The build is failing like this:

function killitwithfire () {
    trap - ALRM
    kill -ALRM $prog 2>/dev/null
    kill -9 $! 2>/dev/null && exit 0
}

function waitforit () {
    trap "killitwithfire" ALRM
    sleep $1& wait
    kill -ALRM $$
}

waitforit $1& prog=$! ; shift ;
+prog=6047
+shift
+waitforit
trap "killitwithfire" ALRM INT
+trap killitwithfire ALRM
+trap killitwithfire ALRM INT
"$@"& wait $!
+wait
+sleep
+wait 6049
sleep: missing operand
Try 'sleep --help' for more information.
RET=$?
+RET=0
if [[ "$(ps -ef | awk -v pid=$prog '$2==pid{print}{}')" != "" ]]; then
    kill -ALRM $prog
    wait $prog
fi
+kill -ALRM 6046
ps -ef | awk -v pid=$prog '$2==pid{print}{}'
killitwithfire
++killitwithfire
++trap - ALRM
++kill -ALRM 6047
++kill -9 6049
++ps -ef
++awk -v pid=6047 '$2==pid{print}{}'
/home/travis/.travis/job_stages: line 57:  6046 Segmentation fault      (core dumped) ./.travis.test.sh 30 ./bin/oonib

The command "chmod +x .travis.test.sh && ./.travis.test.sh 30 ./bin/oonib" exited with 139.

I don't understand very well what it could be, and it seems like this is something related to the way in which travis is handling the test. I am going to restart it to see if it's deterministic (I actually hope so).

I am experiencing more errors with the build, like that the keyserver doesn't know the key. 😡

Alright, it seems it's time to play golf.

Coverage increased (+0.04%) to 78.469% when pulling 9e0eca4 on fix/version_regexp into 03d738b on master.

Coverage increased (+0.09%) to 78.527% when pulling ec91e00 on fix/version_regexp into 03d738b on master.

Coverage increased (+0.04%) to 78.469% when pulling 9bb6db5 on fix/version_regexp into 03d738b on master.

I really hate these coveralls annoying comments that don't serve any purpose.

Coverage increased (+0.09%) to 78.527% when pulling d09c1c5 on fix/version_regexp into 03d738b on master.

In the end, I decided to rewrite the test that caused a segfault.

I believe the problem appeared now that travis has upgraded its infrastructure to 14.04: the previous travis build was for 03d738b and occurred on Apr, 6 when travis was still using 12.04.

The original script was brilliant: it did two nested waits to make sure the running process was either killed gracefully or with fire. But probably was too brilliant and triggered some edge case.

I did not want to wrestle too much with travis. Also, reading the script, it seems to me it's fine to rewrite it such that, if we cannot kill the background process, the build will hang and then fail (on travis).

I guess having the local build hang and the travis build fail is good enough for our purposes.

Alright, I have read the diff once more. I am going to self bless this as good, given that @hellais already blessed the diff improving the regexp on Slack and that the test changes "look good to me".