[query] no-sharing in matrix ir lowering

Author: ehigham

hail/hail/ir-gen/src/Main.scala

@@ -582,7 +582,8 @@ object Main {
     r += node("I64", in("x", att("Long"))).withTraits(Atom)
     r += node("F32", in("x", att("Float"))).withTraits(Atom)
     r += node("F64", in("x", att("Double"))).withTraits(Atom)
-    r += node("Str", in("x", att("String"))).withTraits(Atom)
+    // Making Str < Atom would lead to code bloat
+    r += node("Str", in("x", att("String")))
       .withPreamble(
         "override def toString(): String = s\"\"\"Str(\"${StringEscapeUtils.escapeString(x)}\")\"\"\""
       ): @nowarn("msg=possible missing interpolator")
@@ -671,7 +672,9 @@ object Main {
     )
 
     r += node("MakeArray", in("args", child.*), _typ("TArray")).withCompanionExtension
-    r += node("MakeStream", in("args", child.*), _typ("TStream"), mmPerElt).withCompanionExtension
+    r += node("MakeStream", in("args", child.*), _typ("TStream"), mmPerElt)
+      .typed("TStream")
+      .withCompanionExtension
     r += node("ArrayRef", in("a", child), in("i", child), errorID)
     r += node(
       "ArraySlice",

hail/hail/src/is/hail/backend/driver/BatchQueryDriver.scala

@@ -113,7 +113,7 @@ object BatchQueryDriver extends HttpLikeRpc with Logging {
           tempFileManager = new OwningTempFileManager(env.fs),
           theHailClassLoader = env.hcl,
           flags = env.flags,
-          irMetadata = new IrMetadata(),
+          irMetadata = new IrMetadata,
           blockMatrixCache = ImmutableMap.empty,
           compileCache = ImmutableMap.empty,
           irCache = ImmutableMap.empty,

hail/hail/src/is/hail/backend/driver/Py4JQueryDriver.scala

@@ -346,7 +346,7 @@ final class Py4JQueryDriver(backend: Backend) extends Closeable with Logging {
             else new OwningTempFileManager(tmpFileManager.fs),
           theHailClassLoader = hcl,
           flags = flags,
-          irMetadata = new IrMetadata(),
+          irMetadata = new IrMetadata,
           blockMatrixCache = blockMatrixCache,
           compileCache = compiledCodeCache,
           irCache = irCache,

hail/hail/src/is/hail/expr/ir/Binds.scala

@@ -2,6 +2,7 @@ package is.hail.expr.ir
 
 import is.hail.collection.FastSeq
 import is.hail.collection.compat.immutable.ArraySeq
+import is.hail.expr.ir.Scope.AGG
 import is.hail.expr.ir.defs._
 import is.hail.types.tcoerce
 import is.hail.types.virtual._
@@ -224,38 +225,40 @@ object Bindings {
   private def childEnvValue(ir: IR, i: Int): Bindings[Type] =
     ir match {
       case Block(bindings, _) =>
-        val bindingsTypes = bindings.view.take(i).map(b => b.name -> b.value.typ).to(ArraySeq)
+        val types = ArraySeq.newBuilder[(Name, Type)]
+        types.sizeHint(i)
+
         val eval = ArraySeq.newBuilder[Int]
+        eval.sizeHint(i) // most likely binding in eval
+
         val agg = ArraySeq.newBuilder[Int]
         val scan = ArraySeq.newBuilder[Int]
-        for (k <- 0 until i) bindings(k) match {
-          case Binding(_, _, Scope.EVAL) =>
-            eval += k
-          case Binding(_, _, Scope.AGG) =>
-            agg += k
-          case Binding(_, _, Scope.SCAN) =>
-            scan += k
-        }
-        if (i < bindings.length) bindings(i).scope match {
-          case Scope.EVAL =>
-            Bindings(
-              bindingsTypes,
-              eval.result(),
-              AggEnv.bindOrNoOp(agg.result()),
-              AggEnv.bindOrNoOp(scan.result()),
-            )
-          case Scope.AGG =>
-            Bindings(bindingsTypes, agg.result(), AggEnv.Promote, AggEnv.bindOrNoOp(scan.result()))
-          case Scope.SCAN =>
-            Bindings(bindingsTypes, scan.result(), AggEnv.bindOrNoOp(agg.result()), AggEnv.Promote)
+
+        for (k <- 0 until i) {
+          val Binding(name, value, scope) = bindings(k)
+          types += name -> value.typ
+          scope match {
+            case Scope.EVAL =>
+              eval += k
+            case Scope.AGG =>
+              agg += k
+            case Scope.SCAN =>
+              scan += k
+          }
         }
-        else
+
+        if (i == bindings.length || bindings(i).scope == Scope.EVAL)
           Bindings(
-            bindingsTypes,
+            types.result(),
             eval.result(),
             AggEnv.bindOrNoOp(agg.result()),
             AggEnv.bindOrNoOp(scan.result()),
           )
+        else if (bindings(i).scope == AGG)
+          Bindings(types.result(), agg.result(), AggEnv.Promote, AggEnv.bindOrNoOp(scan.result()))
+        else // SCAN
+          Bindings(types.result(), scan.result(), AggEnv.bindOrNoOp(agg.result()), AggEnv.Promote)
+
       case TailLoop(name, args, resultType, _) if i == args.length =>
         Bindings(
           args.map { case (name, ir) => name -> ir.typ } :+