Validate generated crates against RO-Crate 1.2

.github/workflows/realapps.yml

@@ -71,8 +71,17 @@ jobs:
           done
 
       - name: Validate the generated crate
+        # rocrate-validator needs rich<14, which current Flower cannot share, so it
+        # gets its own virtual environment. The federation has finished by now.
+        # flwrcrate goes in without its dependencies (as in tests.yml), plus
+        # ro-crate-py, so the script can import flwrcrate.validation.
+        # offline=True only reads the validator's cache, so fill it first.
         run: |
-          python tests/e2e/validate_crate.py \
+          python -m venv /tmp/validator
+          /tmp/validator/bin/pip install "roc-validator>=0.13" "rocrate>=0.15"
+          /tmp/validator/bin/pip install -e . --no-deps
+          /tmp/validator/bin/rocrate-validator cache warm -p ro-crate-1.2 -p process-run-crate-0.5
+          /tmp/validator/bin/python tests/e2e/validate_crate.py \
             /tmp/flcrate_out/ro-crate/ro-crate-metadata.json \
             "${{ matrix.strategy }}" "${{ matrix.framework }}"
 

.github/workflows/tests.yml

@@ -28,8 +28,13 @@ jobs:
         # works without a Flower install.
         run: |
           python -m pip install --upgrade pip
-          pip install pytest pytest-cov "rocrate>=0.13" "tomli>=2.0; python_version < '3.11'"
+          pip install pytest pytest-cov "rocrate>=0.15" "roc-validator>=0.13" "tomli>=2.0; python_version < '3.11'"
           pip install -e . --no-deps
+
+      - name: Cache the validator's profile resources
+        # The tests validate with offline=True, which only reads rocrate-validator's
+        # cache. A fresh runner's cache is empty, so fill it first.
+        run: rocrate-validator cache warm -p ro-crate-1.2 -p process-run-crate-0.5
 
       - name: Run tests with coverage
         run: pytest -q --cov=flwrcrate --cov-report=term-missing --cov-fail-under=85

README.md

@@ -46,7 +46,7 @@ grep -rn "flwr.serverapp\|strategy.start\|server_fn\|ServerAppComponents" <app>/
 - `flwr.serverapp` **and** `strategy.start` → ✅ compatible
 - `server_fn` or `ServerAppComponents` → ❌ classic API, not supported (yet)
 
-Other requirements: Python ≥ 3.11, `rocrate ≥ 0.13`.
+Other requirements: Python ≥ 3.11, `rocrate ≥ 0.15`.
 
 ## Install
 
@@ -243,11 +243,15 @@ series.
 
 ### Validation
 
-Crates validate against the official
-[`rocrate-validator`](https://github.com/crs4/rocrate-validator). Note:
-`ro-crate-py` ≥ 0.15 writes RO-Crate **1.2**, while `rocrate-validator` 0.10
-ships only a **1.1** profile, producing a single false-positive `MUST 5.3`
-(`conformsTo` version). All other REQUIRED checks pass.
+Crates declare and validate against **RO-Crate 1.2**: flwrCrate sets the
+version explicitly, so the output does not depend on which `ro-crate-py` is
+installed. The test suite and the end-to-end workflow check generated crates
+with [`rocrate-validator`](https://github.com/crs4/rocrate-validator) at the
+REQUIRED level, against the `ro-crate-1.2` and `process-run-crate-0.5`
+profiles. Process Run Crate 0.5 is based on RO-Crate 1.1, so its inherited
+checks are not reported; the RO-Crate 1.2 check covers the base rules.
+The validator's opt-in extra checks are enabled too; they flag likely mistakes, 
+such as a plain string where a link to another entity was meant.
 
 ## Tested with
 
@@ -294,7 +298,8 @@ it produced.
 ## Development
 
 ```bash
-pip install -e ".[dev]"                      # installs pytest + pytest-cov
+pip install -e ".[dev]"                      # installs pytest + pytest-cov and roc-validator
+rocrate-validator cache warm -p ro-crate-1.2 -p process-run-crate-0.5   # once, for the validation tests
 pytest                                        # run the full unit + integration suite
 pytest --cov=flwrcrate --cov-report=term-missing   # with a coverage report
 ```

flwrcrate/crate_builder.py

@@ -56,7 +56,7 @@ def build_crate(captured: dict, crate_dir, metrics_log_path=None, model_path=Non
     """Assemble and write an RO-Crate. Returns the crate directory path."""
     uri_map = uri_map or {}
     crate_dir = Path(crate_dir)
-    crate = ROCrate()
+    crate = ROCrate(version="1.2")
 
     crate.name = captured.get("app_name") or "Federated learning run"
     crate.description = (

flwrcrate/validation.py

@@ -0,0 +1,31 @@
+"""Validate RO-Crates with rocrate-validator.
+Shared by the test suite and the end-to-end check, and meant for validating
+crates after a run.
+"""
+
+from io import StringIO
+
+from rocrate_validator import models, services
+from rocrate_validator.utils.io_helpers.output.console import Console
+from rocrate_validator.utils.io_helpers.output.text import TextOutputFormatter
+
+
+def validate(crate_dir, profile, **settings):
+    """Validate a crate folder against one profile: REQUIRED level, offline, extra checks on."""
+    return services.validate(services.ValidationSettings(
+        rocrate_uri=str(crate_dir),
+        profile_identifier=profile,
+        requirement_severity=models.Severity.REQUIRED,
+        offline=True,
+        extra_checks=True,
+        **settings,
+    ))
+
+
+def report(result):
+    """The validator's full text report, which names the entity behind each issue."""
+    out = StringIO()
+    console = Console(file=out, width=120)
+    console.register_formatter(TextOutputFormatter())
+    console.print(result)
+    return out.getvalue()