Drake: Mysterious CI Failure - Certain Text in PR's Description Results in Failure to Identify MATLAB Root

[liangfok]

In #3358, I encountered the following mysterious pre-merge CI failure in configuration gcc_linux_matlab:

CMake Error at /home/ubuntu/workspace/experimental/c5b85b52/cmake/mex.cmake:102 (string):
  string sub-command STRIP requires two arguments.
Call Stack (most recent call first):
  CMakeLists.txt:143 (mex_setup)


SWIG_EXECUTABLE: /home/ubuntu/workspace/experimental/c5b85b52/build/install/bin/swig
Checking if Swig has Matlab support... yes
Found Swig with Matlab support
CMake Error at /home/ubuntu/workspace/experimental/c5b85b52/build/install/share/cmake/SwigMatlab.cmake:43 (message):
  Please run 'mex_setup(REQUIRED)' which is provided by mex.cmake in
  https://github.com/RobotLocomotion/cmake before using this macro.
Call Stack (most recent call first):
  bindings/swig/CMakeLists.txt:55 (add_swig_matlab_module)

To debug this, I instrumented mex.cmake with a debug print statement. This resulted in the following being logged:

18:35:00 -- ************************** MATLAB ROOT: _matlab_root=This change was shown to decrease build times by

Since _matlab_root appeared to get part of the PR's description, I removed the following text from the description:

Using an HP z640 workstation with an Intel(R) Xeon(R) CPU E5-2620 v3 @ 2.40GHz, 40GB of Buffered DDR4 1866 MHz RAM, and commit 6d1f32d1c351e4c646009c3ee6e3e906bb2a4395, the build time and memory footprint without dReal is ~14 minutes and 5.1GB:

$ cd drake-distro
$ rm -rf build; externals
$ git reset --hard HEAD
$ mkdir build; cd build
$ cmake .. -DCMAKE_BUILD_TYPE:STRING=RelWithDebInfo -DDISABLE_MATLAB=TRUE -DWITH_DREAL=FALSE -DWITH_SNOPT=FALSE 
$ /usr/bin/time --format=%E,%U,%M make -j12
13:55.23,4906.59,5097528

This resulted in the PR's pre-merge CI tests to pass: https://drake-jenkins.csail.mit.edu/job/experimental/4412/.

Thus, I believe there is a bug in mex.cmake that arises when the PR's description contains certain text.

[david-german-tri]

Congratulations, this is the most surprising bug report of the month.

@mwoehlke-kitware, @BetsyMcPhail, @jamiesnape any ideas?

[jamiesnape]

I will look into it.

[mwoehlke-kitware]

Why does mex.cmake look at the PR description? Or indeed have any concept of PR's? I suspect the problem is in the CI scripts...

[jamiesnape]

Among other things, Jenkins sets various environment variables, one of which contains the PR description.

[jamiesnape]

Fail: https://drake-jenkins.csail.mit.edu/job/experimental/4407/compiler=gcc,label=linux_matlab,type=release/injectedEnvVars/
Pass: https://drake-jenkins.csail.mit.edu/job/experimental/4412/injectedEnvVars/

[mwoehlke-kitware]

Yes, but why would mex.cmake look at a PR description?

Oh, I see the problem... looking at that error, it looks like matlab -e | grep -e MATLAB= | cut -d'=' -f2 went sideways... and the PR has ...STRING=RelWithDebInfo -DDISABLE_MATLAB=TRUE -DWITH_DREAL... The grep needs to be -E -e '^MATLAB=.