Skip a bunch of work for tests that we don't actually want.

[bangerth]

When setting TEST_PICKUP_REGEX, the path through the macro that is used to construct the list of tests iterates over all tests, checks the regex, checks a bunch of other things, and at the very end decides whether we actually want that test. That is a lot of work if a test doesn't match the regex -- what we should really be doing in that case is a quick exit: just check the next test.

This patch implements that: Instead of setting a flag about whether the test should be defined or not and then keep going with more work, we just call cmake's CONTINUE and skip to the next test.

There are two problems:

• CONTINUE requires cmake 3.2, but we currently only require 3.1. That might be an acceptable thing to require these days.
• This patch doesn't actually work and instead produces a lot of CMake warnings of the kind

CMake Error at /home/fac/g/bangerth/p/deal.II/1/build-strange/share/deal.II/macros/macro_deal_ii_add_test.cmake:351 (ADD_TEST):
  ADD_TEST given test NAME "multigrid/transfer_04b.mpirun=2.release" which
  already exists in this directory.
Call Stack (most recent call first):
  /home/fac/g/bangerth/p/deal.II/1/build-strange/share/deal.II/macros/macro_deal_ii_pickup_tests.cmake:311 (DEAL_II_ADD_TEST)
  CMakeLists.txt:4 (DEAL_II_PICKUP_TESTS)

I fail to see where that is from -- some other eyes might be quite useful. I thought I'd post the patch anyway to get the idea out there.

@tamiko?

[bangerth]

@tamiko ?

[tamiko]

@bangerth On it.

[tamiko]

@bangerth CONTINUE() was in the wrong for loop ^^

[tamiko]

/rebuild

[tamiko]

@bangerth You missed this sneaky ENDFOREACH() here...

[bangerth]

Ah, and there I have been railing for many years that break and continue in C/C++ are dangerous because it is so easy to get things wrong when there are nested levels of loops -- and because one can so easily change the meaning of a break/continue when refactoring and introducing or reducing the number of nested loops. Now I'm stepping into that very same trap.

I so wished that the authors of C back in the 1960s had only allowed break and continue with a label that identifies the loop that we're working on...

[tamiko]

I verified that after this changeset we still configure the same number of tests as on master.

[bangerth]

Nice, thanks for fixing my patch! Let's hope this works.