[cmake-developers] [PATCH][CPack][RPM] multiple path relocation prefixes

Domen Vrankar domen.vrankar at gmail.com
Fri Jan 16 17:06:46 EST 2015

Resending as I accidentally dropped the developers mailing list...

2015-01-16 22:38 GMT+01:00 Brad King <brad.king at kitware.com>:
> On 01/16/2015 08:42 AM, Brad King wrote:
>> Go ahead and merge to 'next' for testing and follow up with any
>> dashboard trouble.  I'll still be able to review it there.
> AFAICT the change to GNUInstallDirs is just to help projects use
> the new feature, and not actually part of the new feature itself.
> The feature itself just happens to use the same set of <dir> values.

Yes I added variables to GNUInstallDirs as a convenience method. Users
can still choose to set values without the use of GNUInstallDirs

> It feels a bit strange to have CPACK_ variables set by GNUInstallDirs,
> and to require CPACK_PACKAGING_INSTALL_PREFIX to be set before that
> module is included.
> Perhaps instead CPackRPM can provide an option to compute a default
> for CPACK_PACKAGING_INSTALL_FULL_<dir> based on the GNUInstallDirs
> values.  This way GNUInstallDirs does not have to change at all,
> and users can choose to include GNUInstallDirs and still set their
> own CPACK_PACKAGING_INSTALL_FULL_<dir> values separately.

I tried to do that first but the problem is that only CPACK_*
variables are propagated to CPackRPM script and I thought that
redefining the variables with new names in GNUInstallDirs is better
than creating a new module just for that.

I could write a new module for converting CMAKE_INSTALL* variables to
CPACK_PACKAGING_INSTALL* variables and then generate
dir> inside CPackRPM.

Alternative that comes to mind is that I just remove the convenience
method of setting variables through GNUInstallDirs.

I'm opened to suggestions.

> Typically I quote the second argument in this case.  It is not
> strictly necessary but looks cleaner IMO.

I agree.

> Please look at adding test cases, perhaps in Tests/RunCMake,
> to check the warning message about non-relocatable paths.

Haven't tried to write negative tests or parsing of command line
during test execution. Are there any good examples of such tests?

> Thanks,
> -Brad

More information about the cmake-developers mailing list