Description of problem:

  When running the GlusterFS Test Framework on EL6.4, the rpm.t test
  incorrectly reports itself as failing if the DEBUG environment
  variable is set.

  It's caused by this line of code near the top of the test:

    exec &> ${RESULT_DIR}/log

  Commenting it out stops the incorrect "fail" reporting.

  Looking at the log file generated when DEBUG=1, all of the tests
  are actually passing ("ok").

  There's just something going wrong passing that result back to
  the test framework.


Version-Release number of selected component (if applicable):

  Upstream GlusterFS git master, as of Tue 9th July 2013
  commit 60bdca792b7e572b4d79382dada1c6b93bebdd0e


How reproducible:

  Every time.


Steps to Reproduce:
1.  Setup the Gluster Test Framework on EL 6.4 (either RHEL or CentOS)
2.  export DEBUG=1
3.  Run the test suite ($ sudo ./run-tests.sh)

    The failure occurs here.

Actual results:

  # export DEBUG=1
  # prove -vf tests/basic/rpm.t 
  tests/basic/rpm.t .. 
  1..5
  Failed 5/5 subtests 

  Test Summary Report
  -------------------
  tests/basic/rpm.t (Wstat: 0 Tests: 0 Failed: 0)
    Parse errors: Bad plan.  You planned 5 tests but ran 0.
  Files=1, Tests=0, 283 wallclock secs ( 0.01 usr  0.03 sys + 133.63 cusr 45.61 csys = 179.28 CPU)
  Result: FAIL

  (and now showing a completely separate run without DEBUG set)

  # prove -vf tests/basic/rpm.t 
  tests/basic/rpm.t .. 
  1..5
  [extra stuff snipped]
  ok
  All tests successful.
  Files=1, Tests=5, 287 wallclock secs ( 0.02 usr  0.04 sys + 131.02 cusr 53.90 csys = 184.98 CPU)
  Result: PASS


Expected results:

  Setting debug mode shouldn't cause test failure to be incorrectly reported.


Additional info: