I'm looking into the failed tests to see if i can figure out a better solution

Please rewiew, and especially the "Handle weird bool" commit, and if someone can figure out why it happens in the first place (coming from the cuda.py nextend calls somehow) i'd be interested.

I just can't figure out that part....

@akesandgren Can you clarify the context in which you were hitting this?

I think there should be no need for this, since -fPIC -['someflag', 'flag2'] should get flattened to -fPIC -someflag -flag2 in time before it's actually used to define $CFLAGS.

The bool change you had to include seems be required for handling openmp being defined, it may be a sign of a different problem (that does not occur when sticking to nappend).

It's clear that the way in which the toolchain supports deals with variables is quite complex, it should probably be thoroughly revised/simplified at some point...

I was hitting it when trying to add a 'ieee' toolchain option (PR 1975), where the gcc definition is:
'ieee': ['mieee-fp', 'fno-trapping-math'],

So if you take PR 1975 by itself you will see a bunch of errors that this PR deals with.

@akesandgren we already have various other options that relate to a list of flags rather than a single flag, and that's working fine?

I don't see what's different about ieee in that respect?

Where are you seeing errors pop up without this change?

Hmm, wait, the precision options relate to lists of flags, but those are already handled via nextend.

I'm adding a test for ieee in #1975, and I'm now indeed seeing something like:

======================================================================
FAIL: test_precision_flags (__main__.ToolchainTest)
Test whether precision flags are being set correctly.
----------------------------------------------------------------------
Traceback (most recent call last):
  File "/Users/kehoste/work/easybuild-framework/test/framework/toolchain.py", line 389, in test_precision_flags
    self.assertEqual(os.getenv(var), "-O2 -march=native %s" % prec_flags[prec])
  File "/Users/kehoste/work/vsc-install/lib/vsc/install/testing.py", line 79, in assertEqual
    raise AssertionError("%s:\nDIFF%s:\n%s" % (msg, limit, ''.join(diff[:self.ASSERT_MAX_DIFF])))
AssertionError: "-['mieee-fp', 'fno-trapping-math'] -O2 -march=native" != '-O2 -march=native -mieee-fp -fno-trapping-math':
DIFF:
- -['mieee-fp', 'fno-trapping-math'] -O2 -march=native

So, this change indeed makes sense. That leaves the bool weirdness that we need to figure out...

@akesandgren OK, I think the problem is that using defining the openmp toolchain option as True is not being handled well in case there's no mapping for openmp, like is the case in cuda.py.
The openmp flag should probably be skipped all together in cuda.py...

I'll see if I can come up with a better fix, your change in easybuild/tools/variables.py is basically sweeping something under the rug, rather than an actual fix...

Yeah, i thought it was suspicious.

@akesandgren I fixed the problem differently by making sure there's always a compiler flag defined for openmp, see akesandgren#2

It's not airtight either, but at least I'm dealing with the root of the problem you ran into, I feel it's better than the change you made to nextend...

lgtm, thanks @akesandgren!