Add GMT_KNOWN_FAILURE flag to "pass" the known failures

[seisman]

Fixes #852.

[seisman]

mgd77/test_cm4.sh and psxy/antwrap.sh still fail.

[joa-quim]

I still don't understand well this idea of hiding the failing tests. What then are they good for?

[seisman]

I still don't understand well this idea of hiding the failing tests. What then are they good for?

We run all the tests nightly to make sure recent changes don't break the codes. Usually, we have two ways to know if all the tests pass:

1. Check the badges in README.md

2. Check the build logs.

It's obvious that the first one is much simpler. However, it doesn't work well for our case. Since we always have some "known to fail" tests, we have to read build logs to check if there're any new "failing" tests. After hiding these "known to fail" tests, we can know if all the tests pass at a glance.

[seisman]

There is still one failing test (mapproject/proj4.sh) on Linux. Probably it's caused by the old versions of gdal (1.11.3) or proj (4.9.2).

[joa-quim]

I see. But the price of this is to almost throw away the failing tests. We should move them into a separate category (directory?) to not hide them under the carpet.

[joa-quim]

The mapproject/proj4.sh failures are boring. They mostly happen due to changes in PROJ4.

[PaulWessel]

We need a way to run the filling test so they fail without having to find and edit individual scripts. but we can discuss this at Scripps.

[seisman]

Can we merge this? Or leave it to the summit?

[PaulWessel]

Either way, we will discuss this later this summer so perhaps nice to have things pass for now? I won't forget about the failures!

[seisman]

OK. merged.