Implementing cmake build

[rhyolight]

Replacement for PR #645.
Fixes: #647.
Fixes: #661.

[rhyolight]

Discussion continues from #645, see that PR for previous conversations.

[oxtopus]

Is it possible to encapsulate the git submodule commands into the cmake configuration? I just tried this in a stock VM and it wasn't obvious that I needed to run git submodule init and git submodule update before cmake would run.

[rhyolight]

@oxtopus Fine by me, but if we do that, we need to add back

git:
 -  submodules: false

into the .travis.yml. Also need to take out the instructions to manually git submodule update from the README.

[david-ragazzi]

SOLVED!!!! Now Travis is building fine with GCC!!!

This helped me a lot:
https://groups.google.com/forum/#!msg/trema-dev/8DGcZ-om6Bk/7Vm-OXYsxroJ

The only explanation that I could find think of was that the linker used by Travis CI uses the --as-needed flag. This has a side-effect [1]. If all the libraries requiring libpthread are placed after it in the linker flags, libpthread will be discarded despite being needed.

By placing "-lpthread" in extconf.rb, it ended up near the beginning of the final LDFLAGS and my guess is that it was subsequently ignored by the linker for the aforementioned reason.

I included "-Wl,--no-as-needed " at begining of NTA_PLATFORM_LINKFLAGS variable for Linux platforms. Travis simply was ignoring the libraries which HtmTest and TestEveything would be linked against (-ldl -lpthread, etc ).

[subutai]

@david-ragazzi Nice work! That's really encouraging.

[david-ragazzi]

@subutai

I include these lines:

if("$ENV{NTAX_DEVELOPER_BUILD}" STREQUAL "")
  set_environment_variable(NTAX_DEVELOPER_BUILD "1" OFF)
else()
  show_environment_variable(NTAX_DEVELOPER_BUILD "$ENV{NTAX_DEVELOPER_BUILD}")
endif()

The code above set NTAX_DEVELOPER_BUILD to 1 for first time, and once set, it only shows its value.

Feel free to change its initial value..

[rhyolight]

I looked through it too. Only used for optimization in a handful of places.

---

Matt Taylor
OS Community Flag-Bearer
Numenta

On Thu, Feb 20, 2014 at 9:59 AM, Subutai Ahmad notifications@github.comwrote:

@rhyolight https://github.com/rhyolight Just looking into veclib issue.
I'm going to probably propose we get rid of that include. That is an
optimization that is only used in a couple of functions that I don't think
are really being used anyway. I will verify first though.

Reply to this email directly or view it on GitHubhttps://github.com//pull/657#issuecomment-35650024
.

[subutai]

@david-ragazzi I think it can just show it. It shouldn't set it - that is up to the developer. We also need to add the code to do the actual symlink. In the past this was handled by the script in build_system/pybuild/install.py

[subutai]

@david-ragazzi The following lines (179-182) are causing problems on some Mac's:

# Check if default C/C++ compiler is GNU or CLANG, otherwise exit with error.  
if(NOT(CMAKE_COMPILER_IS_GNUCXX OR ${CMAKE_CXX_COMPILER} MATCHES "clang"))
  message(FATAL_ERROR "${CMAKE_CXX_COMPILER} compiler not supported yet. If you have GNU or CLANG C++ compiler installed on your machine change $CXX environment variable or pass its location through USER_CXX_COMPILER option.")
endif()

The check fails even if CXX=clang++ and CC=clang. By taking it out the compilation works fine. On my Mac it works fine either way. @rhyolight suggests we take out the check and just put it in the documentation.

What do you think?

Good news is that without this check and after fixing #663, it works on Matt's and Austin's Mac's as well, so three total.

[david-ragazzi]

@subutai I think is fine. Although a Borland or VC++ will cause trouble if user don't read doc ....

I'm checking symlinks issue.. NTAX_DELELOPER_BUILD is a "boolean" value, right? If true, CMake should create symlinks, if false copy the files..

[david-ragazzi]

@subutai I wrote the code for symlinks. Please look for NTAX_DEVELOPER_BUILD word in CMake file.

I also noted that it doesn't replace a file to link and vice-versa after first configuration. Could CMake remove the entire $NTA folder before CMake starts the copy or linking?

[subutai]

@david-ragazzi Awesome! I just tried it out and it works great.

I also noted that it doesn't replace a file to link and vice-versa after first configuration. Could CMake remove the entire $NTA folder before CMake starts the copy or linking?

Hmm, that's a good question. If you remove $NTA every time you run CMake, is there a problem with that?

[rhyolight]

I usually remove and recreate the $NTA directory when I do a clean build. I think it should be okay.

[subutai]

@david-ragazzi We are ok to merge this once build passes. After that I will continue some more Grok tests, and we can start on the next phase!

[subutai]

I usually remove and recreate the $NTA directory when I do a clean build. I think it should be okay.

OK, so running CMake is the equivalent of doing a clean build? I was wondering about that. If so, it should also clean out the build directory. For incremental builds the user can just run make right?

[david-ragazzi]

OK, so running CMake is the equivalent of doing a clean build? I was wondering about that. If so, it should also clean out the build directory. For incremental builds the user can just run make right?

@subutai Yes. At CMake file we could simply put a command to delete $NTA (future build/release) directory to accomodate new changes in a clean folder. This would save a lot of code, mainly in symlinks issues!

[subutai]

@david-ragazzi Great, thanks. That is all very clean and simple.

[subutai]

@david-ragazzi Really nice job on this PR! I have a question. It looks like all the CPP/HPP files are explicitly listed in the CMakeLists.txt file. Is that necessary or can we eventually just list directories and it can automatically pick up the files? Currently each directory has a unittests sub-directory, so perhaps we have to wait until we remove those?

[scottpurdy]

I just cleaned everything out and followed the README instructions and it worked well. Great work!

[david-ragazzi]

It looks like all the CPP/HPP files are explicitly listed in the CMakeLists.txt file. Is that necessary or can we eventually just list directories and it can automatically pick up the files? Currently each directory has a unittests sub-directory, so perhaps we have to wait until we remove those?

@subutai In future, we could automate the extraction of these files names. There is a macro in CMake file that transverse a folder and get list of all files that fit a given criteria. I created this macro to get .cpp and .hpp files used by TestEverything project.

[rhyolight]

@david-ragazzi said:

@subutai In future, we could automate the extraction of these files names. There is a macro in CMake file that transverse a folder and get list of all files that fit a given criteria.

Let's make this part of the discussion for numenta/nupic.core-legacy#4.

[subutai]

@david-ragazzi Thanks for the explanation.