NormaliseToMonitor race condition

[Russell Taylor]

Saw this failure on a windows build:

In NormaliseToMonitorTest::testExec_inplace:
C:/jenkins/workspace/ornl_windows7_master/Code/Mantid/Framework/Algorithms/test/NormaliseToMonitorTest.h:115: Error: Expected (output->readE(i)[j] == 3.05941) up to 0.00001 (0.0000), found (3.1176 != 3.0594)
C:/jenkins/workspace/ornl_windows7_master/Code/Mantid/Framework/Algorithms/test/NormaliseToMonitorTest.h:115: Error: Expected (output->readE(i)[j] == 3.05941) up to 0.00001 (0.0000), found (3.1176 != 3.0594)
C:/jenkins/workspace/ornl_windows7_master/Code/Mantid/Framework/Algorithms/test/NormaliseToMonitorTest.h:115: Error: Expected (output->readE(i)[j] == 3.05941) up to 0.00001 (0.0000), found (3.1176 != 3.0594)
C:/jenkins/workspace/ornl_windows7_master/Code/Mantid/Framework/Algorithms/test/NormaliseToMonitorTest.h:115: Error: Expected (output->readE(i)[j] == 3.05941) up to 0.00001 (0.0000), found (3.1176 != 3.0594)
C:/jenkins/workspace/ornl_windows7_master/Code/Mantid/Framework/Algorithms/test/NormaliseToMonitorTest.h:115: Error: Expected (output->readE(i)[j] == 3.05941) up to 0.00001 (0.0000), found (3.1176 != 3.0594)
C:/jenkins/workspace/ornl_windows7_master/Code/Mantid/Framework/Algorithms/test/NormaliseToMonitorTest.h:115: Error: Expected (output->readE(i)[j] == 3.05941) up to 0.00001 (0.0000), found (3.1176 != 3.0594)
C:/jenkins/workspace/ornl_windows7_master/Code/Mantid/Framework/Algorithms/test/NormaliseToMonitorTest.h:115: Error: Expected (output->readE(i)[j] == 3.05941) up to 0.00001 (0.0000), found (3.1176 != 3.0594)
C:/jenkins/workspace/ornl_windows7_master/Code/Mantid/Framework/Algorithms/test/NormaliseToMonitorTest.h:115: Error: Expected (output->readE(i)[j] == 3.05941) up to 0.00001 (0.0000), found (3.1176 != 3.0594)
C:/jenkins/workspace/ornl_windows7_master/Code/Mantid/Framework/Algorithms/test/NormaliseToMonitorTest.h:115: Error: Expected (output->readE(i)[j] == 3.05941) up to 0.00001 (0.0000), found (3.1176 != 3.0594)

Seems likely to be a race condition shown up by the additional test coverage introduced in #7216.

[Russell Taylor]

Wrong ticket quoted in commit message:

Re #7216. Ensure non-const ref is obtained before const one.

As it may trigger a copy inside the cow_ptr and invalidate a previously obtained reference if the input & output workspaces are the same. In reality, Y & E vectors are almost never shared (outside of tests), but it's better to be fully safe/correct.

Changeset: ea3af02a52d6e923f971fa8a19af36e99b7598cd

[Russell Taylor]

Tester: Roll up for a good opportunity to learn what the deal is with the cow_ptr if you don't know already. Branch is feature/7234_nomalisetomonitor_race_condition.

Test by inspecting build jobs for failure of the above test. This has been seen (that I've noticed) on the ORNL Windows & Ubuntu incremental builds & the ISIS Windows one.

[Martyn Gigg]

Ran the test in a thin loop 50 times and had about 4-5 failures before the fix and no failures after.

[Stuart Campbell]

This ticket has been transferred to github issue ​8080