Add travis CI config

scripts/travis-build.sh

Index: scripts/travis-build.sh
diff --git a/scripts/travis-build.sh b/scripts/travis-build.sh
new file mode 100755
index 0000000000000000000000000000000000000000..60890deddcf827ea74931122e3bf4abb750c4a13
--- /dev/null
+++ b/scripts/travis-build.sh
@@ -0,0 +1,31 @@
+#!/bin/sh
+
+set -ex
+
+setup_env() {
+  # Travis sets CC/CXX to the system toolchain, so our .travis.yml
+  # exports USE_{CC,CXX} for this script to use.
+  if test -n "$USE_CC"; then

[vapier, 2016/04/11 17:13:35]

more standard imo is to use [ instead of test

 if [ -n "${USE_CC}" ]; then

[Ted Mielczarek, 2016/04/12 10:03:06]

Done.

+      export CC=$USE_CC
+  fi
+  if test -n "$USE_CXX"; then

[vapier, 2016/04/11 17:13:35]

same here

[Ted Mielczarek, 2016/04/12 10:03:06]

Done.

+      export CXX=$USE_CXX
+  fi
+}
+
+build() {
+  ./configure
+  # Use -jN for faster builds. Travis build machines under Docker
+  # have a lot of cores, but resource limits will kill the build
+  # if we try to use them all, so use at most 4.

[vapier, 2016/04/11 17:13:35]

you should reference this report:
 https://github.com/travis-ci/travis-ci/issues/1972

it's not a core utilization issue, but more the compiler uses a lot of memory,
and when you allow a lot of cores like that, the kernel starts OOM-ing.

i would move this section up to setup_env and have it export like NCPUS and
JOBS.  that way we can allow a higher # depending on the toolchain (sounds like
clang doesn't use as much memory).

[Ted Mielczarek, 2016/04/11 17:30:17]

Yeah, I understood the root cause, I may have just not phrased my explanation
clearly.

[Ted Mielczarek, 2016/04/12 10:03:06]

Done.

+  ncpus=$(getconf _NPROCESSORS_ONLN)
+  jobs=$(($ncpus<4?$ncpus:4))

[vapier, 2016/04/11 17:13:35]

there's no need to omit spaces ... that just makes it harder to read.

 jobs=$(( ncpus < 4 ? ncpus : 4 ))

[Ted Mielczarek, 2016/04/12 10:03:06]

Done.

+  make -j${jobs} check
+}
+
+main() {
+  setup_env
+  build
+}
+
+main "$@"