Merge pull request #2115 from sbt/wip/caller

Fixes #2105/#1763. Cached resolution: removes duplicate callers during merge

ivy/src/main/scala/sbt/ivyint/CachedResolutionResolveEngine.scala

@@ -27,7 +27,7 @@ private[sbt] object CachedResolutionResolveCache {
   def createID(organization: String, name: String, revision: String) =
     ModuleRevisionId.newInstance(organization, name, revision)
   def sbtOrgTemp = JsonUtil.sbtOrgTemp
-  def graphVersion = "0.13.9B"
+  def graphVersion = "0.13.9C"
   val buildStartup: Long = System.currentTimeMillis
   lazy val todayStr: String = toYyyymmdd(buildStartup)
   lazy val tomorrowStr: String = toYyyymmdd(buildStartup + (1 day).toMillis)
@@ -470,8 +470,10 @@ private[sbt] trait CachedResolutionResolveEngine extends ResolveEngine {
       def mergeModuleReports(org: String, name: String, version: String, xs: Seq[ModuleReport]): ModuleReport = {
         val completelyEvicted = xs forall { _.evicted }
         val allCallers = xs flatMap { _.callers }
+        // Caller info is often repeated across the subprojects. We only need ModuleID info for later, so xs.head is ok.
+        val distinctByModuleId = allCallers.groupBy({ _.caller }).toList map { case (k, xs) => xs.head }
         val allArtifacts = (xs flatMap { _.artifacts }).distinct
-        xs.head.copy(artifacts = allArtifacts, evicted = completelyEvicted, callers = allCallers)
+        xs.head.copy(artifacts = allArtifacts, evicted = completelyEvicted, callers = distinctByModuleId)
       }
       val merged = (modules groupBy { m => (m.module.organization, m.module.name, m.module.revision) }).toSeq.toVector flatMap {
         case ((org, name, version), xs) =>

ivy/src/test/scala/BaseIvySpecification.scala

@@ -59,6 +59,11 @@ trait BaseIvySpecification extends Specification {
     IvyActions.updateEither(module, config, UnresolvedWarningConfiguration(), LogicalClock.unknown, Some(currentDependency), log)
   }
 
+  def cleanIvyCache: Unit =
+    {
+      IO.delete(currentTarget / "cache")
+    }
+
   def cleanCachedResolutionCache(module: IvySbt#Module): Unit =
     {
       IvyActions.cleanCachedResolutionCache(module, log)

ivy/src/test/scala/CachedResolutionSpec.scala

@@ -28,6 +28,7 @@ class CachedResolutionSpec extends BaseIvySpecification {
   import ShowLines._
 
   def e1 = {
+    cleanIvyCache
     val m = module(ModuleID("com.example", "foo", "0.1.0", Some("compile")),
       Seq(commonsIo13), Some("2.10.2"), UpdateOptions().withCachedResolution(true))
     val report = ivyUpdate(m)
@@ -65,7 +66,8 @@ class CachedResolutionSpec extends BaseIvySpecification {
   // avro:1.4.0 will be evicted by avro:1.7.7.
   // #2046 says that netty:3.2.0.Final is incorrectly evicted by netty:3.2.1.Final
   def e3 = {
-    log.setLevel(Level.Debug)
+    // log.setLevel(Level.Debug)
+    cleanIvyCache
     val m = module(ModuleID("com.example", "foo", "0.3.0", Some("compile")),
       Seq(avro177, dataAvro1940, netty320),
       Some("2.10.2"), UpdateOptions().withCachedResolution(true))

notes/0.13.9.markdown

@@ -119,7 +119,7 @@ On a larger dependency graph, the JSON file growing to be 100MB+
 with 97% of taken up by *caller* information.
 To make the matter worse, these large JSON files were never cleaned up.
 
-sbt 0.13.9 filters out artificial callers,
+sbt 0.13.9 filters out artificial or duplicate callers,
 which fixes `OutOfMemoryException` seen on some builds.
 This generally shrinks the size of JSON, so it should make the IO operations faster.
 Dynamic graphs will be rotated with directories named after `yyyy-mm-dd`,