validate.sh: pass repo root as a directory argument (fix xargs split that produces phantom 'doesn't exist' errors)

[boosc]

Summary

validate.sh currently fails on PRs that bring the repository past roughly 2500-3000 .ttl files because of an xargs argument-list split, not because of any real ontology defect. This PR rewrites validate.sh to invoke the validator once with the repo root as a directory argument, which is the form already documented in the validator's own --help output.

One file changed: validate.sh. No ontology files touched.

The bug

validate.sh was a single line:

find . -type f ! -name 'graphit-ontology.*' -name "*.ttl" | xargs java -jar bin/ogit-validator.jar

GNU xargs splits the argument list into multiple command invocations whenever the joined argument string exceeds the per-invocation byte limit (around 128 KB on GitHub Actions Linux defaults, derived from POSIX ARG_MAX). Each split Java invocation receives only a subset of the .ttl files. It cannot resolve any cross-reference whose target definition lives in the other subset.

This is latent: with the current OGIT master at 1853 .ttl files it fits inside one invocation (the joined arguments are about 70 KB). As soon as a PR adds enough new .ttl files to push the total past the xargs split point, the CI starts to produce two or more Count of errors: summary blocks. The errors are all of the same shape:

ERROR: type id: http://www.purl.org/ogit/<Namespace>/<Class>, attribute id : http://www.purl.org/ogit/name doesn't exist!
ERROR: edge id: http://www.purl.org/ogit/<Namespace>/<Class>, head connection id : http://www.purl.org/ogit/Location doesn't exist!
ERROR: edge id: http://www.purl.org/ogit/<Namespace>/<Class>, connection id e: http://www.purl.org/ogit/generates doesn't exist!

The targets in those errors (ogit:name, ogit:Location, ogit:generates, ogit:Node, ogit:Timeseries, etc.) are all defined correctly in SGO/sgo/attributes/ and SGO/sgo/verbs/. They simply lived in the .ttl files handed to the other xargs-split Java invocation. The errors are phantoms produced by the split, not real ontology defects.

A recent example from CI run 26597172588: the run printed Count of errors: 2943 followed shortly by Count of errors: 312. Two summary blocks = two Java invocations = xargs split. Every error message in both blocks is a doesn't exist! against an ogit:-prefix definition that the master already ships and that this branch did not touch.

The fix

The validator already accepts directory arguments and recurses on its own. From its --help output:

<file|directory>...  .ttl files and directories to recursively validate

Example:
  java -jar ogit-validator.jar ../OGIT/
  Recursively validate the given directory

Passing the repo root as a single directory argument runs the validator in one JVM invocation. The validator's directory walk sees every .ttl file in the repo at once, and cross-reference resolution always succeeds when the references actually exist.

The new validate.sh:

#!/usr/bin/env bash
# ... explanatory comment ...
set -e
java -jar bin/ogit-validator.jar .

Local verification

OpenJDK 21, OGIT working copy at master plus an in-progress NTO/Utilities contribution that adds ~1290 .ttl files (total 3144):

$ java -jar bin/ogit-validator.jar .
Validation successful.
$ echo $?
0

Single Java invocation, exit code 0, zero error lines. The same 3144-file set when piped through find ... | xargs java -jar ... on a system that triggers the xargs split produces the multi-block phantom-error output described above.

Why this matters now

This isn't a hypothetical problem. It is currently blocking the NTO/Utilities/Electricity PR that adds ~1290 .ttl files (CGMES, UCTE-DEF, ENTSO-E NCs, IEC 60870-5, TASE.2). That PR was withdrawn from CI because of this exact phantom-error storm. The same issue will hit every future medium-to-large namespace contribution unless validate.sh is fixed.

This is the smallest possible fix: change one line in validate.sh, no ontology files touched, no validator code change.

Test plan

• CI on this PR succeeds (validator passes on master state)
• After merge: NTO/Utilities/Electricity PR can be re-opened and its CI should pass on real ontology content

🤖 Generated with Claude Code

[boosc]

Update -- der erste Fix (Directory-Argument java -jar bin/ogit-validator.jar .) war defekt. Viktor hatte recht.

Was ich verifiziert habe

Lokal mit OpenJDK 21 gegen das volle Repo:

• java -jar bin/ogit-validator.jar . -> Validation successful. (1 Zeile, ~1 Sek, hat gar nichts validiert)
• java -jar bin/ogit-validator.jar SGO -> Validation failed: No input files specified
• java -jar bin/ogit-validator.jar NTO -> Validation failed: No input files specified
• Mit absichtlich kaputter *.ttl im Root -> CI/local crasht beim Parsen, also einzelne Datei pro Argument funktioniert, aber Directory-Recursion durch den Validator selbst funktioniert nicht.

Der CI-Pass auf dem ersten Commit dieser PR (22cbbb71a6) war ein False Positive: 0 Dateien validiert = 0 Fehler. Der Validator-Binary advertised Directory-Recursion in --help, aber der Code-Pfad ist defekt.

Neuer Fix (Commit f18391b32b -- jetzt gepusht)

Zurueck zur find | xargs-Struktur, aber eine einzige Java-Invocation erzwingen mit xargs --no-run-if-empty -s 1900000. Der xargs -s-Wert liegt unterhalb des Kernel-ARG_MAX (typisch 2 MB auf Linux) und erlaubt comfortable 30k+ TTL-Files in einem Call.

Zusaetzlich Instrumentierung: das Script druckt vor dem Java-Call die File-Anzahl und Joined-Argument-Bytes. Die Count of errors:-Summary-Zeile des Validators muss in der CI-Log EXAKT EINMAL erscheinen -- mehrfache Bloecke wuerden bedeuten dass xargs erneut gesplittet hat.

CI laeuft jetzt auf dem neuen Commit; ich melde sobald die Ergebnis-Zeile da ist.

[boosc]

CI-Run bestaetigt den neuen Fix (Job 78545519748):

```
validate.sh: 1853 TTL files, 74286 bytes of paths
Validation successful.
```

Beweis:

• Vorab-Zeile zeigt 1853 TTL-Files, 74 KB Argumentstring -- die Instrumentierung dokumentiert was tatsaechlich validiert wurde
• 5.6 Sekunden Validator-Laufzeit (vs 0.6 Sek auf dem broken .-Argument-Stand der nichts validiert hat)
• EINE Summary-Zeile (Validation successful.), nicht zwei oder mehr -- der xargs-Split passiert nicht, weil 74 KB unter dem -s 1900000-Budget liegen.

Bereit fuer Review/Merge.