osx8.2 is deprecated.

Problem was likely due to an upgrade of mpi in homebrew on travis

There was some "return 1". Now that exit code is checked, this would
have triggered an error

Report a failure in the following two cases:
- unknown test type
- test program exiting with non-zero code

While doing plumedcheck, verify that autoconf has been applied correctly

pushdoc script uses it to know if branches are supported or not

Merging debug flags and doc build (949397c0)
was a bad idea since the resulting job takes too long (doesn't
fit in 50 minutes for master branch)

I split them again, resulting in 6 jobs.

goes with a5daa52d

changed in order to work according to
e9aa6730

appliable -> applicable

failures are moved to the end of the list

(suggested by @valsson)

manual is always built. Developer doc has been updated accordingly
A total of 5 travis jobs is run each time

See #286

[makedoc]

Added in the list of options to be tested:
$($CXX --print-prog-name=ld) -r -o
As suggested by @alalazo

Since this option appears to work also with old compilers, I try
it as a first option.

See #291

Spot out by cppcheck on master branch. Since the problem is relevant
also for v2.3, I fix it here.

This command should be more portable. Notice that with old
gcc (e.g. 4.6) this does not work. I added a test such that
when this does not work "ld -r -o" is tried instead

removed extra comma

This change allows to load libraries locate in src/lib more easily
on a mac. Notice that traditionally this was solved by setting
DYLD_LIBRARY_PATH. However, as of osx Capitan, DYLD_LIBRARY_PATH
does not work correctly. Fixing the installed path with
install_name_tool should solve the issue.

(backported from master)

after 18b63365

instead of looking for last commit, sort them by year

Thanks to @valsson

Related to fb70153e