[2.x] perf: Intern the values an UpdateReport is built of

The projects of a build mostly depend on the same libraries, and each project's
report materializes its own copies of every coordinate it names, so one value
exists once per project that mentions it: on a 302-module monorepo the cached
reports hold 949,492 ModuleReports for 2,036 distinct values.

UpdateReportInterner adds weak pools for ConfigRef, InclExclRule, File,
Artifact, ModuleID, Caller and ModuleReport itself. Pooling the report is worth
more than pooling its parts alone, because sharing it also shares its licenses
vector, extraAttributes map, homepage string and artifacts vector. Reports
carrying a publicationDate are canonicalized but never pooled, since that
java.util.Calendar is mutable; everything else reachable from a ModuleReport is
immutable, so sharing is semantically invisible.

The pools are weak, so a value lives exactly as long as some report references
it and nothing accumulates across a reload.

Two sites cover a freshly resolved report. SbtUpdateReport interns each report
as it builds it rather than sweeping the finished one, so only the module under
construction is ever un-interned. That alone would not survive, though:
coursier memoizes moduleReport on a key that includes the dependees, so each
project builds its own instance of a shared coordinate, and transformDetails
then rebuilt every report to drop the callers -- discarding the sharing and
leaving a copy per module per configuration. Dropping the callers is what makes
those reports value-equal in the first place, so transformDetails now re-interns
what it rebuilds, and only rebuilds when there is something to drop. A scripted
test pins it: two projects resolving one coordinate must end up holding one
ModuleReport instance, on the fresh path and from the cache.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>

lm-core/src/main/scala/sbt/internal/librarymanagement/JsonUtil.scala

@@ -40,26 +40,30 @@ private[sbt] object JsonUtil {
             oar.organization,
             oar.name,
             oar.modules map { mr =>
-              ModuleReport(
-                mr.module,
-                mr.artifacts,
-                mr.missingArtifacts,
-                mr.status,
-                mr.publicationDate,
-                mr.resolver,
-                mr.artifactResolver,
-                mr.evicted,
-                mr.evictedData,
-                mr.evictedReason,
-                mr.problem,
-                mr.homepage,
-                mr.extraAttributes,
-                mr.isDefault,
-                mr.branch,
-                mr.configurations,
-                mr.licenses,
-                filterOutArtificialCallers(mr.callers)
-              )
+              val callers = filterOutArtificialCallers(mr.callers)
+              // Reuse the instance when filtering changed nothing, so interning is not undone here.
+              if (callers eq mr.callers) mr
+              else
+                ModuleReport(
+                  mr.module,
+                  mr.artifacts,
+                  mr.missingArtifacts,
+                  mr.status,
+                  mr.publicationDate,
+                  mr.resolver,
+                  mr.artifactResolver,
+                  mr.evicted,
+                  mr.evictedData,
+                  mr.evictedReason,
+                  mr.problem,
+                  mr.homepage,
+                  mr.extraAttributes,
+                  mr.isDefault,
+                  mr.branch,
+                  mr.configurations,
+                  mr.licenses,
+                  callers
+                )
             }
           )
         }

lm-core/src/main/scala/sbt/internal/librarymanagement/UpdateReportInterner.scala

@@ -0,0 +1,74 @@
+/*
+ * sbt
+ * Copyright 2023, Scala center
+ * Copyright 2011 - 2022, Lightbend, Inc.
+ * Copyright 2008 - 2010, Mark Harrah
+ * Licensed under Apache License 2.0 (see LICENSE)
+ */
+
+package sbt.internal.librarymanagement
+
+import java.io.File
+import sbt.librarymanagement.*
+
+/**
+ * Intern pools for the immutable `UpdateReport` components to save memory.
+ *
+ * The pools are weak, so an entry lives only as long as some report references it.
+ */
+object UpdateReportInterner {
+
+  private val configRefs = new WeakInterner[ConfigRef]
+  private val rules = new WeakInterner[InclExclRule]
+  private val artifacts = new WeakInterner[Artifact]
+  private val moduleIds = new WeakInterner[ModuleID]
+  private val callers = new WeakInterner[Caller]
+  private val files = new WeakInterner[File]
+  private val moduleReports = new WeakInterner[ModuleReport]
+
+  def intern(c: ConfigRef): ConfigRef = configRefs.intern(c)
+
+  def intern(r: InclExclRule): InclExclRule = rules.intern(r)
+
+  def intern(f: File): File = files.intern(f)
+
+  def intern(a: Artifact): Artifact =
+    artifacts.internWith(a) { artifact =>
+      // Canonicalize the nested vectors so value-equal artifacts share their pieces even when only
+      // encountered once.
+      if (artifact.configurations.isEmpty) artifact
+      else artifact.withConfigurations(artifact.configurations.map(intern))
+    }
+
+  def intern(m: ModuleID): ModuleID =
+    moduleIds.internWith(m) { moduleId =>
+      if (
+        moduleId.inclusions.isEmpty && moduleId.exclusions.isEmpty &&
+        moduleId.explicitArtifacts.isEmpty
+      ) moduleId
+      else
+        moduleId
+          .withInclusions(moduleId.inclusions.map(intern))
+          .withExclusions(moduleId.exclusions.map(intern))
+          .withExplicitArtifacts(moduleId.explicitArtifacts.map(intern))
+    }
+
+  def intern(c: Caller): Caller =
+    callers.internWith(c) { caller =>
+      caller
+        .withCaller(intern(caller.caller))
+        .withCallerConfigurations(caller.callerConfigurations.map(intern))
+    }
+
+  def intern(mr: ModuleReport): ModuleReport =
+    // A publicationDate is a mutable Calendar, so canonicalize such a report but never pool it.
+    if (mr.publicationDate.isDefined) canonicalize(mr)
+    else moduleReports.internWith(mr)(canonicalize)
+
+  private def canonicalize(mr: ModuleReport): ModuleReport =
+    mr.withModule(intern(mr.module))
+      .withArtifacts(mr.artifacts.map { case (a, f) => (intern(a), intern(f)) })
+      .withMissingArtifacts(mr.missingArtifacts.map(intern))
+      .withConfigurations(mr.configurations.map(intern))
+      .withCallers(mr.callers.map(intern))
+}

lm-core/src/main/scala/sbt/internal/librarymanagement/UpdateReportPersistence.scala

@@ -26,32 +26,13 @@ final case class UpdateReportCache(
 object UpdateReportPersistence:
 
   /**
-   * The generated library-management codecs, with the artifact content hash disabled. Persisted update
-   * reports are the only thing that uses them; everything else keeps the stock `LibraryManagementCodec`
-   * object, including the `inputs` store, so `Tracked.inputChanged` still hashes contents for
-   * invalidation.
+   * The generated library-management codecs, with the artifact content hash disabled: nothing reads the
+   * hash back, and computing it re-reads the whole downloaded classpath.
    *
-   * sjsonnew serializes a `File` as a `(uri, Long)` pair whose Long is
-   * `HashUtil.sha256ToLong(file.toPath())` -- a full content hash of the file. Nothing reads it back:
-   * `IsoStringLong[File].from` parses the URI and drops the Long, and `update` decides staleness in
-   * `LibraryManagement.fileUptodate`, which checks `File.exists` and the modification time against
-   * `UpdateReport.stamps`. Meanwhile a report names an artifact once per configuration it resolved in,
-   * and the projects of a build largely share their dependencies, so writing the caches re-reads the
-   * whole downloaded classpath many times over -- easily the dominant cost of writing them -- to produce
-   * bytes no reader looks at.
-   *
-   * `fileStringLongIso` is an `implicit lazy val` in `sjsonnew.FileIsoStringLongs`, so it is a virtual
-   * member and every generated format resolves `JsonFormat[File]` as
-   * `isoStringLongFormat[File](fileStringLongIso)` through its self-type. Overriding it here therefore
-   * also reaches the `Vector[(Artifact, File)]` nested inside the generated `ModuleReportFormat`, which
-   * a locally-scoped `JsonFormat[File]` could not.
-   *
-   * The JSON shape is unchanged -- only the Long's value is -- so caches stay readable by sbt versions
-   * that still write the hash, and the ones written here stay readable by them.
+   * `fileStringLongIso` is virtual, so overriding it also reaches the `Vector[(Artifact, File)]` nested
+   * inside the generated `ModuleReportFormat`.
    */
   private[sbt] object CacheCodec extends LibraryManagementCodec:
-
-    /** `IO.toURI` emits the same text the stock iso puts in `first`, and `IO.toFile` inverts it. */
     override implicit lazy val fileStringLongIso: IsoStringLong[File] =
       IsoStringLong.iso[File](
         (f: File) => (IO.toURI(f).toASCIIString, 0L),
@@ -60,10 +41,27 @@ object UpdateReportPersistence:
 
   end CacheCodec
 
-  // Not the stock `LibraryManagementCodec`: see `CacheCodec` for why persisted reports must not
-  // content-hash the artifacts they name.
   import CacheCodec.given
 
+  /** Interns the modules of a decoded cache, in a pass since the generated reader has no hook. */
+  private def internModules(cache: UpdateReportCache): UpdateReportCache =
+    cache.copy(lite =
+      UpdateReportLite(
+        cache.lite.configurations.map(cr =>
+          ConfigurationReportLite(
+            cr.configuration,
+            cr.details.map(d =>
+              OrganizationArtifactReport(
+                d.organization,
+                d.name,
+                d.modules.map(UpdateReportInterner.intern)
+              )
+            )
+          )
+        )
+      )
+    )
+
   given updateReportCacheFormat: JsonFormat[UpdateReportCache] =
     new JsonFormat[UpdateReportCache]:
       override def read[J](
@@ -78,7 +76,7 @@ object UpdateReportPersistence:
             val stamps = unbuilder.readField[Map[String, Long]]("stamps")
             val cachedDescriptor = unbuilder.readField[File]("cachedDescriptor")
             unbuilder.endObject()
-            UpdateReportCache(lite, stats, stamps, cachedDescriptor)
+            internModules(UpdateReportCache(lite, stats, stamps, cachedDescriptor))
           case None =>
             deserializationError("Expected JsObject but found None")
 
@@ -109,6 +107,7 @@ object UpdateReportPersistence:
       .orElse(
         Try(store.read[UpdateReport]()).toOption
           .map(toCache)
+          .map(internModules)
       )
 
   def writeTo(store: CacheStore, cache: UpdateReportCache): Unit =

lm-core/src/test/scala/sbt/internal/librarymanagement/UpdateReportInternerSpec.scala

@@ -0,0 +1,132 @@
+/*
+ * sbt
+ * Copyright 2023, Scala center
+ * Copyright 2011 - 2022, Lightbend, Inc.
+ * Copyright 2008 - 2010, Mark Harrah
+ * Licensed under Apache License 2.0 (see LICENSE)
+ */
+
+package sbt.internal.librarymanagement
+
+import java.io.File
+import java.lang.ref.WeakReference
+import java.util.Calendar
+import sbt.io.IO
+import sbt.librarymanagement.*
+
+object UpdateReportInternerSpec extends verify.BasicTestSuite:
+
+  test("interning a value-equal ConfigRef returns the same instance"):
+    val a = ConfigRef("compile")
+    val b = ConfigRef("compile")
+    val ia = UpdateReportInterner.intern(a)
+    assert(ia eq UpdateReportInterner.intern(b), "value-equal ConfigRefs should share one instance")
+    assert(ia == a, "interning must preserve value equality")
+
+  test("interning a value-equal InclExclRule shares one instance"):
+    val r1 = InclExclRule().withOrganization("org.bad").withName("bad-lib")
+    val r2 = InclExclRule().withOrganization("org.bad").withName("bad-lib")
+    assert(r1 ne r2)
+    val i1 = UpdateReportInterner.intern(r1)
+    assert(i1 eq UpdateReportInterner.intern(r2))
+    assert(i1 == r1)
+
+  test("interning a value-equal File shares one instance"):
+    val f1 = new File("/tmp/shared/module.jar")
+    val f2 = new File("/tmp/shared/module.jar")
+    assert(f1 ne f2)
+    assert(UpdateReportInterner.intern(f1) eq UpdateReportInterner.intern(f2))
+
+  test("interning a value-equal Artifact shares one instance and preserves equality"):
+    def artifact =
+      Artifact("lib", "jar", "jar", None, Vector(ConfigRef("compile")), None, Map.empty, None)
+    val a1 = artifact
+    val a2 = artifact
+    assert(a1 ne a2)
+    val i1 = UpdateReportInterner.intern(a1)
+    assert(i1 eq UpdateReportInterner.intern(a2))
+    assert(i1 == a1)
+
+  test("interning a value-equal ModuleID shares one instance and canonicalizes nested rules"):
+    def excl = InclExclRule().withOrganization("org.bad").withName("bad")
+    val m1 = ModuleID("org.example", "lib", "1.0.0").withExclusions(Vector(excl))
+    val m2 = ModuleID("org.example", "lib", "1.0.0").withExclusions(Vector(excl))
+    assert(m1 ne m2)
+    val i1 = UpdateReportInterner.intern(m1)
+    val i2 = UpdateReportInterner.intern(m2)
+    assert(i1 eq i2, "value-equal ModuleIDs should share one instance")
+    assert(i1 == m1)
+    assert(i1.exclusions.head eq i2.exclusions.head, "nested exclusion rules should be shared too")
+
+  test("interning a value-equal Caller shares one instance"):
+    def caller =
+      Caller(
+        ModuleID("org.parent", "parent", "2.0.0"),
+        Vector(ConfigRef("compile")),
+        Map.empty,
+        isForceDependency = false,
+        isChangingDependency = false,
+        isTransitiveDependency = true,
+        isDirectlyForceDependency = false
+      )
+    val c1 = caller
+    val c2 = caller
+    assert(c1 ne c2)
+    assert(UpdateReportInterner.intern(c1) eq UpdateReportInterner.intern(c2))
+
+  test("interning a value-equal ModuleReport shares one instance"):
+    IO.withTemporaryDirectory: baseDir =>
+      val jar = new File(baseDir, "pooled.jar")
+      IO.touch(jar)
+      def mr = plainModuleReport("org.example", "pooled-lib", jar)
+      val a = mr
+      val b = mr
+      assert(a ne b)
+      val ia = UpdateReportInterner.intern(a)
+      assert(ia eq UpdateReportInterner.intern(b), "value-equal reports should share one instance")
+      assert(ia == a, "pooling must preserve value equality")
+
+  test("a ModuleReport carrying a publicationDate is canonicalized but not pooled"):
+    IO.withTemporaryDirectory: baseDir =>
+      val jar = new File(baseDir, "dated.jar")
+      IO.touch(jar)
+      val epoch = Calendar.getInstance()
+      epoch.setTimeInMillis(0L)
+      def mr = plainModuleReport("org.example", "dated-lib", jar).withPublicationDate(Some(epoch))
+      val a = UpdateReportInterner.intern(mr)
+      val b = UpdateReportInterner.intern(mr)
+      assert(a ne b, "sharing would alias a mutable java.util.Calendar")
+      assert(a == b, "the two instances must still be value-equal")
+      assert(a.module eq b.module, "the immutable coordinate inside is still interned")
+
+  test("an interned value is released once nothing else references it"):
+    // ModuleID has no factory cache and the name is unique, so the pool is the only thing that could
+    // retain it. A strong pool would pin it for the classloader's life.
+    val weak = new WeakReference(
+      UpdateReportInterner.intern(ModuleID("org.example.interner", "released-coordinate", "1.0.0"))
+    )
+    assert(awaitCleared(weak), "the interner must release a value once no report references it")
+
+  test("a pooled ModuleReport is released once nothing else references it"):
+    IO.withTemporaryDirectory: baseDir =>
+      val jar = new File(baseDir, "transient.jar")
+      IO.touch(jar)
+      val weak = new WeakReference(
+        UpdateReportInterner.intern(
+          plainModuleReport("org.example.interner", "released-report", jar)
+        )
+      )
+      assert(awaitCleared(weak), "the pool must not pin a report no caller references")
+
+  private def plainModuleReport(org: String, name: String, jar: File): ModuleReport =
+    val artifact = Artifact(name, "jar", "jar", None, Vector.empty, None, Map.empty, None)
+    ModuleReport(ModuleID(org, name, "1.0.0"), Vector((artifact, jar)), Vector.empty)
+      .withConfigurations(Vector(ConfigRef("compile")))
+
+  private def awaitCleared(ref: WeakReference[?]): Boolean =
+    var i = 0
+    while i < 50 && ref.get != null do
+      System.gc()
+      Thread.sleep(20)
+      i += 1
+    ref.get == null

lm-coursier/src/main/scala/lmcoursier/internal/SbtUpdateReport.scala

@@ -16,6 +16,7 @@ import coursier.core.{
 }
 import coursier.maven.MavenAttributes
 import coursier.util.Artifact
+import sbt.internal.librarymanagement.UpdateReportInterner
 import sbt.librarymanagement.{ Artifact as _, Configuration as _, * }
 import sbt.util.Logger
 import scala.annotation.nowarn
@@ -147,23 +148,26 @@ private[internal] object SbtUpdateReport {
       sbtMissingArtifacts.toVector
     )
 
-    rep
-      // .withStatus(None)
-      .withPublicationDate(publicationDate)
-      // .withResolver(None)
-      // .withArtifactResolver(None)
-      // .withEvicted(false)
-      // .withEvictedData(None)
-      // .withEvictedReason(None)
-      // .withProblem(None)
-      .withHomepage(Some(project.info.homePage).filter(_.nonEmpty))
-      .withLicenses(project.info.licenses.toVector)
-      .withExtraAttributes(dependency.module.attributes ++ infoProperties(project))
-      // .withIsDefault(None)
-      // .withBranch(None)
-      .withConfigurations(project.configurations.keys.toVector.map(c => ConfigRef(c.value)))
-      .withLicenses(project.info.licenses.toVector)
-      .withCallers(callers.toVector)
+    // Intern as each report is built, so a coordinate is one instance across every project.
+    UpdateReportInterner.intern(
+      rep
+        // .withStatus(None)
+        .withPublicationDate(publicationDate)
+        // .withResolver(None)
+        // .withArtifactResolver(None)
+        // .withEvicted(false)
+        // .withEvictedData(None)
+        // .withEvictedReason(None)
+        // .withProblem(None)
+        .withHomepage(Some(project.info.homePage).filter(_.nonEmpty))
+        .withLicenses(project.info.licenses.toVector)
+        .withExtraAttributes(dependency.module.attributes ++ infoProperties(project))
+        // .withIsDefault(None)
+        // .withBranch(None)
+        .withConfigurations(project.configurations.keys.toVector.map(c => ConfigRef(c.value)))
+        .withLicenses(project.info.licenses.toVector)
+        .withCallers(callers.toVector)
+    )
   }
 
   @nowarn

main/src/main/scala/sbt/internal/LibraryManagement.scala

@@ -265,7 +265,12 @@ private[sbt] object LibraryManagement {
       else
         crs1 map { cr =>
           val mrs0 = cr.modules
-          val mrs1 = mrs0 map { _.withCallers(Vector()) }
+          // Re-intern what stripping rebuilds: dropping the callers is what makes two projects'
+          // reports of the same coordinate value-equal, so this is where they collapse to one.
+          val mrs1 = mrs0 map { mr =>
+            if (mr.callers.isEmpty) mr
+            else UpdateReportInterner.intern(mr.withCallers(Vector()))
+          }
           cr.withModules(mrs1)
         }
     ur.withConfigurations(crs2)

sbt-app/src/sbt-test/dependency-management/update-report-interning/build.sbt

@@ -0,0 +1,43 @@
+ThisBuild / scalaVersion := "2.13.16"
+
+lazy val checkSharedAcrossProjects = taskKey[Unit]("Validates cross-project module report sharing")
+
+// Two projects resolving the same coordinate independently. Coursier memoizes moduleReport on a key
+// that includes the dependees, so each project builds its own instance; they become value-equal only
+// once `update` strips the callers, which is where interning collapses them.
+lazy val a = project.settings(
+  libraryDependencies += "org.scala-lang.modules" %% "scala-xml" % "2.4.0"
+)
+
+lazy val b = project.settings(
+  libraryDependencies += "org.scala-lang.modules" %% "scala-xml" % "2.4.0"
+)
+
+lazy val root = (project in file("."))
+  .aggregate(a, b)
+  .settings(
+    checkSharedAcrossProjects := {
+      def xml(ur: UpdateReport) =
+        ur.configurations
+          .find(_.configuration.name == "compile")
+          .getOrElse(sys.error("no compile configuration"))
+          .modules
+          .find(_.module.name.startsWith("scala-xml"))
+          .getOrElse(sys.error("scala-xml not in the report"))
+      val fromA = xml((a / update).value)
+      val fromB = xml((b / update).value)
+      require(fromA == fromB, s"expected value-equal reports, got\n$fromA\nand\n$fromB")
+      require(
+        fromA.callers.isEmpty,
+        s"update should have stripped the callers, got ${fromA.callers}"
+      )
+      require(
+        fromA eq fromB,
+        "two projects resolving one coordinate must share a single ModuleReport instance"
+      )
+      require(
+        fromA.module eq fromB.module,
+        "the ModuleID inside must be shared too"
+      )
+    }
+  )

sbt-app/src/sbt-test/dependency-management/update-report-interning/test

@@ -0,0 +1,6 @@
+# Both projects resolve fresh here, so this exercises the coursier construction path
+# rather than the cached-read path.
+> checkSharedAcrossProjects
+
+# And again with the reports coming back from the cache.
+> checkSharedAcrossProjects