https://issues.apache.org/jira/browse/ARROW-3144 changed a C++ API and required downstream bindings to be updated. Romain wasn't immediately available to update R, so we marked the R job on Travis as an "allowed failure". That failure looked like this: https://travis-ci.org/apache/arrow/jobs/538795366#L3711-L3830 The C++ library built fine, but then the R package failed to build because it didn't line up with what's in C++.
Then, the C++ local file system patch (https://issues.apache.org/jira/browse/ARROW-5378) landed. Travis passed, though we were still ignoring the R build, which continued to fail. But, it started failing differently. Here's what the R build failure looks like on that PR, and on master since then: https://travis-ci.org/apache/arrow/jobs/539207245#L2520-L2640 The C++ library is failing to build, so we're not even getting to the expected R failure.
For reference, the "C++ & GLib & Ruby w/ gcc 5.4" build has the most similar setup to the R build, and it's still passing. One difference between the two jobs is that the GLib one has `ARROW_TRAVIS_USE_VENDORED_BOOST=1`, which sounds related to some open R issues, and `boost::filesystem` appears all over the error in the R job.