build mantid with clang

[Owen Arnold]

clang is a new compiler that aims to provide very user friendly outputs, is good for template meta programming and is claimed to be faster (according to some reports) than gcc both in terms of build times and the runtime execution of the output binaries.

Aside from acting as a good development tool (seems to catch more & different programming errors than gcc). It might be useful to run as our target compiler on the mac, because it has good support for c++11. At the moment, we are severely limited in the number of modern c++ features we can use in Mantid because we are bounded by our most historical compilers.

Explore the possibilities of building mantid using a recent version of llvm clang.

[Owen Arnold]

refs #5679. Can build and run minimal mantid plot

Changeset: 340fde75a089eb69499c1e17f1937d482d51a34e

[Russell Taylor]

Definitely worth looking into as a developer tool but unfortunately it doesn't have OpenMP, which is why it was rejected when Intel was selected as the current compiler on the Mac (see http://www.mantidproject.org/Mac_OS_X_64_bit_build#Alternative_compilers_considered here).

(Not sure why you needed to reintroduce spaces between nested templates - the c++11 way should be supported by clang since 2.9 and it worked fine for me.)

[Owen Arnold]

refs #5679. minimal build and run unit tests.

Changeset: 5b6bffb0ce982752268a9c743a45fc4c19bac940

[Owen Arnold]

Built with MakeVates switched on, and ran the VatesAPITests. That all works too.

[Owen Arnold]

Replying to Russell Taylor:

Definitely worth looking into as a developer tool but unfortunately it doesn't have OpenMP, which is why it was rejected when Intel was selected as the current compiler on the Mac (see http://www.mantidproject.org/Mac_OS_X_64_bit_build#Alternative_compilers_considered here).

(Not sure why you needed to reintroduce spaces between nested templates - the c++11 way should be supported by clang since 2.9 and it worked fine for me.)

How exactly did you set -std=c++0x in cmake for the clang compiler. When I try and switch it on it starts complaining about the boost::shared_pointer constructors? I've seen a patch associated with the move constructors ​https://svn.boost.org/trac/boost/changeset/73202.

[Nick Draper]

http://www.mantidproject.org/Release_Notes_2_2_Full_List

[Owen Arnold]

I had this building and running all the tests. Only test that fails is the CurveFittingTest_IkedaCarpenterPVTest test, which is probably due to a precision issue.

I'm going to stop the investigations at this point, because we need a more mature version of clang that is available at the moment to make it feasible for mantid production.

[Nick Draper]

until clang supports openmp we are unlikely to move to it, unless another tool can be used via the multithreaded wrapper.

[Stuart Campbell]

This ticket has been transferred to github issue ​6525