cmake fails after Metis merge

[Frank Dellaert]

-- Building 3rdparty CMake Error at gtsam/3rdparty/metis/CMakeLists.txt:38 (include): include could not find load file:

/Users/frank/git/gtsam/gtsam/3rdparty/metis-5.1.0/GKlib/GKlibSystem.cmake

[Jing Dong]

This could be solved by just delete the build folder and have a new one.

[Frank Dellaert]

Is there another way? I don't want to redo all my settings...

[Jing Dong]

Maybe manually set the correct GKLIB_PATH in CMakeCache.txt? I didn't try this one..

[Frank Dellaert]

I fixed this by replacing the path to GKlib to /Users/frank/git/gtsam/gtsam/3rdparty/metis/GKlib manually, after seeing this was the value generated by another build. A bit weird, though, @amelim , that it is an absolute path? And, if GKlib is distributed in 3rdparty, why do we need a setting at all? Finally, if you are changing cmake files, please note we should not have informational messages, only errors or warnings.

[Frank Dellaert]

Changes to minor as this issue provides a fix, but I think it needs @amelim to clean this up a bit.

[Frank Dellaert]

Also, @amelim , I have the impression compilation takes longer now. Could you sanity-check whether metis headers (both 3dparty and yours) are sufficiently isolated? Ideally they would only be included in some .cpp file, not in a .h that is included everywhere.

[Chris Beall]

The easiest way without clearing the cache is to clear the filed. CMake will rediscover the value.

[Frank Dellaert]

The output that is new and should be suppressed, IMHO:

 GTSAM Additional: optimized;/opt/local/lib/libtbb.dylib;optimized;/opt/local/lib/libtbbmalloc.dylib;debug;/Users/dellaert/lib/libtbb_debug.dylib;debug;/Users/dellaert/lib/libtbbmalloc_debug.dylib;metis

 GTSAM Exports: CppUnitLite;wrap;metis

 GTSAM export: CppUnitLite;wrap;metis;gtsam;gtsam_unstable

[Frank Dellaert]

Also, all of a sudden complains about Doxygen?

-- Could NOT find Doxygen (missing:  DOXYGEN_EXECUTABLE) 

[Chris Beall]

Doxygen detection by CMake works fine for me. Does it work if you manually set the DOXYGEN_EXECUTABLE variable to /opt/local/bin/doxygen in CMake?

[Frank Dellaert]

It's just that there is a behavior change: I had to install doxygen to make it work, that tells me that before it did not care about whether doxygen was there or not.

[Andrew Melim]

It's possible there was a regression in develop between when I forked and then merged the feature branch. What is the desired behavior?

[Frank Dellaert]

Desired = doxygen is not needed, complain but allow user to turn off that warning by setting a SMAKe variable to false (as with TBB/OpenMP)

[Frank Dellaert]

@amelim , have you been able to look into the header issue I mentioned above?

[Andrew Melim]

Fixed one minor header issue with MetisIndex-inl.h Not sure if it addressed compile time issues though

[Andrew Melim]

Resolving this issue since comments have been addressed and original issue was corrected.