Inadequate testing in ropp_fm test scripts

[Ian Culverwell]

During ROPP5 testing, a bug (missing file from distribution) became clear when a tester ran the executable t_twodop. This bug had not appeared when the build was passed through the ropp test folder, or when the "in situ" tests, as carried out by buildpack, had been run, because neither of these tests tries out t_twodop (or t_fascod, or any of the _ad or _tl tools).

New tests to examine these tools should be added and run automatically as part of the "make tests" component of buildpack. It's probably not necessary to extend the test folder.

[Ian Culverwell]

I don't know what I meant by that: t_twodop etc are tested by buildpack (though not in the test folder). I think the reason I failed to spot the missing file was that the results fly past the screen. I've added a "Hit any key to continue" statement at the end of all the standalone programs {t_fascod.f90, t_fascod_ad.f90, t_fascod_tl.f90, t_twodad.f90, t_twodop.f90, t_twodtl.f90}.

Analogous things need to be done for the scripts that produce graphical output, but this requirement is dealt with in ticket #241.

I won't close the ticket until these changes have been thoroughly tested.

[Ian Culverwell]

Realised that although t_twodop were called by make test, t_twodop -comp were not. They just weren't invoked in the make test section of ropp_fm/test/Makefile.am.

They are now.

As a result of running these extended tests, it was realised that t_twodtl had a bug in it: x_new%non_ideal was being initialised (correctly) to true if the -comp option was on, but then reinitialised (incorrectly) to false. So there was a contradiction between the non-linear and tangent linear models, which led to t_twodtl -comp failing its test.

A suggested correction has been approved by Sean Healy and is in the base code for ROPP5.1.

(All the other tests seemed OK - at least, they pass the tests.)