Re-introduce monkey patched GSON to make converter run better.

fixup! Re-introduce monkey patched GSON to make converter run better.

fixup! Re-introduce monkey patched GSON to make converter run better.

code/process-models/converting-model/build.gradle

@@ -11,6 +11,9 @@ java {
     }
 }
 dependencies {
+
+    implementation project(':third-party:monkey-patch-gson')
+
     implementation project(':code:common:model')
     implementation project(':code:api:index-api')
     implementation project(':code:common:service-discovery')

code/processes/converting-process/build.gradle

@@ -19,6 +19,9 @@ application {
 tasks.distZip.enabled = false
 
 dependencies {
+
+    implementation project(':third-party:monkey-patch-gson')
+
     implementation project(':code:common:process')
 
     implementation project(':third-party:porterstemmer')

settings.gradle

@@ -75,6 +75,7 @@ include 'third-party:uppend'
 include 'third-party:openzim'
 include 'third-party:count-min-sketch'
 include 'third-party:monkey-patch-opennlp'
+include 'third-party:monkey-patch-gson'
 
 
 dependencyResolutionManagement {

third-party/README.md

@@ -17,3 +17,4 @@ or lack an artifact, or to override some default that is inappropriate for the t
 
 ### Monkey Patched
 * [Stanford OpenNLP](monkey-patch-opennlp/) - Apache-2.0
+* [GSON](monkey-patch-gson/) - Apache-2.0

third-party/monkey-patch-gson/build.gradle

@@ -0,0 +1,17 @@
+plugins {
+    id 'java'
+}
+
+java {
+    toolchain {
+        languageVersion.set(JavaLanguageVersion.of(17))
+    }
+}
+
+dependencies {
+    implementation  libs.bundles.gson
+}
+
+test {
+    useJUnitPlatform()
+}

third-party/monkey-patch-gson/readme.md

@@ -0,0 +1,13 @@
+# Monkey Patched GSON
+
+Stanford OpenNLP - Apache-2.0
+
+## Rationale
+
+GSON makes some assumptions that make it not work very well
+for deserializing extremely large JSON objects. This patch 
+makes the code technically leak memory, but the way it's used
+makes this not much of a problem.
+
+It should only be applied to the converter or possibly
+loader processes, not the services.