Chandler, would you be able to evaluate this (how severe it is,
whether the patch is good, etc)? We are switching to cmake, but will
probably need to support autotools as a legacy build script for a
while. If it's a big pain for the users, we should fix it. Thanks,
--
Zhanyong
On Wed, Jun 9, 2010 at 11:10 AM, Thierry GAYET <thierr...@gmail.com> wrote:
> hi,
>
> I am currently use it and it work fine so if you need something more about
> don't hesitate to ask to me.
>
> Can u also telle me if you will accept it or not. thus i will follow the
> original release instead of my fixed one instead.
>
> Regards
>
> Thierry GAYET
>
> Zhanyong Wan (λx.x x) a écrit :
--
Zhanyong
C++ is mostly a superset of C, and can link with C code. Therefore
you can use gtest to test C code.
Thanks for contributing, Thierry!
Thanks, Vlad. I didn't know that pkg-config is for installed gtest. :)
I'd like to restate that we do *not* intend to support installing
gtest into system directories. It just lead to big pain down the road
too easily.
--
Zhanyong
On Wed, Jun 9, 2010 at 12:40 PM, Vlad Losev <vl...@google.com> wrote:Thanks, Vlad. I didn't know that pkg-config is for installed gtest. :)
> Zhanyong --
>
> In light of the decision not to support installing Google Test into system
> directories, accepting this patch makes little sense. Of course, it is
> possible to reconsider that decision, given how many people seem to be using
> the feature, even with all its drawbacks. ;-)
I'd like to restate that we do *not* intend to support installing
gtest into system directories. It just lead to big pain down the road
too easily.
Adding back the mailing list and Chandler...
hi,
For your information the pkg-config is today really well integrated to the autotools and not dependant to a specific distribution.
Thus, this is possible to use some commands:
pkg-config --exist --print-errors gtest : test if the libgest is installed
pkg-config --cflags gtest : give all the cflags for the compilation step
pkg-config --libs gtest : give all the ldflags for the link step
Then in another autotools files (configure.ac) this is possible to do :
PKG_CHECK_MODULES([GTEST],[gtest],[have_libgtest=yes],[have_libgtest=no])
if test "$have_libgtest" = no ; then
AC_MSG_ERROR([Missing libgtest library!!])
fi
Easy and very usefull isn't it thanks to the buildin m4 macro. Today linux distributions are using .pc files as a standard through autotools projects.
What do you think about it ? Still opposed on the integration of my patch ?
hi,
Okay for your reply.
I am working for a big industrial that develop some software on specific hardware.
The google test have been selected among a big list available for free and according for their features.
We are using a Continuous Integration server with Hudson and the google test package is used as well as xunit for unit and no-regression test.
We are working in a cross-compilation environment through Buildroot that generate root filesystem for our embedded system.
Thus this is not possible to include it with each system and it was easier to install it on our system and 'cos all our packages are autotools based, its was needed to have a pkg-config support through a .pc file.
Is this view interesting for you ?
Hi,I'm not sure this is the same problem, but I actually do have a need to 'install' gtest/gmock. Not necessarily with autotools though....We maintain a thirdparty binary repository to avoid recompiling code that rarely changes. The compilation options are guaranteed to be consistent with the 'user' code. When deploying a new version of gtest, I used to be able to 'make install'. Now I have to manually copy the binaries and includes into their respective directories which is error-prone. I agree that installing into system folders is not very safe, but what about this particular case?
That being said, would it be possible to add an 'INSTALL' command to the CMake script?
On Thu, Jun 10, 2010 at 1:29 AM, Hady Zalek <hady....@gmail.com> wrote:Hi,I'm not sure this is the same problem, but I actually do have a need to 'install' gtest/gmock. Not necessarily with autotools though....We maintain a thirdparty binary repository to avoid recompiling code that rarely changes. The compilation options are guaranteed to be consistent with the 'user' code. When deploying a new version of gtest, I used to be able to 'make install'. Now I have to manually copy the binaries and includes into their respective directories which is error-prone. I agree that installing into system folders is not very safe, but what about this particular case?This is a more common case I suspect.
That being said, would it be possible to add an 'INSTALL' command to the CMake script?I wouldn't be hugely opposed to some command to the CMake build system that would "bundle" the build outputs and the headers into a dumb tree (ie, a flat set of ./{bin,lib,include}/...) perhaps with facilities to build tarballs of it. Focus it on 1) static library linking, and 2) assuming little-to-nothing about the underlying OS. This becomes more of a binary distribution tool than an installation tool.All this said, does it really merit this complexity? the includes are trivial -- it's already a separate tree. The libraries should consist of only 2 files. It seems like overkill at the end of the day.
On Thu, Jun 10, 2010 at 1:29 AM, Hady Zalek <hady....@gmail.com> wrote:Hi,I'm not sure this is the same problem, but I actually do have a need to 'install' gtest/gmock. Not necessarily with autotools though....We maintain a thirdparty binary repository to avoid recompiling code that rarely changes. The compilation options are guaranteed to be consistent with the 'user' code. When deploying a new version of gtest, I used to be able to 'make install'. Now I have to manually copy the binaries and includes into their respective directories which is error-prone. I agree that installing into system folders is not very safe, but what about this particular case?This is a more common case I suspect.That being said, would it be possible to add an 'INSTALL' command to the CMake script?I wouldn't be hugely opposed to some command to the CMake build system that would "bundle" the build outputs and the headers into a dumb tree (ie, a flat set of ./{bin,lib,include}/...) perhaps with facilities to build tarballs of it. Focus it on 1) static library linking, and 2) assuming little-to-nothing about the underlying OS. This becomes more of a binary distribution tool than an installation tool.
All this said, does it really merit this complexity? the includes are trivial -- it's already a separate tree. The libraries should consist of only 2 files. It seems like overkill at the end of the day.