Update CP2K ReFrame test

[RMeli]

Update CP2K ReFrame test in preparation for eth-cscs/alps-uenv#259.

• Changed tolerances for SCF cycles so that the follow-up RPA calculation is more robust
  • This changes the reference runtime
• Created a new input file addressing a change on how the maximum number of SCF steps is counted in CP2K
• Updated CMake options according to changes in 2025.2
  • As previously discussed, it would be good if the uenv itself was aware of how to do the compilation (decoupling compilation instructions from the ReFrame test), but this is not addressed in this PR
• Use xfail to mark known failures
  • Requires Reframe-HPC>=4.9

• find a way to get version from env or uenv

[RMeli]

cscs-ci run alps-eiger-uenv;MY_UENV=cp2k/2025.1:v3

[RMeli]

cscs-ci run alps-daint-uenv;MY_UENV=cp2k/2025.1:v2

[Copilot]

Pull Request Overview

This PR updates the CP2K ReFrame test suite to support CP2K version 2025.2 and improve the robustness of RPA calculations by tightening SCF convergence tolerances.

• Updates SCF tolerance (EPS_SCF) from 1.0E-6 to 1.0E-9 in input files for better RPA calculation convergence
• Adds version-specific logic to handle changes in SCF step counting introduced in CP2K 2025.2
• Updates CMake build options to reflect changes in CP2K 2025.2 defaults (MPI, LIBVORI, CUDA architectures)
• Implements xfail mechanism for known v2025.1 linking issues

Reviewed Changes

Copilot reviewed 3 out of 3 changed files in this pull request and generated 8 comments.

File	Description
checks/apps/cp2k/src/H2O-128-PBE-TZ.inp	Tightened SCF convergence tolerances from 1.0E-6 to 1.0E-9 for both inner and outer SCF loops
checks/apps/cp2k/src/H2O-128-PBE-TZ-max_scf_16.inp	New input file with MAX_SCF=16 to maintain consistent runtime behavior with CP2K 2025.2's revised SCF step counting
checks/apps/cp2k/cp2k_uenv.py	Updated reference runtimes, added version detection logic, updated CMake options for 2025.2, refactored wrapper setup, and added xfail decorator for v2025.1

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[Copilot]

The xfail decorator on line 78 already marks this test as expected to fail for v2025.1. The additional check sn.assert_ne(self.version, 'v2025.1') in the sanity function is redundant. When using xfail, the test should validate its primary behavior, and the xfail mechanism will handle the expected failure case.

Suggested change

	return sn.and_(sn.assert_eq(self.job.exitcode, 0), sn.assert_ne(self.version, 'v2025.1'))
	return sn.assert_eq(self.job.exitcode, 0)

[RMeli]

This is not true. The test fails precisely because of this assertion. The tests that really fails is the following compilation test, but since it fails at the linking phase it would be wasteful to run it in full. Therefore, I made the download fail. One alternative could be to simply remove the archive from JFrog.

[Copilot]

The version_from_uenv() function could raise unclear errors if the UENV environment variable format is unexpected (e.g., IndexError if the splits don't produce enough elements). Consider adding error handling or validation, such as:

def version_from_uenv():
    try:
        return os.environ['UENV'].split('/')[1].split(':')[0]
    except (KeyError, IndexError) as e:
        raise ValueError(f"Unable to parse version from UENV environment variable: {os.environ.get('UENV', 'not set')}") from e

Suggested change

	return os.environ['UENV'].split('/')[1].split(':')[0]
	try:
	return os.environ['UENV'].split('/')[1].split(':')[0]
	except (KeyError, IndexError) as e:
	raise ValueError(f"Unable to parse version from UENV environment variable: {os.environ.get('UENV', 'not set')}") from e

[Copilot]

Converting the version to float may cause precision issues. For example, float('2025.2') equals 2025.2, but float('2025.10') would be 2025.1, which could lead to incorrect version comparisons. Consider using string comparison or a proper version comparison library like packaging.version.Version, or compare the version as a tuple of integers. For example:

version_str = version_from_uenv()
major, minor = map(int, version_str.split('.'))
if (major, minor) > (2025, 1):

[RMeli]

While this is true, the assumption is that CP2K releases only twice a year (with one or two bug fixes releases if they are really important). The function can definitely be improved, and also needs to be made more robust since it doesn't seem to work when releasing a new uenv (see eth-cscs/alps-uenv#259).

[Copilot]

Converting the version to float may cause precision issues. For example, float('2025.2') equals 2025.2, but float('2025.10') would be 2025.1, which could lead to incorrect version comparisons. Consider using string comparison or a proper version comparison library like packaging.version.Version, or compare the version as a tuple of integers. For example:

version_str = version_from_uenv()
major, minor = map(int, version_str.split('.'))
if (major, minor) > (2025, 1):

Suggested change

	version = float(version_from_uenv())
	if version > 2025.1:
	version_str = version_from_uenv()
	major, minor = map(int, version_str.split('.'))
	if (major, minor) > (2025, 1):