v.concrete does not mean isinstance(v, Version)

[haampie]

Steps to reproduce

After #24639

In [2]: spack.spec.Spec('mpich@3.1,3.1.1:3.1.2')
---------------------------------------------------------------------------
AttributeError                            Traceback (most recent call last)
~/spack/lib/spack/spack/cmd/python.py in <module>
----> 1 spack.spec.Spec('mpich@3.1,3.1.1:3.1.2')

~/spack/lib/spack/spack/spec.py in __init__(self, spec_like, normal, concrete, external_path, external_modules)
   1092 
   1093         if isinstance(spec_like, six.string_types):
-> 1094             spec_list = SpecParser(self).parse(spec_like)
   1095             if len(spec_list) > 1:
   1096                 raise ValueError("More than one spec in string: " + spec_like)

~/spack/lib/spack/spack/parse.py in parse(self, text)
    151     def parse(self, text):
    152         self.setup(text)
--> 153         return self.do_parse()
    154 
    155 

~/spack/lib/spack/spack/spec.py in do_parse(self)
   4687             # Cannot do lookups for versions in anonymous specs
   4688             # Only allow concrete versions using git for now
-> 4689             if spec.name and spec.versions.concrete and spec.version.is_commit:
   4690                 pkg = spec.package
   4691                 if hasattr(pkg, 'git'):

AttributeError: 'VersionRange' object has no attribute 'is_commit'

Before #24639

In [1]: spack.spec.Spec('mpich@3.1,3.1.1:3.1.2')
Out[1]: mpich@3.1:3.1

Note that this is technically incorrect, since it should be a version list of a concrete version and a range. This also hits yet another bug in spack, namely:

In [2]: spack.spec.Spec('mpich@3.1,3.1.1:3.1.2').versions.concrete
Out[2]: Version('3.1')

:/

[haampie]

@becker33 I guess the issue is not on the git version side, but rather with VersionRange x:x being concrete -- which is a bug imho. I've tried to address that here: #26208 but spack seems to rely on this behavior

[tgamblin]

VersionRange is broken

Is this title helpful?

[haampie]

What title would you give to it? There's multiple bugs going on.

1. The new git versions PR assumes that version.concrete being truthy implies that instance(version, Version), which is incorrect.
2. VersionRange can be concrete, which doesn't make sense
3. Spack relies on VersionRange being concrete in some cases.

[haampie]

The tl;dr VersionList incorrectly simplifies ranges x,x.y:x.z to x:x, which is considered a concrete version, which triggers a bug in the new git commit code which assume that concrete versions === Version objects.