 2.0). I build my package on Debian and I have tested it
on various freshly installed Linux distribution and it works fine.

Cheers,

On Tue, Oct 28, 2008 at 9:47 PM, David Cole <david.cole at kitware.com> wrote:
> You are not being stupid. It does need to be used at installation time to be
> useful.
> The intent is to analyze a given built executable to determine the required
> set of (possibly 3rd-party, possibly built elsewhere on the build machine)
> libraries that it needs to run. Then, given that set, you need to make sure
> that you have install rules to cover all of those libraries.
>
> The first working example of its use is in the ParaView code base where it
> is used to copy libraries into ParaView's Mac bundle application. It is used
> in conjunction with BundleUtilities.cmake to "fix up" the bundle app so that
> it is standalone and contains all of its referenced libraries in the bundle
> directory structure. Then the bundle can be copied to any directory or to
> another Mac and it should all "just work."
> Having said that, there is more work to be done to make this a genuinely
> easy to use feature of CMake. I envision being able to do the same thing at
> make install time for Linux and Win32 builds, but according to the
> conventions of those platforms. Dlls copied into the same dir with the exe
> on Win32 and "whatever's appropriate" on Linux builds... (Feel free to make
> suggestions.)
> So... ideally it would all just work automatically and all you would have to
> do in CMake is add an install rule for your executable and call a
> post-install script that copies/fixes up libraries appropriately in a
> platform-independent manner.
> Thanks for asking questions / contributing to the discussion / improving
> this development feature of CMake.....
>   David
>
> On Tue, Oct 28, 2008 at 7:36 AM, Mike Arthur <mike at mikearthur.co.uk> wrote:
>>
>> Perhaps I'm just being stupid but I can't see how this can be used as it
>> needs
>> to be called at installation time and I'm not aware that modules can be?
>>
>> Does anyone have a working example of how to install its outputted
>> libraries?
>> --
>> Cheers,
>> Mike Arthur
>> http://mikearthur.co.uk/
>> _______________________________________________
>> CMake mailing list
>> CMake at cmake.org
>> http://www.cmake.org/mailman/listinfo/cmake
>
>
> _______________________________________________
> CMake mailing list
> CMake at cmake.org
> http://www.cmake.org/mailman/listinfo/cmake
>



-- 
Nicolas Desprès