Blank NVS altLabel (R41 tombstone R4100053) makes empty PROGRAM_NAME a warning instead of an error

[killett]

Summary

ArgoNVSReferenceTableParser.resolveMembers indexes every collection member by
altLabel and prefLabel without checking for blank labels. NVS retires a
concept by keeping it in the collection and clearing its label, so a retired
concept becomes the map entry for "". An empty PROGRAM_NAME in a meta file
then matches a real concept and is reported as a warning instead of an error,
and the file is still FILE-ACCEPTED.

Cause

Commit 1520a0a ("Replace R41.jsonld", 2026-06-04) retired R4100053,
formerly Argo ROMANIA:

"skos:altLabel": "",
"skos:prefLabel": { "@language": "en", "@value": "Concept created in error" },
"owl:deprecated": "true"

In ArgoMetadataFileValidator.checkParameterValueAgainstRefTable,
refTable.get("") now returns that concept, so the isDeprecated() branch
fires instead of the "not in reference table" branch.

Reproduce

On a clean main checkout:

cd file_checker_exec && ./mvnw verify

validateProgramNameCheckInMetaFileIT.fileChecker_ShouldRaiseWarning_WhenBadProgramNameValue
fails. Actual output for 1902735_meta_empty_programName.nc:

<warning>PROGRAM_NAME: '' Status: Deprecated</warning>

Expected:

PROGRAM_NAME: '' Status: Invalid (not in reference table)

The test dates from 2025-12-16, before R41 handling was added in 4f01384
(2026-03-25). It appears to have been failing since the June table update.

Scope

ArgoNVSReferenceTable.RELEVANT_TABLES routes 26 collections through this same
line, so any future retirement in any table reopens the same hole. A scan of
file_checker_spec/NVS/*.jsonld currently finds exactly one blank label
(R41::R4100053), so today only PROGRAM_NAME is affected.

Proposed fix

Skip null and blank labels when building the two label indexes. Concepts remain
resolvable by id, so skos:related / narrower / broader lookups are
unaffected.

putIfLabelled(collection.getConceptMembersByAltLabelMap(), conceptMember.getAltLabel(), conceptMember);
putIfLabelled(collection.getConceptMembersByPrefLabelMap(), conceptMember.getPrefLabel(), conceptMember);

private void putIfLabelled(Map<String, SkosConcept> map, String label, SkosConcept concept) {
    if (label != null && !label.trim().isEmpty()) {
        map.put(label, concept);
    }
}

(trim().isEmpty() rather than isBlank(), since the build targets Java 8.)

With this change the full suite passes: Tests run: 67, Failures: 0, Errors: 0.

Branch: https://github.com/killett/ArgoFormatChecker/tree/fix/blank-nvs-label-tombstone

I would have opened a PR, but this repository restricts pull requests to
collaborators. Happy to have the commit cherry-picked, or to submit it another
way.

Aside

R41's dc:description states that PROGRAM_NAME is populated by R41
prefLabel, but the failure shows the value being matched against the
altLabel map. That may be intentional; flagging it separately since it does not
affect this fix.

[delphinedobler]

Dear @killett,

your aside note is the correct interpretation of the issue. In the next release of the FileChecker, the R41 will be checked against the PrefLabel, as mentioned in the NVS table's metadata.

In addition, having program_name ="" is still accepted, and will just raise a warning. This is a will at early stage of implementation of this field, to provide some flexibility to DACs. It may be revised in some time to a rejection, but for the moment, we'll keep it with this level.

warm regards,

Delphine

[killett]

Dear @delphinedobler,

Thank you for the reply and for confirming the prefLabel interpretation.

One point I'd like to raise on severity: this bug currently breaks the documented install process. On a clean
clone of main, ./mvnw clean install (README section IV.1) ends in BUILD FAILURE, because the failing
integration test aborts the build:

[ERROR] Tests run: 67, Failures: 1, Errors: 0, Skipped: 0
[INFO] BUILD FAILURE

So anyone building from source today gets a red build out of the box.

To make this easy to confirm, I've put together a clean-room verification script. It clones upstream fresh
into a temp directory, pins to v3.0.5 (b4b6ade), runs the suite unpatched (67 tests, exactly 1 failure), then
cherry-picks my fix (3c7640d) and runs again (67/0/0, BUILD SUCCESS). I've run it on two hosts with
different JDKs — macOS 26.5.1/arm64 with Temurin 21, and Linux/aarch64 with OpenJDK 25 — with identical
results on both, which rules out a toolchain artifact:

• Script: verification/verify-argoformatchecker.sh
  (https://github.com/killett/ArgoFormatChecker/blob/51e7bbf/verification/verify-argoformatchecker.sh)
• macOS output: results/macos-26.5.1-arm64-temurin21.txt (https://github.com/killett/ArgoFormatChecker/blob/51e7bbf/verification/results/macos-26.5.1-arm64-temurin21.txt)
• Linux output: results/linux-aarch64-openjdk25.txt
  (https://github.com/killett/ArgoFormatChecker/blob/51e7bbf/verification/results/linux-aarch64-openjdk25.txt)

On your point about empty PROGRAM_NAME remaining accepted: my fix does not change that behavior. The file is
FILE-ACCEPTED both before and after the patch, and a warning is raised both before and after — the only
change is the warning text, from Status: Deprecated (matching the retired tombstone concept) to Status:
Invalid (not in reference table), which is what the existing test expects. The DAC flexibility you describe
is preserved. The fix is also complementary to the prefLabel change you mention for the next release: that
addresses R41 specifically, while skipping blank labels closes the same hole for all 26 tables routed through
resolveMembers, should another collection retire a concept the same way.

One unrelated note from the same exercise: file_checker_exec/mvnw is committed without the executable bit
(mode 100644), so ./mvnw fails with Permission denied on a fresh clone until a chmod +x — happy to file that
separately if useful.

Warm regards,
Emmy