Fix rdf validate command in AoT builds (#211)

* Fix rdf validate command in AoT builds

We substituted the whole class instead of a single method...

I've found a better way around this, removing SecureRandom entirely. This reduces binary size from 65.11 MB to 63.96 MB. Incidental win, yay.

* fix

Author: Ostrzyciel

.github/workflows/aot-test.yml

@@ -41,8 +41,9 @@ jobs:
           target/graalvm-native-image/jelly-cli version | grep "JVM reflection: supported"
           
           # Test RDF conversions
-          echo '_:b <http://t.org/> _:b .' | target/graalvm-native-image/jelly-cli \
-            rdf to-jelly --in-format=nt > out.jelly && \
+          echo '_:b <http://t.org/> _:b .' > in.nt
+          target/graalvm-native-image/jelly-cli \
+            rdf to-jelly --in-format=nt in.nt > out.jelly && \
             [ -s out.jelly ]
           target/graalvm-native-image/jelly-cli \
             rdf from-jelly --out-format=jelly-text out.jelly > out.txt && \
@@ -56,6 +57,10 @@ jobs:
           echo '<?xml version="1.0"?><rdf:RDF xmlns:rdf="http://www.w3.org/1999/02/22-rdf-syntax-ns#"><rdf:Seq rdf:about="http://example.org/favourite-fruit"></rdf:Seq></rdf:RDF>' | \
             target/graalvm-native-image/jelly-cli rdf to-jelly --in-format "rdfxml" > rdfxml.jelly && \
             [ -s rdfxml.jelly ]
+          
+          # Test rdf validate
+          target/graalvm-native-image/jelly-cli \
+            rdf validate out.jelly --compare-to-rdf-file in.nt
 
       - name: Upload binary
         uses: actions/upload-artifact@v4

build.sbt

@@ -34,6 +34,8 @@ lazy val graalOptions = Seq(
   "--initialize-at-build-time=org.glassfish.json.UnicodeDetectingInputStream",
   "-H:+TrackPrimitiveValues", // SkipFlow optimization -- will be default in GraalVM 25
   "-H:+UsePredicates", // SkipFlow optimization -- will be default in GraalVM 25
+  // Make sure we don't include stuff that should be unreachable in the native image
+  "-H:AbortOnMethodReachable=*UUID.randomUUID*",
 )
 
 lazy val TestSerial = config("test-serial") extend Test

src/main/java/eu/neverblink/jelly/cli/graal/GraalSubstitutes.java

@@ -9,8 +9,8 @@
 
 import java.net.URI;
 import java.nio.charset.Charset;
-import java.security.Provider;
-import java.util.List;
+import java.util.UUID;
+import java.util.concurrent.ThreadLocalRandom;
 
 // Substitutions of classes and methods for GraalVM native image builds.
 // These try to remove things from the static analysis (and, in effect, from the final binary)
@@ -53,14 +53,33 @@ final class HttpLibSubstitute { }
 final class HttpEnvSubstitute { }
 
 /**
- * UUID generation is used by Jena for blank node IDs, but we can fall back to the default implementation.
+ * Use pseudo-random UUIDs instead of secure random ones for seeding the blank node allocator.
+ * The secure random number generation pulls in a lot of stuff we don't need.
+ * <p>
+ * For conversion commands, this is not used at all, because we preserve blank node IDs.
  */
-@Substitute
-@TargetClass(className = "sun.security.jca.ProviderList")
-final class ProviderListSubstitute { 
+@TargetClass(org.apache.jena.riot.lang.BlankNodeAllocatorHash.class)
+final class BlankNodeAllocatorHashSubstitute {
+    @Substitute
+    UUID freshSeed() {
+        ThreadLocalRandom r = ThreadLocalRandom.current();
+        return new UUID(r.nextLong(), r.nextLong());
+    }
+}
+
+/**
+ * Jena uses secure random number generation to create blank node IDs in its Model API.
+ * <p>
+ * Replaced with pseudo-random UUIDs to avoid including secure random number generation in the native image.
+ * This should be good enough for our purposes, because this is only used in init code for vocabularies
+ * and not for user data.
+ */
+@TargetClass(org.apache.jena.graph.BlankNodeId.class)
+final class BlankNodeIdSubstitute {
     @Substitute
-    public List<Provider> providers() {
-        return List.of();
+    public static String createFreshId() {
+        ThreadLocalRandom r = ThreadLocalRandom.current();
+        return new UUID(r.nextLong(), r.nextLong()).toString();
     }
 }