Merge branch 'upstream' into merge

Analysis/include/Luau/InsertionOrderedMap.h

@@ -0,0 +1,134 @@
+// This file is part of the Luau programming language and is licensed under MIT License; see LICENSE.txt for details
+#pragma once
+
+#include "Luau/Common.h"
+
+#include <unordered_map>
+#include <vector>
+#include <type_traits>
+#include <iterator>
+
+namespace Luau
+{
+
+template<typename K, typename V>
+struct InsertionOrderedMap
+{
+    static_assert(std::is_trivially_copyable_v<K>, "key must be trivially copyable");
+
+    private:
+    using vec = std::vector<std::pair<K, V>>;
+
+    public:
+    using iterator = typename vec::iterator;
+    using const_iterator = typename vec::const_iterator;
+
+    void insert(K k, V v)
+    {
+        if (indices.count(k) != 0)
+            return;
+
+        pairs.push_back(std::make_pair(k, std::move(v)));
+        indices[k] = pairs.size() - 1;
+    }
+
+    void clear()
+    {
+        pairs.clear();
+        indices.clear();
+    }
+
+    size_t size() const
+    {
+        LUAU_ASSERT(pairs.size() == indices.size());
+        return pairs.size();
+    }
+
+    bool contains(const K& k) const
+    {
+        return indices.count(k) > 0;
+    }
+
+    const V* get(const K& k) const
+    {
+        auto it = indices.find(k);
+        if (it == indices.end())
+            return nullptr;
+        else
+            return &pairs.at(it->second).second;
+    }
+
+    V* get(const K& k)
+    {
+        auto it = indices.find(k);
+        if (it == indices.end())
+            return nullptr;
+        else
+            return &pairs.at(it->second).second;
+    }
+
+    const_iterator begin() const
+    {
+        return pairs.begin();
+    }
+
+    const_iterator end() const
+    {
+        return pairs.end();
+    }
+
+    iterator begin()
+    {
+        return pairs.begin();
+    }
+
+    iterator end()
+    {
+        return pairs.end();
+    }
+
+    const_iterator find(K k) const
+    {
+        auto indicesIt = indices.find(k);
+        if (indicesIt == indices.end())
+            return end();
+        else
+            return begin() + indicesIt->second;
+    }
+
+    iterator find(K k)
+    {
+        auto indicesIt = indices.find(k);
+        if (indicesIt == indices.end())
+            return end();
+        else
+            return begin() + indicesIt->second;
+    }
+
+    void erase(iterator it)
+    {
+        if (it == pairs.end())
+            return;
+
+        K k = it->first;
+        auto indexIt = indices.find(k);
+        if (indexIt == indices.end())
+            return;
+
+        size_t removed = indexIt->second;
+        indices.erase(indexIt);
+        pairs.erase(it);
+
+        for (auto& [_, index] : indices)
+        {
+            if (index > removed)
+                --index;
+        }
+    }
+
+private:
+    vec pairs;
+    std::unordered_map<K, size_t> indices;
+};
+
+}

Analysis/include/Luau/Normalize.h

@@ -64,6 +64,7 @@ public:
 
     bool operator==(const TypeIds& there) const;
     size_t getHash() const;
+    bool isNever() const;
 };
 
 } // namespace Luau
@@ -269,12 +270,24 @@ struct NormalizedType
     NormalizedType& operator=(NormalizedType&&) = default;
 
     // IsType functions
+    /// Returns true if the type is exactly a number. Behaves like Type::isNumber()
+    bool isExactlyNumber() const;
 
-    /// Returns true if the type is a subtype of function.  This includes any and unknown.
+    /// Returns true if the type is a subtype of string(it could be a singleton). Behaves like Type::isString()
-    bool isFunction() const;
+    bool isSubtypeOfString() const;
 
-    /// Returns true if the type is a subtype of number.  This includes any and unknown.
+    // Helpers that improve readability of the above (they just say if the component is present)
-    bool isNumber() const;
+    bool hasTops() const;
+    bool hasBooleans() const;
+    bool hasClasses() const;
+    bool hasErrors() const;
+    bool hasNils() const;
+    bool hasNumbers() const;
+    bool hasStrings() const;
+    bool hasThreads() const;
+    bool hasTables() const;
+    bool hasFunctions() const;
+    bool hasTyvars() const;
 };
 
 

Analysis/src/ConstraintGraphBuilder.cpp

@@ -16,6 +16,7 @@
 #include "Luau/TypeFamily.h"
 #include "Luau/Simplify.h"
 #include "Luau/VisitType.h"
+#include "Luau/InsertionOrderedMap.h"
 
 #include <algorithm>
 
@@ -196,7 +197,7 @@ struct RefinementPartition
     bool shouldAppendNilType = false;
 };
 
-using RefinementContext = std::unordered_map<DefId, RefinementPartition>;
+using RefinementContext = InsertionOrderedMap<DefId, RefinementPartition>;
 
 static void unionRefinements(NotNull<BuiltinTypes> builtinTypes, NotNull<TypeArena> arena, const RefinementContext& lhs, const RefinementContext& rhs,
     RefinementContext& dest, std::vector<ConstraintV>* constraints)
@@ -229,8 +230,9 @@ static void unionRefinements(NotNull<BuiltinTypes> builtinTypes, NotNull<TypeAre
         TypeId rightDiscriminantTy =
             rhsIt->second.discriminantTypes.size() == 1 ? rhsIt->second.discriminantTypes[0] : intersect(rhsIt->second.discriminantTypes);
 
-        dest[def].discriminantTypes.push_back(simplifyUnion(builtinTypes, arena, leftDiscriminantTy, rightDiscriminantTy).result);
+        dest.insert(def, {});
-        dest[def].shouldAppendNilType |= partition.shouldAppendNilType || rhsIt->second.shouldAppendNilType;
+        dest.get(def)->discriminantTypes.push_back(simplifyUnion(builtinTypes, arena, leftDiscriminantTy, rightDiscriminantTy).result);
+        dest.get(def)->shouldAppendNilType |= partition.shouldAppendNilType || rhsIt->second.shouldAppendNilType;
     }
 }
 
@@ -285,11 +287,12 @@ static void computeRefinement(NotNull<BuiltinTypes> builtinTypes, NotNull<TypeAr
         }
 
         RefinementContext uncommittedRefis;
-        uncommittedRefis[proposition->breadcrumb->def].discriminantTypes.push_back(discriminantTy);
+        uncommittedRefis.insert(proposition->breadcrumb->def, {});
+        uncommittedRefis.get(proposition->breadcrumb->def)->discriminantTypes.push_back(discriminantTy);
 
         // When the top-level expression is `t[x]`, we want to refine it into `nil`, not `never`.
         if ((sense || !eq) && getMetadata<SubscriptMetadata>(proposition->breadcrumb))
-            uncommittedRefis[proposition->breadcrumb->def].shouldAppendNilType = true;
+            uncommittedRefis.get(proposition->breadcrumb->def)->shouldAppendNilType = true;
 
         for (NullableBreadcrumbId current = proposition->breadcrumb; current && current->previous; current = current->previous)
         {
@@ -302,17 +305,20 @@ static void computeRefinement(NotNull<BuiltinTypes> builtinTypes, NotNull<TypeAr
             {
                 TableType::Props props{{field->prop, Property{discriminantTy}}};
                 discriminantTy = arena->addType(TableType{std::move(props), std::nullopt, TypeLevel{}, scope.get(), TableState::Sealed});
-                uncommittedRefis[current->previous->def].discriminantTypes.push_back(discriminantTy);
+                uncommittedRefis.insert(current->previous->def, {});
+                uncommittedRefis.get(current->previous->def)->discriminantTypes.push_back(discriminantTy);
             }
         }
 
         // And now it's time to commit it.
         for (auto& [def, partition] : uncommittedRefis)
         {
-            for (TypeId discriminantTy : partition.discriminantTypes)
+            (*refis).insert(def, {});
-                (*refis)[def].discriminantTypes.push_back(discriminantTy);
 
-            (*refis)[def].shouldAppendNilType |= partition.shouldAppendNilType;
+            for (TypeId discriminantTy : partition.discriminantTypes)
+                (*refis).get(def)->discriminantTypes.push_back(discriminantTy);
+
+            (*refis).get(def)->shouldAppendNilType |= partition.shouldAppendNilType;
         }
     }
 }

Analysis/src/ConstraintSolver.cpp

@@ -785,7 +785,8 @@ bool ConstraintSolver::tryDispatch(const BinaryConstraint& c, NotNull<const Cons
         const NormalizedType* normLeftTy = normalizer->normalize(leftType);
         if (hasTypeInIntersection<FreeType>(leftType) && force)
             asMutable(leftType)->ty.emplace<BoundType>(anyPresent ? builtinTypes->anyType : builtinTypes->numberType);
-        if (normLeftTy && normLeftTy->isNumber())
+        // We want to check if the left type has tops because `any` is a valid type for the lhs
+        if (normLeftTy && (normLeftTy->isExactlyNumber() || get<AnyType>(normLeftTy->tops)))
         {
             unify(leftType, rightType, constraint->scope);
             asMutable(resultType)->ty.emplace<BoundType>(anyPresent ? builtinTypes->anyType : leftType);
@@ -805,9 +806,11 @@ bool ConstraintSolver::tryDispatch(const BinaryConstraint& c, NotNull<const Cons
     // For concatenation, if the LHS is a string, the RHS must be a string as
     // well. The result will also be a string.
     case AstExprBinary::Op::Concat:
+    {
         if (hasTypeInIntersection<FreeType>(leftType) && force)
             asMutable(leftType)->ty.emplace<BoundType>(anyPresent ? builtinTypes->anyType : builtinTypes->stringType);
-        if (isString(leftType))
+        const NormalizedType* leftNormTy = normalizer->normalize(leftType);
+        if (leftNormTy && leftNormTy->isSubtypeOfString())
         {
             unify(leftType, rightType, constraint->scope);
             asMutable(resultType)->ty.emplace<BoundType>(anyPresent ? builtinTypes->anyType : leftType);
@@ -823,14 +826,33 @@ bool ConstraintSolver::tryDispatch(const BinaryConstraint& c, NotNull<const Cons
         }
 
         break;
+    }
     // Inexact comparisons require that the types be both numbers or both
     // strings, and evaluate to a boolean.
     case AstExprBinary::Op::CompareGe:
     case AstExprBinary::Op::CompareGt:
     case AstExprBinary::Op::CompareLe:
     case AstExprBinary::Op::CompareLt:
-        if ((isNumber(leftType) && isNumber(rightType)) || (isString(leftType) && isString(rightType)) || get<NeverType>(leftType) ||
+    {
-            get<NeverType>(rightType))
+        const NormalizedType* lt = normalizer->normalize(leftType);
+        const NormalizedType* rt = normalizer->normalize(rightType);
+        // If the lhs is any, comparisons should be valid.
+        if (lt && rt && (lt->isExactlyNumber() || get<AnyType>(lt->tops)) && rt->isExactlyNumber())
+        {
+            asMutable(resultType)->ty.emplace<BoundType>(builtinTypes->booleanType);
+            unblock(resultType);
+            return true;
+        }
+
+        if (lt && rt && (lt->isSubtypeOfString() || get<AnyType>(lt->tops)) && rt->isSubtypeOfString())
+        {
+            asMutable(resultType)->ty.emplace<BoundType>(builtinTypes->booleanType);
+            unblock(resultType);
+            return true;
+        }
+
+
+        if (get<NeverType>(leftType) || get<NeverType>(rightType))
         {
             asMutable(resultType)->ty.emplace<BoundType>(builtinTypes->booleanType);
             unblock(resultType);
@@ -838,6 +860,8 @@ bool ConstraintSolver::tryDispatch(const BinaryConstraint& c, NotNull<const Cons
         }
 
         break;
+    }
+
     // == and ~= always evaluate to a boolean, and impose no other constraints
     // on their parameters.
     case AstExprBinary::Op::CompareEq:
@@ -1776,7 +1800,7 @@ struct FindRefineConstraintBlockers : TypeOnceVisitor
     }
 };
 
-}
+} // namespace
 
 static bool isNegatedAny(TypeId ty)
 {

Analysis/src/Frontend.cpp

@@ -35,7 +35,6 @@ LUAU_FASTINTVARIABLE(LuauAutocompleteCheckTimeoutMs, 100)
 LUAU_FASTFLAGVARIABLE(DebugLuauDeferredConstraintResolution, false)
 LUAU_FASTFLAGVARIABLE(DebugLuauLogSolverToJson, false)
 LUAU_FASTFLAGVARIABLE(DebugLuauReadWriteProperties, false)
-LUAU_FASTFLAGVARIABLE(LuauTypeCheckerUseCorrectScope, false)
 
 namespace Luau
 {
@@ -1196,8 +1195,7 @@ ModulePtr Frontend::check(const SourceModule& sourceModule, Mode mode, std::vect
     }
     else
     {
-        TypeChecker typeChecker(FFlag::LuauTypeCheckerUseCorrectScope ? (forAutocomplete ? globalsForAutocomplete.globalScope : globals.globalScope)
+        TypeChecker typeChecker(forAutocomplete ? globalsForAutocomplete.globalScope : globals.globalScope,
-                                                                      : globals.globalScope,
             forAutocomplete ? &moduleResolverForAutocomplete : &moduleResolver, builtinTypes, &iceHandler);
 
         if (prepareModuleScope)

Analysis/src/Normalize.cpp

@@ -108,6 +108,14 @@ size_t TypeIds::getHash() const
     return hash;
 }
 
+bool TypeIds::isNever() const
+{
+    return std::all_of(begin(), end(), [&](TypeId i) {
+        // If each typeid is never, then I guess typeid's is also never?
+        return get<NeverType>(i) != nullptr;
+    });
+}
+
 bool TypeIds::operator==(const TypeIds& there) const
 {
     return hash == there.hash && types == there.types;
@@ -228,14 +236,72 @@ NormalizedType::NormalizedType(NotNull<BuiltinTypes> builtinTypes)
 {
 }
 
-bool NormalizedType::isFunction() const
+bool NormalizedType::isExactlyNumber() const
 {
-    return !get<NeverType>(tops) || !functions.parts.empty();
+    return hasNumbers() && !hasTops() && !hasBooleans() && !hasClasses() && !hasErrors() && !hasNils() && !hasStrings() && !hasThreads() &&
+           !hasTables() && !hasFunctions() && !hasTyvars();
 }
 
-bool NormalizedType::isNumber() const
+bool NormalizedType::isSubtypeOfString() const
 {
-    return !get<NeverType>(tops) || !get<NeverType>(numbers);
+    return hasStrings() && !hasTops() && !hasBooleans() && !hasClasses() && !hasErrors() && !hasNils() && !hasNumbers() && !hasThreads() &&
+           !hasTables() && !hasFunctions() && !hasTyvars();
+}
+
+bool NormalizedType::hasTops() const
+{
+    return !get<NeverType>(tops);
+}
+
+
+bool NormalizedType::hasBooleans() const
+{
+    return !get<NeverType>(booleans);
+}
+
+bool NormalizedType::hasClasses() const
+{
+    return !classes.isNever();
+}
+
+bool NormalizedType::hasErrors() const
+{
+    return !get<NeverType>(errors);
+}
+
+bool NormalizedType::hasNils() const
+{
+    return !get<NeverType>(nils);
+}
+
+bool NormalizedType::hasNumbers() const
+{
+    return !get<NeverType>(numbers);
+}
+
+bool NormalizedType::hasStrings() const
+{
+    return !strings.isNever();
+}
+
+bool NormalizedType::hasThreads() const
+{
+    return !get<NeverType>(threads);
+}
+
+bool NormalizedType::hasTables() const
+{
+    return !tables.isNever();
+}
+
+bool NormalizedType::hasFunctions() const
+{
+    return !functions.isNever();
+}
+
+bool NormalizedType::hasTyvars() const
+{
+    return !tyvars.empty();
 }
 
 static bool isShallowInhabited(const NormalizedType& norm)

Analysis/src/TypeChecker2.cpp

@@ -1067,9 +1067,7 @@ struct TypeChecker2
         std::vector<Location> argLocs;
         argLocs.reserve(call->args.size + 1);
 
-        TypeId* maybeOriginalCallTy = module->astOriginalCallTypes.find(call);
+        auto maybeOriginalCallTy = module->astOriginalCallTypes.find(call);
-        TypeId* maybeSelectedOverload = module->astOverloadResolvedTypes.find(call);
-
         if (!maybeOriginalCallTy)
             return;
 
@@ -1093,8 +1091,19 @@ struct TypeChecker2
                 return;
             }
         }
-        else if (get<FunctionType>(originalCallTy) || get<IntersectionType>(originalCallTy))
+        else if (get<FunctionType>(originalCallTy))
         {
+            // ok.
+        }
+        else if (get<IntersectionType>(originalCallTy))
+        {
+            auto norm = normalizer.normalize(originalCallTy);
+            if (!norm)
+                return reportError(CodeTooComplex{}, call->location);
+
+            // NormalizedType::hasFunction returns true if its' tops component is `unknown`, but for soundness we want the reverse.
+            if (get<UnknownType>(norm->tops) || !norm->hasFunctions())
+                return reportError(CannotCallNonFunction{originalCallTy}, call->func->location);
         }
         else if (auto utv = get<UnionType>(originalCallTy))
         {
@@ -1164,7 +1173,7 @@ struct TypeChecker2
 
         TypePackId expectedArgTypes = testArena.addTypePack(args);
 
-        if (maybeSelectedOverload)
+        if (auto maybeSelectedOverload = module->astOverloadResolvedTypes.find(call))
         {
             // This overload might not work still: the constraint solver will
             // pass the type checker an instantiated function type that matches
@@ -1414,7 +1423,7 @@ struct TypeChecker2
         {
             // Nothing
         }
-        else if (!normalizedFnTy->isFunction())
+        else if (!normalizedFnTy->hasFunctions())
         {
             ice->ice("Internal error: Lambda has non-function type " + toString(inferredFnTy), fn->location);
         }
@@ -1793,12 +1802,14 @@ struct TypeChecker2
         case AstExprBinary::Op::CompareGt:
         case AstExprBinary::Op::CompareLe:
         case AstExprBinary::Op::CompareLt:
-            if (isNumber(leftType))
+        {
+            const NormalizedType* leftTyNorm = normalizer.normalize(leftType);
+            if (leftTyNorm && leftTyNorm->isExactlyNumber())
             {
                 reportErrors(tryUnify(scope, expr->right->location, rightType, builtinTypes->numberType));
                 return builtinTypes->numberType;
             }
-            else if (isString(leftType))
+            else if (leftTyNorm && leftTyNorm->isSubtypeOfString())
             {
                 reportErrors(tryUnify(scope, expr->right->location, rightType, builtinTypes->stringType));
                 return builtinTypes->stringType;
@@ -1810,6 +1821,8 @@ struct TypeChecker2
                     expr->location);
                 return builtinTypes->errorRecoveryType();
             }
+        }
+
         case AstExprBinary::Op::And:
         case AstExprBinary::Op::Or:
         case AstExprBinary::Op::CompareEq:

Analysis/src/TypeFamily.cpp

@@ -346,7 +346,6 @@ TypeFamilyReductionResult<TypeId> addFamilyFn(std::vector<TypeId> typeParams, st
     TypeId rhsTy = log->follow(typeParams.at(1));
     const NormalizedType* normLhsTy = normalizer->normalize(lhsTy);
     const NormalizedType* normRhsTy = normalizer->normalize(rhsTy);
-
     if (!normLhsTy || !normRhsTy)
     {
         return {std::nullopt, false, {}, {}};
@@ -355,7 +354,7 @@ TypeFamilyReductionResult<TypeId> addFamilyFn(std::vector<TypeId> typeParams, st
     {
         return {builtins->anyType, false, {}, {}};
     }
-    else if (normLhsTy->isNumber() && normRhsTy->isNumber())
+    else if ((normLhsTy->hasNumbers() || normLhsTy->hasTops()) && (normRhsTy->hasNumbers() || normRhsTy->hasTops()))
     {
         return {builtins->numberType, false, {}, {}};
     }

CLI/Compile.cpp

@@ -0,0 +1,346 @@
+// This file is part of the Luau programming language and is licensed under MIT License; see LICENSE.txt for details
+#include "lua.h"
+#include "lualib.h"
+
+#include "Luau/CodeGen.h"
+#include "Luau/Compiler.h"
+#include "Luau/BytecodeBuilder.h"
+#include "Luau/Parser.h"
+#include "Luau/TimeTrace.h"
+
+#include "FileUtils.h"
+#include "Flags.h"
+
+#include <memory>
+
+#ifdef _WIN32
+#include <io.h>
+#include <fcntl.h>
+#endif
+
+LUAU_FASTFLAG(DebugLuauTimeTracing)
+
+enum class CompileFormat
+{
+    Text,
+    Binary,
+    Remarks,
+    Codegen,        // Prints annotated native code including IR and assembly
+    CodegenAsm,     // Prints annotated native code assembly
+    CodegenIr,      // Prints annotated native code IR
+    CodegenVerbose, // Prints annotated native code including IR, assembly and outlined code
+    CodegenNull,
+    Null
+};
+
+struct GlobalOptions
+{
+    int optimizationLevel = 1;
+    int debugLevel = 1;
+} globalOptions;
+
+static Luau::CompileOptions copts()
+{
+    Luau::CompileOptions result = {};
+    result.optimizationLevel = globalOptions.optimizationLevel;
+    result.debugLevel = globalOptions.debugLevel;
+
+    return result;
+}
+
+static std::optional<CompileFormat> getCompileFormat(const char* name)
+{
+    if (strcmp(name, "text") == 0)
+        return CompileFormat::Text;
+    else if (strcmp(name, "binary") == 0)
+        return CompileFormat::Binary;
+    else if (strcmp(name, "text") == 0)
+        return CompileFormat::Text;
+    else if (strcmp(name, "remarks") == 0)
+        return CompileFormat::Remarks;
+    else if (strcmp(name, "codegen") == 0)
+        return CompileFormat::Codegen;
+    else if (strcmp(name, "codegenasm") == 0)
+        return CompileFormat::CodegenAsm;
+    else if (strcmp(name, "codegenir") == 0)
+        return CompileFormat::CodegenIr;
+    else if (strcmp(name, "codegenverbose") == 0)
+        return CompileFormat::CodegenVerbose;
+    else if (strcmp(name, "codegennull") == 0)
+        return CompileFormat::CodegenNull;
+    else if (strcmp(name, "null") == 0)
+        return CompileFormat::Null;
+    else
+        return std::nullopt;
+}
+
+static void report(const char* name, const Luau::Location& location, const char* type, const char* message)
+{
+    fprintf(stderr, "%s(%d,%d): %s: %s\n", name, location.begin.line + 1, location.begin.column + 1, type, message);
+}
+
+static void reportError(const char* name, const Luau::ParseError& error)
+{
+    report(name, error.getLocation(), "SyntaxError", error.what());
+}
+
+static void reportError(const char* name, const Luau::CompileError& error)
+{
+    report(name, error.getLocation(), "CompileError", error.what());
+}
+
+static std::string getCodegenAssembly(const char* name, const std::string& bytecode, Luau::CodeGen::AssemblyOptions options)
+{
+    std::unique_ptr<lua_State, void (*)(lua_State*)> globalState(luaL_newstate(), lua_close);
+    lua_State* L = globalState.get();
+
+    if (luau_load(L, name, bytecode.data(), bytecode.size(), 0) == 0)
+        return Luau::CodeGen::getAssembly(L, -1, options);
+
+    fprintf(stderr, "Error loading bytecode %s\n", name);
+    return "";
+}
+
+static void annotateInstruction(void* context, std::string& text, int fid, int instpos)
+{
+    Luau::BytecodeBuilder& bcb = *(Luau::BytecodeBuilder*)context;
+
+    bcb.annotateInstruction(text, fid, instpos);
+}
+
+struct CompileStats
+{
+    size_t lines;
+    size_t bytecode;
+    size_t codegen;
+
+    double readTime;
+    double miscTime;
+    double parseTime;
+    double compileTime;
+    double codegenTime;
+};
+
+static double recordDeltaTime(double& timer)
+{
+    double now = Luau::TimeTrace::getClock();
+    double delta = now - timer;
+    timer = now;
+    return delta;
+}
+
+static bool compileFile(const char* name, CompileFormat format, CompileStats& stats)
+{
+    double currts = Luau::TimeTrace::getClock();
+
+    std::optional<std::string> source = readFile(name);
+    if (!source)
+    {
+        fprintf(stderr, "Error opening %s\n", name);
+        return false;
+    }
+
+    stats.readTime += recordDeltaTime(currts);
+
+    // NOTE: Normally, you should use Luau::compile or luau_compile (see lua_require as an example)
+    // This function is much more complicated because it supports many output human-readable formats through internal interfaces
+
+    try
+    {
+        Luau::BytecodeBuilder bcb;
+
+        Luau::CodeGen::AssemblyOptions options;
+        options.outputBinary = format == CompileFormat::CodegenNull;
+
+        if (!options.outputBinary)
+        {
+            options.includeAssembly = format != CompileFormat::CodegenIr;
+            options.includeIr = format != CompileFormat::CodegenAsm;
+            options.includeOutlinedCode = format == CompileFormat::CodegenVerbose;
+        }
+
+        options.annotator = annotateInstruction;
+        options.annotatorContext = &bcb;
+
+        if (format == CompileFormat::Text)
+        {
+            bcb.setDumpFlags(Luau::BytecodeBuilder::Dump_Code | Luau::BytecodeBuilder::Dump_Source | Luau::BytecodeBuilder::Dump_Locals |
+                             Luau::BytecodeBuilder::Dump_Remarks);
+            bcb.setDumpSource(*source);
+        }
+        else if (format == CompileFormat::Remarks)
+        {
+            bcb.setDumpFlags(Luau::BytecodeBuilder::Dump_Source | Luau::BytecodeBuilder::Dump_Remarks);
+            bcb.setDumpSource(*source);
+        }
+        else if (format == CompileFormat::Codegen || format == CompileFormat::CodegenAsm || format == CompileFormat::CodegenIr ||
+                 format == CompileFormat::CodegenVerbose)
+        {
+            bcb.setDumpFlags(Luau::BytecodeBuilder::Dump_Code | Luau::BytecodeBuilder::Dump_Source | Luau::BytecodeBuilder::Dump_Locals |
+                             Luau::BytecodeBuilder::Dump_Remarks);
+            bcb.setDumpSource(*source);
+        }
+
+        stats.miscTime += recordDeltaTime(currts);
+
+        Luau::Allocator allocator;
+        Luau::AstNameTable names(allocator);
+        Luau::ParseResult result = Luau::Parser::parse(source->c_str(), source->size(), names, allocator);
+
+        if (!result.errors.empty())
+            throw Luau::ParseErrors(result.errors);
+
+        stats.lines += result.lines;
+        stats.parseTime += recordDeltaTime(currts);
+
+        Luau::compileOrThrow(bcb, result, names, copts());
+        stats.bytecode += bcb.getBytecode().size();
+        stats.compileTime += recordDeltaTime(currts);
+
+        switch (format)
+        {
+        case CompileFormat::Text:
+            printf("%s", bcb.dumpEverything().c_str());
+            break;
+        case CompileFormat::Remarks:
+            printf("%s", bcb.dumpSourceRemarks().c_str());
+            break;
+        case CompileFormat::Binary:
+            fwrite(bcb.getBytecode().data(), 1, bcb.getBytecode().size(), stdout);
+            break;
+        case CompileFormat::Codegen:
+        case CompileFormat::CodegenAsm:
+        case CompileFormat::CodegenIr:
+        case CompileFormat::CodegenVerbose:
+            printf("%s", getCodegenAssembly(name, bcb.getBytecode(), options).c_str());
+            break;
+        case CompileFormat::CodegenNull:
+            stats.codegen += getCodegenAssembly(name, bcb.getBytecode(), options).size();
+            stats.codegenTime += recordDeltaTime(currts);
+            break;
+        case CompileFormat::Null:
+            break;
+        }
+
+        return true;
+    }
+    catch (Luau::ParseErrors& e)
+    {
+        for (auto& error : e.getErrors())
+            reportError(name, error);
+        return false;
+    }
+    catch (Luau::CompileError& e)
+    {
+        reportError(name, e);
+        return false;
+    }
+}
+
+static void displayHelp(const char* argv0)
+{
+    printf("Usage: %s [--mode] [options] [file list]\n", argv0);
+    printf("\n");
+    printf("Available modes:\n");
+    printf("   binary, text, remarks, codegen\n");
+    printf("\n");
+    printf("Available options:\n");
+    printf("  -h, --help: Display this usage message.\n");
+    printf("  -O<n>: compile with optimization level n (default 1, n should be between 0 and 2).\n");
+    printf("  -g<n>: compile with debug level n (default 1, n should be between 0 and 2).\n");
+    printf("  --timetrace: record compiler time tracing information into trace.json\n");
+}
+
+static int assertionHandler(const char* expr, const char* file, int line, const char* function)
+{
+    printf("%s(%d): ASSERTION FAILED: %s\n", file, line, expr);
+    return 1;
+}
+
+int main(int argc, char** argv)
+{
+    Luau::assertHandler() = assertionHandler;
+
+    setLuauFlagsDefault();
+
+    CompileFormat compileFormat = CompileFormat::Text;
+
+    for (int i = 1; i < argc; i++)
+    {
+        if (strcmp(argv[i], "-h") == 0 || strcmp(argv[i], "--help") == 0)
+        {
+            displayHelp(argv[0]);
+            return 0;
+        }
+        else if (strncmp(argv[i], "-O", 2) == 0)
+        {
+            int level = atoi(argv[i] + 2);
+            if (level < 0 || level > 2)
+            {
+                fprintf(stderr, "Error: Optimization level must be between 0 and 2 inclusive.\n");
+                return 1;
+            }
+            globalOptions.optimizationLevel = level;
+        }
+        else if (strncmp(argv[i], "-g", 2) == 0)
+        {
+            int level = atoi(argv[i] + 2);
+            if (level < 0 || level > 2)
+            {
+                fprintf(stderr, "Error: Debug level must be between 0 and 2 inclusive.\n");
+                return 1;
+            }
+            globalOptions.debugLevel = level;
+        }
+        else if (strcmp(argv[i], "--timetrace") == 0)
+        {
+            FFlag::DebugLuauTimeTracing.value = true;
+        }
+        else if (strncmp(argv[i], "--fflags=", 9) == 0)
+        {
+            setLuauFlags(argv[i] + 9);
+        }
+        else if (argv[i][0] == '-' && argv[i][1] == '-' && getCompileFormat(argv[i] + 2))
+        {
+            compileFormat = *getCompileFormat(argv[i] + 2);
+        }
+        else if (argv[i][0] == '-')
+        {
+            fprintf(stderr, "Error: Unrecognized option '%s'.\n\n", argv[i]);
+            displayHelp(argv[0]);
+            return 1;
+        }
+    }
+
+#if !defined(LUAU_ENABLE_TIME_TRACE)
+    if (FFlag::DebugLuauTimeTracing)
+    {
+        fprintf(stderr, "To run with --timetrace, Luau has to be built with LUAU_ENABLE_TIME_TRACE enabled\n");
+        return 1;
+    }
+#endif
+
+    const std::vector<std::string> files = getSourceFiles(argc, argv);
+
+#ifdef _WIN32
+    if (compileFormat == CompileFormat::Binary)
+        _setmode(_fileno(stdout), _O_BINARY);
+#endif
+
+    CompileStats stats = {};
+    int failed = 0;
+
+    for (const std::string& path : files)
+        failed += !compileFile(path.c_str(), compileFormat, stats);
+
+    if (compileFormat == CompileFormat::Null)
+        printf("Compiled %d KLOC into %d KB bytecode (read %.2fs, parse %.2fs, compile %.2fs)\n", int(stats.lines / 1000), int(stats.bytecode / 1024),
+            stats.readTime, stats.parseTime, stats.compileTime);
+    else if (compileFormat == CompileFormat::CodegenNull)
+        printf("Compiled %d KLOC into %d KB bytecode => %d KB native code (%.2fx) (read %.2fs, parse %.2fs, compile %.2fs, codegen %.2fs)\n",
+            int(stats.lines / 1000), int(stats.bytecode / 1024), int(stats.codegen / 1024),
+            stats.bytecode == 0 ? 0.0 : double(stats.codegen) / double(stats.bytecode), stats.readTime, stats.parseTime, stats.compileTime,
+            stats.codegenTime);
+
+    return failed ? 1 : 0;
+}

CMakeLists.txt

@@ -36,12 +36,14 @@ if(LUAU_BUILD_CLI)
     add_executable(Luau.Analyze.CLI)
     add_executable(Luau.Ast.CLI)
     add_executable(Luau.Reduce.CLI)
+    add_executable(Luau.Compile.CLI)
 
     # This also adds target `name` on Linux/macOS and `name.exe` on Windows
     set_target_properties(Luau.Repl.CLI PROPERTIES OUTPUT_NAME luau)
     set_target_properties(Luau.Analyze.CLI PROPERTIES OUTPUT_NAME luau-analyze)
     set_target_properties(Luau.Ast.CLI PROPERTIES OUTPUT_NAME luau-ast)
     set_target_properties(Luau.Reduce.CLI PROPERTIES OUTPUT_NAME luau-reduce)
+    set_target_properties(Luau.Compile.CLI PROPERTIES OUTPUT_NAME luau-compile)
 endif()
 
 if(LUAU_BUILD_TESTS)
@@ -186,6 +188,7 @@ if(LUAU_BUILD_CLI)
     target_compile_options(Luau.Reduce.CLI PRIVATE ${LUAU_OPTIONS})
     target_compile_options(Luau.Analyze.CLI PRIVATE ${LUAU_OPTIONS})
     target_compile_options(Luau.Ast.CLI PRIVATE ${LUAU_OPTIONS})
+    target_compile_options(Luau.Compile.CLI PRIVATE ${LUAU_OPTIONS})
 
     target_include_directories(Luau.Repl.CLI PRIVATE extern extern/isocline/include)
 
@@ -206,6 +209,8 @@ if(LUAU_BUILD_CLI)
     target_compile_features(Luau.Reduce.CLI PRIVATE cxx_std_17)
     target_include_directories(Luau.Reduce.CLI PUBLIC Reduce/include)
     target_link_libraries(Luau.Reduce.CLI PRIVATE Luau.Common Luau.Ast Luau.Analysis)
+
+    target_link_libraries(Luau.Compile.CLI PRIVATE Luau.Compiler Luau.VM Luau.CodeGen)
 endif()
 
 if(LUAU_BUILD_TESTS)

CodeGen/include/Luau/AddressA64.h

@@ -14,13 +14,10 @@ namespace A64
 
 enum class AddressKindA64 : uint8_t
 {
-    imm, // reg + imm
     reg,  // reg + reg
-
+    imm,  // reg + imm
-    // TODO:
+    pre,  // reg + imm, reg += imm
-    // reg + reg << shift
+    post, // reg, reg += imm
-    // reg + sext(reg) << shift
-    // reg + uext(reg) << shift
 };
 
 struct AddressA64
@@ -29,13 +26,14 @@ struct AddressA64
     // For example, ldr x0, [reg+imm] is limited to 8 KB offsets assuming imm is divisible by 8, but loading into w0 reduces the range to 4 KB
     static constexpr size_t kMaxOffset = 1023;
 
-    constexpr AddressA64(RegisterA64 base, int off = 0)
+    constexpr AddressA64(RegisterA64 base, int off = 0, AddressKindA64 kind = AddressKindA64::imm)
-        : kind(AddressKindA64::imm)
+        : kind(kind)
         , base(base)
         , offset(xzr)
         , data(off)
     {
         LUAU_ASSERT(base.kind == KindA64::x || base == sp);
+        LUAU_ASSERT(kind != AddressKindA64::reg);
     }
 
     constexpr AddressA64(RegisterA64 base, RegisterA64 offset)

CodeGen/include/Luau/IrAnalysis.h

@@ -1,6 +1,8 @@
 // This file is part of the Luau programming language and is licensed under MIT License; see LICENSE.txt for details
 #pragma once
 
+#include "Luau/Common.h"
+
 #include <bitset>
 #include <utility>
 #include <vector>
@@ -37,6 +39,16 @@ struct RegisterSet
 
 void requireVariadicSequence(RegisterSet& sourceRs, const RegisterSet& defRs, uint8_t varargStart);
 
+struct BlockOrdering
+{
+    uint32_t depth = 0;
+
+    uint32_t preOrder = ~0u;
+    uint32_t postOrder = ~0u;
+
+    bool visited = false;
+};
+
 struct CfgInfo
 {
     std::vector<uint32_t> predecessors;
@@ -45,6 +57,15 @@ struct CfgInfo
     std::vector<uint32_t> successors;
     std::vector<uint32_t> successorsOffsets;
 
+    // Immediate dominators (unique parent in the dominator tree)
+    std::vector<uint32_t> idoms;
+
+    // Children in the dominator tree
+    std::vector<uint32_t> domChildren;
+    std::vector<uint32_t> domChildrenOffsets;
+
+    std::vector<BlockOrdering> domOrdering;
+
     // VM registers that are live when the block is entered
     // Additionally, an active variadic sequence can exist at the entry of the block
     std::vector<RegisterSet> in;
@@ -64,6 +85,18 @@ struct CfgInfo
     RegisterSet captured;
 };
 
+// A quick refresher on dominance and dominator trees:
+// * If A is a dominator of B (A dom B), you can never execute B without executing A first
+// * A is a strict dominator of B (A sdom B) is similar to previous one but A != B
+// * Immediate dominator node N (idom N) is a unique node T so that T sdom N,
+//   but T does not strictly dominate any other node that dominates N.
+// * Dominance frontier is a set of nodes where dominance of a node X ends.
+//   In practice this is where values established by node X might no longer hold because of join edges from other nodes coming in.
+//   This is also where PHI instructions in SSA are placed.
+void computeCfgImmediateDominators(IrFunction& function);
+void computeCfgDominanceTreeChildren(IrFunction& function);
+
+// Function used to update all CFG data
 void computeCfgInfo(IrFunction& function);
 
 struct BlockIteratorWrapper
@@ -90,10 +123,17 @@ struct BlockIteratorWrapper
     {
         return itEnd;
     }
+
+    uint32_t operator[](size_t pos) const
+    {
+        LUAU_ASSERT(pos < size_t(itEnd - itBegin));
+        return itBegin[pos];
+    }
 };
 
 BlockIteratorWrapper predecessors(const CfgInfo& cfg, uint32_t blockIdx);
 BlockIteratorWrapper successors(const CfgInfo& cfg, uint32_t blockIdx);
+BlockIteratorWrapper domChildren(const CfgInfo& cfg, uint32_t blockIdx);
 
 } // namespace CodeGen
 } // namespace Luau

CodeGen/include/Luau/IrData.h

@@ -823,6 +823,7 @@ struct IrFunction
     uint32_t validRestoreOpBlockIdx = 0;
 
     Proto* proto = nullptr;
+    bool variadic = false;
 
     CfgInfo cfg;
 

CodeGen/src/AssemblyBuilderA64.cpp

@@ -876,6 +876,9 @@ void AssemblyBuilderA64::placeA(const char* name, RegisterA64 dst, AddressA64 sr
 
     switch (src.kind)
     {
+    case AddressKindA64::reg:
+        place(dst.index | (src.base.index << 5) | (0b011'0'10 << 10) | (src.offset.index << 16) | (1 << 21) | (opsize << 22));
+        break;
     case AddressKindA64::imm:
         if (unsigned(src.data >> sizelog) < 1024 && (src.data & ((1 << sizelog) - 1)) == 0)
             place(dst.index | (src.base.index << 5) | ((src.data >> sizelog) << 10) | (opsize << 22) | (1 << 24));
@@ -884,8 +887,13 @@ void AssemblyBuilderA64::placeA(const char* name, RegisterA64 dst, AddressA64 sr
         else
             LUAU_ASSERT(!"Unable to encode large immediate offset");
         break;
-    case AddressKindA64::reg:
+    case AddressKindA64::pre:
-        place(dst.index | (src.base.index << 5) | (0b011'0'10 << 10) | (src.offset.index << 16) | (1 << 21) | (opsize << 22));
+        LUAU_ASSERT(src.data >= -256 && src.data <= 255);
+        place(dst.index | (src.base.index << 5) | (0b11 << 10) | ((src.data & ((1 << 9) - 1)) << 12) | (opsize << 22));
+        break;
+    case AddressKindA64::post:
+        LUAU_ASSERT(src.data >= -256 && src.data <= 255);
+        place(dst.index | (src.base.index << 5) | (0b01 << 10) | ((src.data & ((1 << 9) - 1)) << 12) | (opsize << 22));
         break;
     }
 
@@ -1312,23 +1320,37 @@ void AssemblyBuilderA64::log(RegisterA64 reg)
 
 void AssemblyBuilderA64::log(AddressA64 addr)
 {
-    text.append("[");
     switch (addr.kind)
     {
-    case AddressKindA64::imm:
-        log(addr.base);
-        if (addr.data != 0)
-            logAppend(",#%d", addr.data);
-        break;
     case AddressKindA64::reg:
+        text.append("[");
         log(addr.base);
         text.append(",");
         log(addr.offset);
+        text.append("]");
+        break;
+    case AddressKindA64::imm:
+        text.append("[");
+        log(addr.base);
         if (addr.data != 0)
-            logAppend(" LSL #%d", addr.data);
+            logAppend(",#%d", addr.data);
+        text.append("]");
+        break;
+    case AddressKindA64::pre:
+        text.append("[");
+        log(addr.base);
+        if (addr.data != 0)
+            logAppend(",#%d", addr.data);
+        text.append("]!");
+        break;
+    case AddressKindA64::post:
+        text.append("[");
+        log(addr.base);
+        text.append("]!");
+        if (addr.data != 0)
+            logAppend(",#%d", addr.data);
         break;
     }
-    text.append("]");
 }
 
 } // namespace A64

CodeGen/src/AssemblyBuilderX64.cpp

@@ -1415,7 +1415,7 @@ void AssemblyBuilderX64::commit()
 {
     LUAU_ASSERT(codePos <= codeEnd);
 
-    if (unsigned(codeEnd - codePos) < kMaxInstructionLength)
+    if (codeEnd - codePos < kMaxInstructionLength)
         extend();
 }
 

CodeGen/src/CodeAllocator.cpp

@@ -56,10 +56,8 @@ static void makePagesExecutable(uint8_t* mem, size_t size)
 
 static void flushInstructionCache(uint8_t* mem, size_t size)
 {
-#if WINAPI_FAMILY_PARTITION(WINAPI_PARTITION_APP | WINAPI_PARTITION_SYSTEM)
     if (FlushInstructionCache(GetCurrentProcess(), mem, size) == 0)
         LUAU_ASSERT(!"Failed to flush instruction cache");
-#endif
 }
 #else
 static uint8_t* allocatePages(size_t size)

CodeGen/src/CodeGen.cpp

@@ -268,7 +268,7 @@ static bool lowerImpl(AssemblyBuilder& build, IrLowering& lowering, IrFunction&
 [[maybe_unused]] static bool lowerIr(
     A64::AssemblyBuilderA64& build, IrBuilder& ir, NativeState& data, ModuleHelpers& helpers, Proto* proto, AssemblyOptions options)
 {
-    A64::IrLoweringA64 lowering(build, helpers, data, proto, ir.function);
+    A64::IrLoweringA64 lowering(build, helpers, data, ir.function);
 
     return lowerImpl(build, lowering, ir.function, proto->bytecodeid, options);
 }

CodeGen/src/CodeGenA64.cpp

@@ -117,6 +117,81 @@ static void emitReentry(AssemblyBuilderA64& build, ModuleHelpers& helpers)
     build.br(x4);
 }
 
+void emitReturn(AssemblyBuilderA64& build, ModuleHelpers& helpers)
+{
+    // x1 = res
+    // w2 = number of written values
+
+    // x0 = ci
+    build.ldr(x0, mem(rState, offsetof(lua_State, ci)));
+    // w3 = ci->nresults
+    build.ldr(w3, mem(x0, offsetof(CallInfo, nresults)));
+
+    Label skipResultCopy;
+
+    // Fill the rest of the expected results (nresults - written) with 'nil'
+    build.cmp(w2, w3);
+    build.b(ConditionA64::GreaterEqual, skipResultCopy);
+
+    // TODO: cmp above could compute this and flags using subs
+    build.sub(w2, w3, w2); // counter = nresults - written
+    build.mov(w4, LUA_TNIL);
+
+    Label repeatNilLoop = build.setLabel();
+    build.str(w4, mem(x1, offsetof(TValue, tt)));
+    build.add(x1, x1, sizeof(TValue));
+    build.sub(w2, w2, 1);
+    build.cbnz(w2, repeatNilLoop);
+
+    build.setLabel(skipResultCopy);
+
+    // x2 = cip = ci - 1
+    build.sub(x2, x0, sizeof(CallInfo));
+
+    // res = cip->top when nresults >= 0
+    Label skipFixedRetTop;
+    build.tbnz(w3, 31, skipFixedRetTop);
+    build.ldr(x1, mem(x2, offsetof(CallInfo, top))); // res = cip->top
+    build.setLabel(skipFixedRetTop);
+
+    // Update VM state (ci, base, top)
+    build.str(x2, mem(rState, offsetof(lua_State, ci)));      // L->ci = cip
+    build.ldr(rBase, mem(x2, offsetof(CallInfo, base)));      // sync base = L->base while we have a chance
+    build.str(rBase, mem(rState, offsetof(lua_State, base))); // L->base = cip->base
+
+    build.str(x1, mem(rState, offsetof(lua_State, top))); // L->top = res
+
+    // Unlikely, but this might be the last return from VM
+    build.ldr(w4, mem(x0, offsetof(CallInfo, flags)));
+    build.tbnz(w4, countrz(LUA_CALLINFO_RETURN), helpers.exitNoContinueVm);
+
+    // Continue in interpreter if function has no native data
+    build.ldr(w4, mem(x2, offsetof(CallInfo, flags)));
+    build.tbz(w4, countrz(LUA_CALLINFO_NATIVE), helpers.exitContinueVm);
+
+    // Need to update state of the current function before we jump away
+    build.ldr(rClosure, mem(x2, offsetof(CallInfo, func)));
+    build.ldr(rClosure, mem(rClosure, offsetof(TValue, value.gc)));
+
+    build.ldr(x1, mem(rClosure, offsetof(Closure, l.p))); // cl->l.p aka proto
+
+    LUAU_ASSERT(offsetof(Proto, code) == offsetof(Proto, k) + 8);
+    build.ldp(rConstants, rCode, mem(x1, offsetof(Proto, k))); // proto->k, proto->code
+
+    // Get instruction index from instruction pointer
+    // To get instruction index from instruction pointer, we need to divide byte offset by 4
+    // But we will actually need to scale instruction index by 4 back to byte offset later so it cancels out
+    build.ldr(x2, mem(x2, offsetof(CallInfo, savedpc))); // cip->savedpc
+    build.sub(x2, x2, rCode);
+
+    // Get new instruction location and jump to it
+    LUAU_ASSERT(offsetof(Proto, exectarget) == offsetof(Proto, execdata) + 8);
+    build.ldp(x3, x4, mem(x1, offsetof(Proto, execdata)));
+    build.ldr(w2, mem(x3, x2));
+    build.add(x4, x4, x2);
+    build.br(x4);
+}
+
 static EntryLocations buildEntryFunction(AssemblyBuilderA64& build, UnwindBuilder& unwind)
 {
     EntryLocations locations;
@@ -230,6 +305,11 @@ void assembleHelpers(AssemblyBuilderA64& build, ModuleHelpers& helpers)
         build.logAppend("; interrupt\n");
     helpers.interrupt = build.setLabel();
     emitInterrupt(build);
+
+    if (build.logText)
+        build.logAppend("; return\n");
+    helpers.return_ = build.setLabel();
+    emitReturn(build, helpers);
 }
 
 } // namespace A64

CodeGen/src/CodeGenUtils.cpp

@@ -17,8 +17,6 @@
 
 #include <string.h>
 
-LUAU_FASTFLAG(LuauUniformTopHandling)
-
 // All external function calls that can cause stack realloc or Lua calls have to be wrapped in VM_PROTECT
 // This makes sure that we save the pc (in case the Lua call needs to generate a backtrace) before the call,
 // and restores the stack pointer after in case stack gets reallocated
@@ -306,44 +304,6 @@ Closure* callFallback(lua_State* L, StkId ra, StkId argtop, int nresults)
     }
 }
 
-// Extracted as-is from lvmexecute.cpp with the exception of control flow (reentry) and removed interrupts
-Closure* returnFallback(lua_State* L, StkId ra, StkId valend)
-{
-    // ci is our callinfo, cip is our parent
-    CallInfo* ci = L->ci;
-    CallInfo* cip = ci - 1;
-
-    StkId res = ci->func; // note: we assume CALL always puts func+args and expects results to start at func
-    StkId vali = ra;
-
-    int nresults = ci->nresults;
-
-    // copy return values into parent stack (but only up to nresults!), fill the rest with nil
-    // note: in MULTRET context nresults starts as -1 so i != 0 condition never activates intentionally
-    int i;
-    for (i = nresults; i != 0 && vali < valend; i--)
-        setobj2s(L, res++, vali++);
-    while (i-- > 0)
-        setnilvalue(res++);
-
-    // pop the stack frame
-    L->ci = cip;
-    L->base = cip->base;
-    L->top = (nresults == LUA_MULTRET) ? res : cip->top;
-
-    // we're done!
-    if (LUAU_UNLIKELY(ci->flags & LUA_CALLINFO_RETURN))
-    {
-        if (!FFlag::LuauUniformTopHandling)
-            L->top = res;
-        return NULL;
-    }
-
-    // keep executing new function
-    LUAU_ASSERT(isLua(cip));
-    return clvalue(cip->func);
-}
-
 const Instruction* executeGETGLOBAL(lua_State* L, const Instruction* pc, StkId base, TValue* k)
 {
     [[maybe_unused]] Closure* cl = clvalue(L->ci->func);

CodeGen/src/CodeGenUtils.h

@@ -18,7 +18,6 @@ Closure* callProlog(lua_State* L, TValue* ra, StkId argtop, int nresults);
 void callEpilogC(lua_State* L, int nresults, int n);
 
 Closure* callFallback(lua_State* L, StkId ra, StkId argtop, int nresults);
-Closure* returnFallback(lua_State* L, StkId ra, StkId valend);
 
 const Instruction* executeGETGLOBAL(lua_State* L, const Instruction* pc, StkId base, TValue* k);
 const Instruction* executeSETGLOBAL(lua_State* L, const Instruction* pc, StkId base, TValue* k);

CodeGen/src/CodeGenX64.cpp

@@ -189,6 +189,11 @@ void assembleHelpers(X64::AssemblyBuilderX64& build, ModuleHelpers& helpers)
         build.logAppend("; continueCallInVm\n");
     helpers.continueCallInVm = build.setLabel();
     emitContinueCallInVm(build);
+
+    if (build.logText)
+        build.logAppend("; return\n");
+    helpers.return_ = build.setLabel();
+    emitReturn(build, helpers);
 }
 
 } // namespace X64

CodeGen/src/EmitCommon.h

@@ -24,6 +24,7 @@ struct ModuleHelpers
     // A64/X64
     Label exitContinueVm;
     Label exitNoContinueVm;
+    Label return_;
 
     // X64
     Label continueCallInVm;

CodeGen/src/EmitCommonX64.cpp

@@ -352,6 +352,89 @@ void emitContinueCallInVm(AssemblyBuilderX64& build)
     emitExit(build, /* continueInVm */ true);
 }
 
+void emitReturn(AssemblyBuilderX64& build, ModuleHelpers& helpers)
+{
+    // input: ci in r8, res in rdi, number of written values in ecx
+    RegisterX64 ci = r8;
+    RegisterX64 res = rdi;
+    RegisterX64 written = ecx;
+
+    RegisterX64 cip = r9;
+    RegisterX64 nresults = esi;
+
+    build.lea(cip, addr[ci - sizeof(CallInfo)]);
+
+    // nresults = ci->nresults
+    build.mov(nresults, dword[ci + offsetof(CallInfo, nresults)]);
+
+    Label skipResultCopy;
+
+    // Fill the rest of the expected results (nresults - written) with 'nil'
+    RegisterX64 counter = written;
+    build.sub(counter, nresults); // counter = -(nresults - written)
+    build.jcc(ConditionX64::GreaterEqual, skipResultCopy);
+
+    Label repeatNilLoop = build.setLabel();
+    build.mov(dword[res + offsetof(TValue, tt)], LUA_TNIL);
+    build.add(res, sizeof(TValue));
+    build.inc(counter);
+    build.jcc(ConditionX64::NotZero, repeatNilLoop);
+
+    build.setLabel(skipResultCopy);
+
+    build.mov(qword[rState + offsetof(lua_State, ci)], cip);     // L->ci = cip
+    build.mov(rBase, qword[cip + offsetof(CallInfo, base)]);     // sync base = L->base while we have a chance
+    build.mov(qword[rState + offsetof(lua_State, base)], rBase); // L->base = cip->base
+
+    Label skipFixedRetTop;
+    build.test(nresults, nresults);                       // test here will set SF=1 for a negative number and it always sets OF to 0
+    build.jcc(ConditionX64::Less, skipFixedRetTop);       // jl jumps if SF != OF
+    build.mov(res, qword[cip + offsetof(CallInfo, top)]); // res = cip->top
+    build.setLabel(skipFixedRetTop);
+
+    build.mov(qword[rState + offsetof(lua_State, top)], res); // L->top = res
+
+    // Unlikely, but this might be the last return from VM
+    build.test(byte[ci + offsetof(CallInfo, flags)], LUA_CALLINFO_RETURN);
+    build.jcc(ConditionX64::NotZero, helpers.exitNoContinueVm);
+
+    // Returning back to the previous function is a bit tricky
+    // Registers alive: r9 (cip)
+    RegisterX64 proto = rcx;
+    RegisterX64 execdata = rbx;
+
+    // Change closure
+    build.mov(rax, qword[cip + offsetof(CallInfo, func)]);
+    build.mov(rax, qword[rax + offsetof(TValue, value.gc)]);
+    build.mov(sClosure, rax);
+
+    build.mov(proto, qword[rax + offsetof(Closure, l.p)]);
+
+    build.mov(execdata, qword[proto + offsetof(Proto, execdata)]);
+
+    build.test(byte[cip + offsetof(CallInfo, flags)], LUA_CALLINFO_NATIVE);
+    build.jcc(ConditionX64::Zero, helpers.exitContinueVm); // Continue in interpreter if function has no native data
+
+    // Change constants
+    build.mov(rConstants, qword[proto + offsetof(Proto, k)]);
+
+    // Change code
+    build.mov(rdx, qword[proto + offsetof(Proto, code)]);
+    build.mov(sCode, rdx);
+
+    build.mov(rax, qword[cip + offsetof(CallInfo, savedpc)]);
+
+    // To get instruction index from instruction pointer, we need to divide byte offset by 4
+    // But we will actually need to scale instruction index by 4 back to byte offset later so it cancels out
+    build.sub(rax, rdx);
+
+    // Get new instruction location and jump to it
+    build.mov(edx, dword[execdata + rax]);
+    build.add(rdx, qword[proto + offsetof(Proto, exectarget)]);
+    build.jmp(rdx);
+}
+
+
 } // namespace X64
 } // namespace CodeGen
 } // namespace Luau

CodeGen/src/EmitCommonX64.h

@@ -207,6 +207,8 @@ void emitFallback(IrRegAllocX64& regs, AssemblyBuilderX64& build, int offset, in
 
 void emitContinueCallInVm(AssemblyBuilderX64& build);
 
+void emitReturn(AssemblyBuilderX64& build, ModuleHelpers& helpers);
+
 } // namespace X64
 } // namespace CodeGen
 } // namespace Luau

CodeGen/src/EmitInstructionX64.cpp

@@ -166,68 +166,33 @@ void emitInstCall(AssemblyBuilderX64& build, ModuleHelpers& helpers, int ra, int
 void emitInstReturn(AssemblyBuilderX64& build, ModuleHelpers& helpers, int ra, int actualResults)
 {
     RegisterX64 ci = r8;
-    RegisterX64 cip = r9;
     RegisterX64 res = rdi;
-    RegisterX64 nresults = esi;
+    RegisterX64 written = ecx;
 
     build.mov(ci, qword[rState + offsetof(lua_State, ci)]);
-    build.lea(cip, addr[ci - sizeof(CallInfo)]);
-
-    // res = ci->func; note: we assume CALL always puts func+args and expects results to start at func
     build.mov(res, qword[ci + offsetof(CallInfo, func)]);
-    // nresults = ci->nresults
-    build.mov(nresults, dword[ci + offsetof(CallInfo, nresults)]);
-
-    {
-        Label skipResultCopy;
-
-        RegisterX64 counter = ecx;
 
     if (actualResults == 0)
     {
-            // Our instruction doesn't have any results, so just fill results expected in parent with 'nil'
+        build.xor_(written, written);
-            build.test(nresults, nresults);                     // test here will set SF=1 for a negative number, ZF=1 for zero and OF=0
+        build.jmp(helpers.return_);
-            build.jcc(ConditionX64::LessEqual, skipResultCopy); // jle jumps if SF != OF or ZF == 1
-
-            build.mov(counter, nresults);
-
-            Label repeatNilLoop = build.setLabel();
-            build.mov(dword[res + offsetof(TValue, tt)], LUA_TNIL);
-            build.add(res, sizeof(TValue));
-            build.dec(counter);
-            build.jcc(ConditionX64::NotZero, repeatNilLoop);
     }
-        else if (actualResults == 1)
+    else if (actualResults >= 1 && actualResults <= 3)
     {
-            // Try setting our 1 result
+        for (int r = 0; r < actualResults; ++r)
-            build.test(nresults, nresults);
+        {
-            build.jcc(ConditionX64::Zero, skipResultCopy);
+            build.vmovups(xmm0, luauReg(ra + r));
-
+            build.vmovups(xmmword[res + r * sizeof(TValue)], xmm0);
-            build.lea(counter, addr[nresults - 1]);
+        }
-
+        build.add(res, actualResults * sizeof(TValue));
-            build.vmovups(xmm0, luauReg(ra));
+        build.mov(written, actualResults);
-            build.vmovups(xmmword[res], xmm0);
+        build.jmp(helpers.return_);
-            build.add(res, sizeof(TValue));
-
-            // Fill the rest of the expected results with 'nil'
-            build.test(counter, counter);                       // test here will set SF=1 for a negative number, ZF=1 for zero and OF=0
-            build.jcc(ConditionX64::LessEqual, skipResultCopy); // jle jumps if SF != OF or ZF == 1
-
-            Label repeatNilLoop = build.setLabel();
-            build.mov(dword[res + offsetof(TValue, tt)], LUA_TNIL);
-            build.add(res, sizeof(TValue));
-            build.dec(counter);
-            build.jcc(ConditionX64::NotZero, repeatNilLoop);
     }
     else
     {
         RegisterX64 vali = rax;
         RegisterX64 valend = rdx;
 
-            // Copy return values into parent stack (but only up to nresults!)
-            build.test(nresults, nresults);
-            build.jcc(ConditionX64::Zero, skipResultCopy);
-
         // vali = ra
         build.lea(vali, luauRegAddress(ra));
 
@@ -237,89 +202,25 @@ void emitInstReturn(AssemblyBuilderX64& build, ModuleHelpers& helpers, int ra, i
         else
             build.lea(valend, luauRegAddress(ra + actualResults)); // valend = ra + actualResults
 
-            build.mov(counter, nresults);
+        build.xor_(written, written);
 
         Label repeatValueLoop, exitValueLoop;
 
-            build.setLabel(repeatValueLoop);
         build.cmp(vali, valend);
         build.jcc(ConditionX64::NotBelow, exitValueLoop);
 
+        build.setLabel(repeatValueLoop);
         build.vmovups(xmm0, xmmword[vali]);
         build.vmovups(xmmword[res], xmm0);
         build.add(vali, sizeof(TValue));
         build.add(res, sizeof(TValue));
-            build.dec(counter);
+        build.inc(written);
-            build.jcc(ConditionX64::NotZero, repeatValueLoop);
+        build.cmp(vali, valend);
+        build.jcc(ConditionX64::Below, repeatValueLoop);
 
         build.setLabel(exitValueLoop);
-
+        build.jmp(helpers.return_);
-            // Fill the rest of the expected results with 'nil'
-            build.test(counter, counter);                       // test here will set SF=1 for a negative number, ZF=1 for zero and OF=0
-            build.jcc(ConditionX64::LessEqual, skipResultCopy); // jle jumps if SF != OF or ZF == 1
-
-            Label repeatNilLoop = build.setLabel();
-            build.mov(dword[res + offsetof(TValue, tt)], LUA_TNIL);
-            build.add(res, sizeof(TValue));
-            build.dec(counter);
-            build.jcc(ConditionX64::NotZero, repeatNilLoop);
     }
-
-        build.setLabel(skipResultCopy);
-    }
-
-    build.mov(qword[rState + offsetof(lua_State, ci)], cip);     // L->ci = cip
-    build.mov(rBase, qword[cip + offsetof(CallInfo, base)]);     // sync base = L->base while we have a chance
-    build.mov(qword[rState + offsetof(lua_State, base)], rBase); // L->base = cip->base
-
-    // Start with result for LUA_MULTRET/exit value
-    build.mov(qword[rState + offsetof(lua_State, top)], res); // L->top = res
-
-    // Unlikely, but this might be the last return from VM
-    build.test(byte[ci + offsetof(CallInfo, flags)], LUA_CALLINFO_RETURN);
-    build.jcc(ConditionX64::NotZero, helpers.exitNoContinueVm);
-
-    Label skipFixedRetTop;
-    build.test(nresults, nresults);                 // test here will set SF=1 for a negative number and it always sets OF to 0
-    build.jcc(ConditionX64::Less, skipFixedRetTop); // jl jumps if SF != OF
-    build.mov(rax, qword[cip + offsetof(CallInfo, top)]);
-    build.mov(qword[rState + offsetof(lua_State, top)], rax); // L->top = cip->top
-    build.setLabel(skipFixedRetTop);
-
-    // Returning back to the previous function is a bit tricky
-    // Registers alive: r9 (cip)
-    RegisterX64 proto = rcx;
-    RegisterX64 execdata = rbx;
-
-    // Change closure
-    build.mov(rax, qword[cip + offsetof(CallInfo, func)]);
-    build.mov(rax, qword[rax + offsetof(TValue, value.gc)]);
-    build.mov(sClosure, rax);
-
-    build.mov(proto, qword[rax + offsetof(Closure, l.p)]);
-
-    build.mov(execdata, qword[proto + offsetof(Proto, execdata)]);
-
-    build.test(byte[cip + offsetof(CallInfo, flags)], LUA_CALLINFO_NATIVE);
-    build.jcc(ConditionX64::Zero, helpers.exitContinueVm); // Continue in interpreter if function has no native data
-
-    // Change constants
-    build.mov(rConstants, qword[proto + offsetof(Proto, k)]);
-
-    // Change code
-    build.mov(rdx, qword[proto + offsetof(Proto, code)]);
-    build.mov(sCode, rdx);
-
-    build.mov(rax, qword[cip + offsetof(CallInfo, savedpc)]);
-
-    // To get instruction index from instruction pointer, we need to divide byte offset by 4
-    // But we will actually need to scale instruction index by 4 back to byte offset later so it cancels out
-    build.sub(rax, rdx);
-
-    // Get new instruction location and jump to it
-    build.mov(edx, dword[execdata + rax]);
-    build.add(rdx, qword[proto + offsetof(Proto, exectarget)]);
-    build.jmp(rdx);
 }
 
 void emitInstSetList(IrRegAllocX64& regs, AssemblyBuilderX64& build, int ra, int rb, int count, uint32_t index)

CodeGen/src/IrAnalysis.cpp

@@ -661,9 +661,212 @@ static void computeCfgBlockEdges(IrFunction& function)
     }
 }
 
+// Assign tree depth and pre- and post- DFS visit order of the tree/graph nodes
+// Optionally, collect required node order into a vector
+template<auto childIt>
+void computeBlockOrdering(
+    IrFunction& function, std::vector<BlockOrdering>& ordering, std::vector<uint32_t>* preOrder, std::vector<uint32_t>* postOrder)
+{
+    CfgInfo& info = function.cfg;
+
+    LUAU_ASSERT(info.idoms.size() == function.blocks.size());
+
+    ordering.clear();
+    ordering.resize(function.blocks.size());
+
+    // Get depth-first post-order using manual stack instead of recursion
+    struct StackItem
+    {
+        uint32_t blockIdx;
+        uint32_t itPos;
+    };
+    std::vector<StackItem> stack;
+
+    if (preOrder)
+        preOrder->reserve(function.blocks.size());
+    if (postOrder)
+        postOrder->reserve(function.blocks.size());
+
+    uint32_t nextPreOrder = 0;
+    uint32_t nextPostOrder = 0;
+
+    stack.push_back({0, 0});
+    ordering[0].visited = true;
+    ordering[0].preOrder = nextPreOrder++;
+
+    while (!stack.empty())
+    {
+        StackItem& item = stack.back();
+        BlockIteratorWrapper children = childIt(info, item.blockIdx);
+
+        if (item.itPos < children.size())
+        {
+            uint32_t childIdx = children[item.itPos++];
+
+            BlockOrdering& childOrdering = ordering[childIdx];
+
+            if (!childOrdering.visited)
+            {
+                childOrdering.visited = true;
+                childOrdering.depth = uint32_t(stack.size());
+                childOrdering.preOrder = nextPreOrder++;
+
+                if (preOrder)
+                    preOrder->push_back(item.blockIdx);
+
+                stack.push_back({childIdx, 0});
+            }
+        }
+        else
+        {
+            ordering[item.blockIdx].postOrder = nextPostOrder++;
+
+            if (postOrder)
+                postOrder->push_back(item.blockIdx);
+
+            stack.pop_back();
+        }
+    }
+}
+
+// Dominance tree construction based on 'A Simple, Fast Dominance Algorithm' [Keith D. Cooper, et al]
+// This solution has quadratic complexity in the worst case.
+// It is possible to switch to SEMI-NCA algorithm (also quadratic) mentioned in 'Linear-Time Algorithms for Dominators and Related Problems' [Loukas
+// Georgiadis]
+
+// Find block that is common between blocks 'a' and 'b' on the path towards the entry
+static uint32_t findCommonDominator(const std::vector<uint32_t>& idoms, const std::vector<BlockOrdering>& data, uint32_t a, uint32_t b)
+{
+    while (a != b)
+    {
+        while (data[a].postOrder < data[b].postOrder)
+        {
+            a = idoms[a];
+            LUAU_ASSERT(a != ~0u);
+        }
+
+        while (data[b].postOrder < data[a].postOrder)
+        {
+            b = idoms[b];
+            LUAU_ASSERT(b != ~0u);
+        }
+    }
+
+    return a;
+}
+
+void computeCfgImmediateDominators(IrFunction& function)
+{
+    CfgInfo& info = function.cfg;
+
+    // Clear existing data
+    info.idoms.clear();
+    info.idoms.resize(function.blocks.size(), ~0u);
+
+    std::vector<BlockOrdering> ordering;
+    std::vector<uint32_t> blocksInPostOrder;
+    computeBlockOrdering<successors>(function, ordering, /* preOrder */ nullptr, &blocksInPostOrder);
+
+    // Entry node is temporarily marked to be an idom of itself to make algorithm work
+    info.idoms[0] = 0;
+
+    // Iteratively compute immediate dominators
+    bool updated = true;
+
+    while (updated)
+    {
+        updated = false;
+
+        // Go over blocks in reverse post-order of CFG
+        // '- 2' skips the root node which is last in post-order traversal
+        for (int i = int(blocksInPostOrder.size() - 2); i >= 0; i--)
+        {
+            uint32_t blockIdx = blocksInPostOrder[i];
+            uint32_t newIdom = ~0u;
+
+            for (uint32_t predIdx : predecessors(info, blockIdx))
+            {
+                if (uint32_t predIdom = info.idoms[predIdx]; predIdom != ~0u)
+                {
+                    if (newIdom == ~0u)
+                        newIdom = predIdx;
+                    else
+                        newIdom = findCommonDominator(info.idoms, ordering, newIdom, predIdx);
+                }
+            }
+
+            if (newIdom != info.idoms[blockIdx])
+            {
+                info.idoms[blockIdx] = newIdom;
+
+                // Run until a fixed point is reached
+                updated = true;
+            }
+        }
+    }
+
+    // Entry node doesn't have an immediate dominator
+    info.idoms[0] = ~0u;
+}
+
+void computeCfgDominanceTreeChildren(IrFunction& function)
+{
+    CfgInfo& info = function.cfg;
+
+    // Clear existing data
+    info.domChildren.clear();
+
+    info.domChildrenOffsets.clear();
+    info.domChildrenOffsets.resize(function.blocks.size());
+
+    // First we need to know children count of each node in the dominance tree
+    // We use offset array for to hold this data, counts will be readjusted to offsets later
+    for (size_t blockIdx = 0; blockIdx < function.blocks.size(); blockIdx++)
+    {
+        uint32_t domParent = info.idoms[blockIdx];
+
+        if (domParent != ~0u)
+            info.domChildrenOffsets[domParent]++;
+    }
+
+    // Convert counds to offsets using prefix sum
+    uint32_t total = 0;
+
+    for (size_t blockIdx = 0; blockIdx < function.blocks.size(); blockIdx++)
+    {
+        uint32_t& offset = info.domChildrenOffsets[blockIdx];
+        uint32_t count = offset;
+        offset = total;
+        total += count;
+    }
+
+    info.domChildren.resize(total);
+
+    for (size_t blockIdx = 0; blockIdx < function.blocks.size(); blockIdx++)
+    {
+        // We use a trick here, where we use the starting offset of the dominance children list as the position where to write next child
+        // The values will be adjusted back in a separate loop later
+        uint32_t domParent = info.idoms[blockIdx];
+
+        if (domParent != ~0u)
+            info.domChildren[info.domChildrenOffsets[domParent]++] = uint32_t(blockIdx);
+    }
+
+    // Offsets into the dominance children list were used as iterators in the previous loop
+    // That process basically moved the values in the array 1 step towards the start
+    // Here we move them one step towards the end and restore 0 for first offset
+    for (int blockIdx = int(function.blocks.size() - 1); blockIdx > 0; blockIdx--)
+        info.domChildrenOffsets[blockIdx] = info.domChildrenOffsets[blockIdx - 1];
+    info.domChildrenOffsets[0] = 0;
+
+    computeBlockOrdering<domChildren>(function, info.domOrdering, /* preOrder */ nullptr, /* postOrder */ nullptr);
+}
+
 void computeCfgInfo(IrFunction& function)
 {
     computeCfgBlockEdges(function);
+    computeCfgImmediateDominators(function);
+    computeCfgDominanceTreeChildren(function);
     computeCfgLiveInOutRegSets(function);
 }
 
@@ -687,5 +890,15 @@ BlockIteratorWrapper successors(const CfgInfo& cfg, uint32_t blockIdx)
     return BlockIteratorWrapper{cfg.successors.data() + start, cfg.successors.data() + end};
 }
 
+BlockIteratorWrapper domChildren(const CfgInfo& cfg, uint32_t blockIdx)
+{
+    LUAU_ASSERT(blockIdx < cfg.domChildrenOffsets.size());
+
+    uint32_t start = cfg.domChildrenOffsets[blockIdx];
+    uint32_t end = blockIdx + 1 < cfg.domChildrenOffsets.size() ? cfg.domChildrenOffsets[blockIdx + 1] : uint32_t(cfg.domChildren.size());
+
+    return BlockIteratorWrapper{cfg.domChildren.data() + start, cfg.domChildren.data() + end};
+}
+
 } // namespace CodeGen
 } // namespace Luau

CodeGen/src/IrBuilder.cpp

@@ -25,6 +25,7 @@ IrBuilder::IrBuilder()
 void IrBuilder::buildFunctionIr(Proto* proto)
 {
     function.proto = proto;
+    function.variadic = proto->is_vararg != 0;
 
     // Rebuild original control flow blocks
     rebuildBytecodeBasicBlocks(proto);

CodeGen/src/IrLoweringA64.cpp

@@ -185,11 +185,10 @@ static bool emitBuiltin(
     }
 }
 
-IrLoweringA64::IrLoweringA64(AssemblyBuilderA64& build, ModuleHelpers& helpers, NativeState& data, Proto* proto, IrFunction& function)
+IrLoweringA64::IrLoweringA64(AssemblyBuilderA64& build, ModuleHelpers& helpers, NativeState& data, IrFunction& function)
     : build(build)
     , helpers(helpers)
     , data(data)
-    , proto(proto)
     , function(function)
     , regs(function, {{x0, x15}, {x16, x17}, {q0, q7}, {q16, q31}})
     , valueTracker(function)
@@ -1343,19 +1342,71 @@ void IrLoweringA64::lowerInst(IrInst& inst, uint32_t index, IrBlock& next)
         break;
     case IrCmd::RETURN:
         regs.spill(build, index);
+
+        if (function.variadic)
+        {
+            build.ldr(x1, mem(rState, offsetof(lua_State, ci)));
+            build.ldr(x1, mem(x1, offsetof(CallInfo, func)));
+        }
+        else if (intOp(inst.b) != 1)
+            build.sub(x1, rBase, sizeof(TValue)); // invariant: ci->func + 1 == ci->base for non-variadic frames
+
+        if (intOp(inst.b) == 0)
+        {
+            build.mov(w2, 0);
+            build.b(helpers.return_);
+        }
+        else if (intOp(inst.b) == 1 && !function.variadic)
+        {
+            // fast path: minimizes x1 adjustments
+            // note that we skipped x1 computation for this specific case above
+            build.ldr(q0, mem(rBase, vmRegOp(inst.a) * sizeof(TValue)));
+            build.str(q0, mem(rBase, -int(sizeof(TValue))));
+            build.mov(x1, rBase);
+            build.mov(w2, 1);
+            build.b(helpers.return_);
+        }
+        else if (intOp(inst.b) >= 1 && intOp(inst.b) <= 3)
+        {
+            for (int r = 0; r < intOp(inst.b); ++r)
+            {
+                build.ldr(q0, mem(rBase, (vmRegOp(inst.a) + r) * sizeof(TValue)));
+                build.str(q0, mem(x1, sizeof(TValue), AddressKindA64::post));
+            }
+            build.mov(w2, intOp(inst.b));
+            build.b(helpers.return_);
+        }
+        else
+        {
+            build.mov(w2, 0);
+
+            // vali = ra
+            build.add(x3, rBase, uint16_t(vmRegOp(inst.a) * sizeof(TValue)));
+
             // valend = (n == LUA_MULTRET) ? L->top : ra + n
             if (intOp(inst.b) == LUA_MULTRET)
-            build.ldr(x2, mem(rState, offsetof(lua_State, top)));
+                build.ldr(x4, mem(rState, offsetof(lua_State, top)));
             else
-            build.add(x2, rBase, uint16_t((vmRegOp(inst.a) + intOp(inst.b)) * sizeof(TValue)));
+                build.add(x4, rBase, uint16_t((vmRegOp(inst.a) + intOp(inst.b)) * sizeof(TValue)));
-        // returnFallback(L, ra, valend)
-        build.mov(x0, rState);
-        build.add(x1, rBase, uint16_t(vmRegOp(inst.a) * sizeof(TValue)));
-        build.ldr(x3, mem(rNativeContext, offsetof(NativeContext, returnFallback)));
-        build.blr(x3);
 
-        // reentry with x0=closure (NULL will trigger exit)
+            Label repeatValueLoop, exitValueLoop;
-        build.b(helpers.reentry);
+
+            if (intOp(inst.b) == LUA_MULTRET)
+            {
+                build.cmp(x3, x4);
+                build.b(ConditionA64::CarrySet, exitValueLoop); // CarrySet == UnsignedGreaterEqual
+            }
+
+            build.setLabel(repeatValueLoop);
+            build.ldr(q0, mem(x3, sizeof(TValue), AddressKindA64::post));
+            build.str(q0, mem(x1, sizeof(TValue), AddressKindA64::post));
+            build.add(w2, w2, 1);
+            build.cmp(x3, x4);
+            build.b(ConditionA64::CarryClear, repeatValueLoop); // CarryClear == UnsignedLess
+
+            build.setLabel(exitValueLoop);
+            build.b(helpers.return_);
+        }
         break;
     case IrCmd::FORGLOOP:
         // register layout: ra + 1 = table, ra + 2 = internal index, ra + 3 .. ra + aux = iteration variables

CodeGen/src/IrLoweringA64.h

@@ -9,8 +9,6 @@
 
 #include <vector>
 
-struct Proto;
-
 namespace Luau
 {
 namespace CodeGen
@@ -25,7 +23,7 @@ namespace A64
 
 struct IrLoweringA64
 {
-    IrLoweringA64(AssemblyBuilderA64& build, ModuleHelpers& helpers, NativeState& data, Proto* proto, IrFunction& function);
+    IrLoweringA64(AssemblyBuilderA64& build, ModuleHelpers& helpers, NativeState& data, IrFunction& function);
 
     void lowerInst(IrInst& inst, uint32_t index, IrBlock& next);
     void finishBlock();
@@ -58,7 +56,6 @@ struct IrLoweringA64
     AssemblyBuilderA64& build;
     ModuleHelpers& helpers;
     NativeState& data;
-    Proto* proto = nullptr; // Temporarily required to provide 'Instruction* pc' to old emitInst* methods
 
     IrFunction& function;
 

CodeGen/src/NativeState.cpp

@@ -90,7 +90,6 @@ void initFunctions(NativeState& data)
     data.context.callEpilogC = callEpilogC;
 
     data.context.callFallback = callFallback;
-    data.context.returnFallback = returnFallback;
 
     data.context.executeGETGLOBAL = executeGETGLOBAL;
     data.context.executeSETGLOBAL = executeSETGLOBAL;

CodeGen/src/NativeState.h

@@ -86,7 +86,6 @@ struct NativeContext
     void (*callEpilogC)(lua_State* L, int nresults, int n) = nullptr;
 
     Closure* (*callFallback)(lua_State* L, StkId ra, StkId argtop, int nresults) = nullptr;
-    Closure* (*returnFallback)(lua_State* L, StkId ra, StkId valend) = nullptr;
 
     // Opcode fallbacks, implemented in C
     const Instruction* (*executeGETGLOBAL)(lua_State* L, const Instruction* pc, StkId base, TValue* k) = nullptr;

Makefile

@@ -44,6 +44,10 @@ ANALYZE_CLI_SOURCES=CLI/FileUtils.cpp CLI/Flags.cpp CLI/Analyze.cpp
 ANALYZE_CLI_OBJECTS=$(ANALYZE_CLI_SOURCES:%=$(BUILD)/%.o)
 ANALYZE_CLI_TARGET=$(BUILD)/luau-analyze
 
+COMPILE_CLI_SOURCES=CLI/FileUtils.cpp CLI/Flags.cpp CLI/Compile.cpp
+COMPILE_CLI_OBJECTS=$(COMPILE_CLI_SOURCES:%=$(BUILD)/%.o)
+COMPILE_CLI_TARGET=$(BUILD)/luau-compile
+
 FUZZ_SOURCES=$(wildcard fuzz/*.cpp) fuzz/luau.pb.cpp
 FUZZ_OBJECTS=$(FUZZ_SOURCES:%=$(BUILD)/%.o)
 
@@ -55,8 +59,8 @@ ifneq ($(opt),)
 	TESTS_ARGS+=-O$(opt)
 endif
 
-OBJECTS=$(AST_OBJECTS) $(COMPILER_OBJECTS) $(ANALYSIS_OBJECTS) $(CODEGEN_OBJECTS) $(VM_OBJECTS) $(ISOCLINE_OBJECTS) $(TESTS_OBJECTS) $(REPL_CLI_OBJECTS) $(ANALYZE_CLI_OBJECTS) $(FUZZ_OBJECTS)
+OBJECTS=$(AST_OBJECTS) $(COMPILER_OBJECTS) $(ANALYSIS_OBJECTS) $(CODEGEN_OBJECTS) $(VM_OBJECTS) $(ISOCLINE_OBJECTS) $(TESTS_OBJECTS) $(REPL_CLI_OBJECTS) $(ANALYZE_CLI_OBJECTS) $(COMPILE_CLI_OBJECTS) $(FUZZ_OBJECTS)
-EXECUTABLE_ALIASES = luau luau-analyze luau-tests
+EXECUTABLE_ALIASES = luau luau-analyze luau-compile luau-tests
 
 # common flags
 CXXFLAGS=-g -Wall
@@ -132,6 +136,7 @@ $(ISOCLINE_OBJECTS): CXXFLAGS+=-Wno-unused-function -Iextern/isocline/include
 $(TESTS_OBJECTS): CXXFLAGS+=-std=c++17 -ICommon/include -IAst/include -ICompiler/include -IAnalysis/include -ICodeGen/include -IVM/include -ICLI -Iextern -DDOCTEST_CONFIG_DOUBLE_STRINGIFY
 $(REPL_CLI_OBJECTS): CXXFLAGS+=-std=c++17 -ICommon/include -IAst/include -ICompiler/include -IVM/include -ICodeGen/include -Iextern -Iextern/isocline/include
 $(ANALYZE_CLI_OBJECTS): CXXFLAGS+=-std=c++17 -ICommon/include -IAst/include -IAnalysis/include -Iextern
+$(COMPILE_CLI_OBJECTS): CXXFLAGS+=-std=c++17 -ICommon/include -IAst/include -ICompiler/include -IVM/include -ICodeGen/include
 $(FUZZ_OBJECTS): CXXFLAGS+=-std=c++17 -ICommon/include -IAst/include -ICompiler/include -IAnalysis/include -IVM/include -ICodeGen/include
 
 $(TESTS_TARGET): LDFLAGS+=-lpthread
@@ -189,6 +194,9 @@ luau: $(REPL_CLI_TARGET)
 luau-analyze: $(ANALYZE_CLI_TARGET)
 	ln -fs $^ $@
 
+luau-compile: $(COMPILE_CLI_TARGET)
+	ln -fs $^ $@
+
 luau-tests: $(TESTS_TARGET)
 	ln -fs $^ $@
 
@@ -196,8 +204,9 @@ luau-tests: $(TESTS_TARGET)
 $(TESTS_TARGET): $(TESTS_OBJECTS) $(ANALYSIS_TARGET) $(COMPILER_TARGET) $(AST_TARGET) $(CODEGEN_TARGET) $(VM_TARGET) $(ISOCLINE_TARGET)
 $(REPL_CLI_TARGET): $(REPL_CLI_OBJECTS) $(COMPILER_TARGET) $(AST_TARGET) $(CODEGEN_TARGET) $(VM_TARGET) $(ISOCLINE_TARGET)
 $(ANALYZE_CLI_TARGET): $(ANALYZE_CLI_OBJECTS) $(ANALYSIS_TARGET) $(AST_TARGET)
+$(COMPILE_CLI_TARGET): $(COMPILE_CLI_OBJECTS) $(COMPILER_TARGET) $(AST_TARGET) $(CODEGEN_TARGET) $(VM_TARGET)
 
-$(TESTS_TARGET) $(REPL_CLI_TARGET) $(ANALYZE_CLI_TARGET):
+$(TESTS_TARGET) $(REPL_CLI_TARGET) $(ANALYZE_CLI_TARGET) $(COMPILE_CLI_TARGET):
 	$(CXX) $^ $(LDFLAGS) -o $@
 
 # executable targets for fuzzing

Sources.cmake

@@ -141,7 +141,6 @@ target_sources(Luau.Analysis PRIVATE
     Analysis/include/Luau/BuiltinDefinitions.h
     Analysis/include/Luau/Clone.h
     Analysis/include/Luau/Config.h
-    Analysis/include/Luau/Refinement.h
     Analysis/include/Luau/Constraint.h
     Analysis/include/Luau/ConstraintGraphBuilder.h
     Analysis/include/Luau/ConstraintSolver.h
@@ -153,6 +152,7 @@ target_sources(Luau.Analysis PRIVATE
     Analysis/include/Luau/Error.h
     Analysis/include/Luau/FileResolver.h
     Analysis/include/Luau/Frontend.h
+    Analysis/include/Luau/InsertionOrderedMap.h
     Analysis/include/Luau/Instantiation.h
     Analysis/include/Luau/IostreamHelpers.h
     Analysis/include/Luau/JsonEmitter.h
@@ -165,6 +165,7 @@ target_sources(Luau.Analysis PRIVATE
     Analysis/include/Luau/Predicate.h
     Analysis/include/Luau/Quantify.h
     Analysis/include/Luau/RecursionCounter.h
+    Analysis/include/Luau/Refinement.h
     Analysis/include/Luau/RequireTracer.h
     Analysis/include/Luau/Scope.h
     Analysis/include/Luau/Simplify.h
@@ -175,6 +176,7 @@ target_sources(Luau.Analysis PRIVATE
     Analysis/include/Luau/ToString.h
     Analysis/include/Luau/Transpiler.h
     Analysis/include/Luau/TxnLog.h
+    Analysis/include/Luau/Type.h
     Analysis/include/Luau/TypeArena.h
     Analysis/include/Luau/TypeAttach.h
     Analysis/include/Luau/TypeChecker2.h
@@ -183,7 +185,6 @@ target_sources(Luau.Analysis PRIVATE
     Analysis/include/Luau/TypeInfer.h
     Analysis/include/Luau/TypePack.h
     Analysis/include/Luau/TypeUtils.h
-    Analysis/include/Luau/Type.h
     Analysis/include/Luau/Unifiable.h
     Analysis/include/Luau/Unifier.h
     Analysis/include/Luau/UnifierSharedState.h
@@ -198,7 +199,6 @@ target_sources(Luau.Analysis PRIVATE
     Analysis/src/BuiltinDefinitions.cpp
     Analysis/src/Clone.cpp
     Analysis/src/Config.cpp
-    Analysis/src/Refinement.cpp
     Analysis/src/Constraint.cpp
     Analysis/src/ConstraintGraphBuilder.cpp
     Analysis/src/ConstraintSolver.cpp
@@ -216,6 +216,7 @@ target_sources(Luau.Analysis PRIVATE
     Analysis/src/Module.cpp
     Analysis/src/Normalize.cpp
     Analysis/src/Quantify.cpp
+    Analysis/src/Refinement.cpp
     Analysis/src/RequireTracer.cpp
     Analysis/src/Scope.cpp
     Analysis/src/Simplify.cpp
@@ -226,6 +227,7 @@ target_sources(Luau.Analysis PRIVATE
     Analysis/src/ToString.cpp
     Analysis/src/Transpiler.cpp
     Analysis/src/TxnLog.cpp
+    Analysis/src/Type.cpp
     Analysis/src/TypeArena.cpp
     Analysis/src/TypeAttach.cpp
     Analysis/src/TypeChecker2.cpp
@@ -234,7 +236,6 @@ target_sources(Luau.Analysis PRIVATE
     Analysis/src/TypeInfer.cpp
     Analysis/src/TypePack.cpp
     Analysis/src/TypeUtils.cpp
-    Analysis/src/Type.cpp
     Analysis/src/Unifiable.cpp
     Analysis/src/Unifier.cpp
 )
@@ -326,6 +327,7 @@ if(TARGET Luau.Analyze.CLI)
 endif()
 
 if(TARGET Luau.Ast.CLI)
+    # Luau.Ast.CLI Sources
     target_sources(Luau.Ast.CLI PRIVATE
         CLI/Ast.cpp
         CLI/FileUtils.h
@@ -415,6 +417,7 @@ if(TARGET Luau.UnitTest)
         tests/TypeVar.test.cpp
         tests/Variant.test.cpp
         tests/VisitType.test.cpp
+        tests/InsertionOrderedMap.test.cpp
         tests/main.cpp)
 endif()
 
@@ -449,9 +452,20 @@ if(TARGET Luau.Web)
 endif()
 
 if(TARGET Luau.Reduce.CLI)
+    # Luau.Reduce.CLI Sources
     target_sources(Luau.Reduce.CLI PRIVATE
         CLI/Reduce.cpp
         CLI/FileUtils.cpp
         CLI/FileUtils.h
     )
 endif()
+
+if(TARGET Luau.Compile.CLI)
+    # Luau.Compile.CLI Sources
+    target_sources(Luau.Compile.CLI PRIVATE
+        CLI/FileUtils.h
+        CLI/FileUtils.cpp
+        CLI/Flags.h
+        CLI/Flags.cpp
+        CLI/Compile.cpp)
+endif()

VM/src/ldo.cpp

@@ -17,8 +17,6 @@
 
 #include <string.h>
 
-LUAU_FASTFLAGVARIABLE(LuauUniformTopHandling, false)
-
 /*
 ** {======================================================
 ** Error-recovery functions
@@ -252,7 +250,7 @@ void luaD_call(lua_State* L, StkId func, int nresults)
             L->isactive = false;
     }
 
-    if (FFlag::LuauUniformTopHandling && nresults != LUA_MULTRET)
+    if (nresults != LUA_MULTRET)
         L->top = restorestack(L, old_func) + nresults;
 
     L->nCcalls--;

VM/src/lvmexecute.cpp

@@ -16,7 +16,6 @@
 
 #include <string.h>
 
-LUAU_FASTFLAG(LuauUniformTopHandling)
 LUAU_FASTFLAG(LuauGetImportDirect)
 
 // Disable c99-designator to avoid the warning in CGOTO dispatch table
@@ -1043,8 +1042,6 @@ reentry:
                 // we're done!
                 if (LUAU_UNLIKELY(ci->flags & LUA_CALLINFO_RETURN))
                 {
-                    if (!FFlag::LuauUniformTopHandling)
-                        L->top = res;
                     goto exit;
                 }
 

tests/AssemblyBuilderA64.test.cpp

@@ -460,6 +460,25 @@ TEST_CASE_FIXTURE(AssemblyBuilderA64Fixture, "Undefined")
     SINGLE_COMPARE(udf(), 0x00000000);
 }
 
+TEST_CASE_FIXTURE(AssemblyBuilderA64Fixture, "PrePostIndexing")
+{
+    SINGLE_COMPARE(ldr(x0, mem(x1, 1)), 0xF8401020);
+    SINGLE_COMPARE(ldr(x0, mem(x1, 1, AddressKindA64::pre)), 0xF8401C20);
+    SINGLE_COMPARE(ldr(x0, mem(x1, 1, AddressKindA64::post)), 0xF8401420);
+
+    SINGLE_COMPARE(ldr(q0, mem(x1, 1)), 0x3CC01020);
+    SINGLE_COMPARE(ldr(q0, mem(x1, 1, AddressKindA64::pre)), 0x3CC01C20);
+    SINGLE_COMPARE(ldr(q0, mem(x1, 1, AddressKindA64::post)), 0x3CC01420);
+
+    SINGLE_COMPARE(str(x0, mem(x1, 1)), 0xF8001020);
+    SINGLE_COMPARE(str(x0, mem(x1, 1, AddressKindA64::pre)), 0xF8001C20);
+    SINGLE_COMPARE(str(x0, mem(x1, 1, AddressKindA64::post)), 0xF8001420);
+
+    SINGLE_COMPARE(str(q0, mem(x1, 1)), 0x3C801020);
+    SINGLE_COMPARE(str(q0, mem(x1, 1, AddressKindA64::pre)), 0x3C801C20);
+    SINGLE_COMPARE(str(q0, mem(x1, 1, AddressKindA64::post)), 0x3C801420);
+}
+
 TEST_CASE("LogTest")
 {
     AssemblyBuilderA64 build(/* logText= */ true);
@@ -501,6 +520,10 @@ TEST_CASE("LogTest")
 
     build.ubfx(x1, x2, 37, 5);
 
+    build.ldr(x0, mem(x1, 1));
+    build.ldr(x0, mem(x1, 1, AddressKindA64::pre));
+    build.ldr(x0, mem(x1, 1, AddressKindA64::post));
+
     build.setLabel(l);
     build.ret();
 
@@ -534,6 +557,9 @@ TEST_CASE("LogTest")
  tbz         x0,#5,.L1
  fcvt        s1,d2
  ubfx        x1,x2,#3705
+ ldr         x0,[x1,#1]
+ ldr         x0,[x1,#1]!
+ ldr         x0,[x1]!,#1
 .L1:
  ret
 )";

tests/Autocomplete.test.cpp

@@ -3471,8 +3471,6 @@ local a: T@1
 
 TEST_CASE_FIXTURE(ACFixture, "frontend_use_correct_global_scope")
 {
-    ScopedFastFlag sff("LuauTypeCheckerUseCorrectScope", true);
-
     loadDefinition(R"(
         declare class Instance
             Name: string

tests/InsertionOrderedMap.test.cpp

@@ -0,0 +1,140 @@
+// This file is part of the Luau programming language and is licensed under MIT License; see LICENSE.txt for details
+#include "Luau/InsertionOrderedMap.h"
+
+#include <memory>
+
+#include "doctest.h"
+
+using namespace Luau;
+
+struct MapFixture
+{
+    std::vector<std::unique_ptr<int>> ptrs;
+
+    int* makePtr()
+    {
+        ptrs.push_back(std::make_unique<int>(int{}));
+        return ptrs.back().get();
+    }
+};
+
+TEST_SUITE_BEGIN("InsertionOrderedMap");
+
+TEST_CASE_FIXTURE(MapFixture, "map_insertion")
+{
+    InsertionOrderedMap<int*, int> map;
+
+    int* a = makePtr();
+    int* b = makePtr();
+
+    map.insert(a, 1);
+    map.insert(b, 2);
+}
+
+TEST_CASE_FIXTURE(MapFixture, "map_lookup")
+{
+    InsertionOrderedMap<int*, int> map;
+
+    int* a = makePtr();
+    map.insert(a, 1);
+
+    int* r = map.get(a);
+    REQUIRE(r != nullptr);
+    CHECK(*r == 1);
+
+    r = map.get(makePtr());
+    CHECK(r == nullptr);
+}
+
+TEST_CASE_FIXTURE(MapFixture, "insert_does_not_update")
+{
+    InsertionOrderedMap<int*, int> map;
+
+    int* k = makePtr();
+    map.insert(k, 1);
+    map.insert(k, 2);
+
+    int* v = map.get(k);
+    REQUIRE(v != nullptr);
+    CHECK(*v == 1);
+}
+
+TEST_CASE_FIXTURE(MapFixture, "insertion_order_is_iteration_order")
+{
+    // This one is a little hard to prove, in that if the ordering guarantees
+    // fail this test isn't guaranteed to fail, but it is strictly better than
+    // nothing.
+
+    InsertionOrderedMap<int*, int> map;
+    int* a = makePtr();
+    int* b = makePtr();
+    int* c = makePtr();
+    map.insert(a, 1);
+    map.insert(b, 1);
+    map.insert(c, 1);
+
+    auto it = map.begin();
+    REQUIRE(it != map.end());
+    CHECK(it->first == a);
+    CHECK(it->second == 1);
+
+    ++it;
+    REQUIRE(it != map.end());
+    CHECK(it->first == b);
+    CHECK(it->second == 1);
+
+    ++it;
+    REQUIRE(it != map.end());
+    CHECK(it->first == c);
+    CHECK(it->second == 1);
+
+    ++it;
+    CHECK(it == map.end());
+}
+
+TEST_CASE_FIXTURE(MapFixture, "destructuring_iterator_compiles")
+{
+    // This test's only purpose is to successfully compile.
+    InsertionOrderedMap<int*, int> map;
+
+    for (auto [k, v] : map)
+    {
+        // Checks here solely to silence unused variable warnings.
+        CHECK(k);
+        CHECK(v > 0);
+    }
+}
+
+TEST_CASE_FIXTURE(MapFixture, "map_erasure")
+{
+    InsertionOrderedMap<int*, int> map;
+
+    int* a = makePtr();
+    int* b = makePtr();
+
+    map.insert(a, 1);
+    map.insert(b, 2);
+
+    map.erase(map.find(a));
+    CHECK(map.size() == 1);
+    CHECK(!map.contains(a));
+    CHECK(map.get(a) == nullptr);
+
+    int* v = map.get(b);
+    REQUIRE(v);
+}
+
+TEST_CASE_FIXTURE(MapFixture, "map_clear")
+{
+    InsertionOrderedMap<int*, int> map;
+    int* a = makePtr();
+
+    map.insert(a, 1);
+
+    map.clear();
+    CHECK(map.size() == 0);
+    CHECK(!map.contains(a));
+    CHECK(map.get(a) == nullptr);
+}
+
+TEST_SUITE_END();

tests/IrBuilder.test.cpp

@@ -74,6 +74,35 @@ public:
         CHECK(target.f == inst.f);
     }
 
+    void defineCfgTree(const std::vector<std::vector<uint32_t>>& successorSets)
+    {
+        for (const std::vector<uint32_t>& successorSet : successorSets)
+        {
+            build.beginBlock(build.block(IrBlockKind::Internal));
+
+            build.function.cfg.successorsOffsets.push_back(uint32_t(build.function.cfg.successors.size()));
+            build.function.cfg.successors.insert(build.function.cfg.successors.end(), successorSet.begin(), successorSet.end());
+        }
+
+        // Brute-force the predecessor list
+        for (int i = 0; i < int(build.function.blocks.size()); i++)
+        {
+            build.function.cfg.predecessorsOffsets.push_back(uint32_t(build.function.cfg.predecessors.size()));
+
+            for (int k = 0; k < int(build.function.blocks.size()); k++)
+            {
+                for (uint32_t succIdx : successors(build.function.cfg, k))
+                {
+                    if (succIdx == i)
+                        build.function.cfg.predecessors.push_back(k);
+                }
+            }
+        }
+
+        computeCfgImmediateDominators(build.function);
+        computeCfgDominanceTreeChildren(build.function);
+    }
+
     IrBuilder build;
 
     // Luau.VM headers are not accessible
@@ -2164,6 +2193,30 @@ bb_0:
 )");
 }
 
+// 'A Simple, Fast Dominance Algorithm' [Keith D. Cooper, et al]. Figure 2.
+TEST_CASE_FIXTURE(IrBuilderFixture, "DominanceVerification1")
+{
+    defineCfgTree({{1, 2}, {3}, {4}, {4}, {3}});
+
+    CHECK(build.function.cfg.idoms == std::vector<uint32_t>{{~0u, 0, 0, 0, 0}});
+}
+
+// 'A Linear Time Algorithm for Placing Phi-Nodes' [Vugranam C.Sreedhar]. Figure 1.
+TEST_CASE_FIXTURE(IrBuilderFixture, "DominanceVerification2")
+{
+    defineCfgTree({{1, 16}, {2, 3, 4}, {4, 7}, {9}, {5}, {6}, {2, 8}, {8}, {7, 15}, {10, 11}, {12}, {12}, {13}, {3, 14, 15}, {12}, {16}, {}});
+
+    CHECK(build.function.cfg.idoms == std::vector<uint32_t>{~0u, 0, 1, 1, 1, 4, 5, 1, 1, 3, 9, 9, 9, 12, 13, 1, 0});
+}
+
+// 'A Linear Time Algorithm for Placing Phi-Nodes' [Vugranam C.Sreedhar]. Figure 4.
+TEST_CASE_FIXTURE(IrBuilderFixture, "DominanceVerification3")
+{
+    defineCfgTree({{1, 2}, {3}, {3, 4}, {5}, {5, 6}, {7}, {7}, {}});
+
+    CHECK(build.function.cfg.idoms == std::vector<uint32_t>{~0u, 0, 0, 0, 2, 0, 4, 0});
+}
+
 TEST_SUITE_END();
 
 TEST_SUITE_BEGIN("ValueNumbering");

tests/Normalize.test.cpp

@@ -791,14 +791,21 @@ TEST_CASE_FIXTURE(NormalizeFixture, "normalize_blocked_types")
     CHECK_EQ(normalizer.typeFromNormal(*norm), &blocked);
 }
 
-TEST_CASE_FIXTURE(NormalizeFixture, "normalize_pending_expansion_types")
+TEST_CASE_FIXTURE(NormalizeFixture, "normalize_is_exactly_number")
 {
-    AstName name;
+    const NormalizedType* number = normalizer.normalize(builtinTypes->numberType);
-    Type pending{PendingExpansionType{std::nullopt, name, {}, {}}};
+    // 1. all types for which Types::number say true for, NormalizedType::isExactlyNumber should say true as well
+    CHECK(Luau::isNumber(builtinTypes->numberType) == number->isExactlyNumber());
+    // 2. isExactlyNumber should handle cases like `number & number`
+    TypeId intersection = arena.addType(IntersectionType{{builtinTypes->numberType, builtinTypes->numberType}});
+    const NormalizedType* normIntersection = normalizer.normalize(intersection);
+    CHECK(normIntersection->isExactlyNumber());
 
-    const NormalizedType* norm = normalizer.normalize(&pending);
+    // 3. isExactlyNumber should reject things that are definitely not precisely numbers `number | any`
 
-    CHECK_EQ(normalizer.typeFromNormal(*norm), &pending);
+    TypeId yoonion = arena.addType(UnionType{{builtinTypes->anyType, builtinTypes->numberType}});
+    const NormalizedType* unionIntersection = normalizer.normalize(yoonion);
+    CHECK(!unionIntersection->isExactlyNumber());
 }
 
 TEST_SUITE_END();

tests/TypeInfer.functions.test.cpp

@@ -14,7 +14,7 @@
 
 using namespace Luau;
 
-LUAU_FASTFLAG(LuauInstantiateInSubtyping);
+LUAU_FASTFLAG(LuauInstantiateInSubtyping)
 
 TEST_SUITE_BEGIN("TypeInferFunctions");
 
@@ -2073,4 +2073,20 @@ TEST_CASE_FIXTURE(BuiltinsFixture, "param_1_and_2_both_takes_the_same_generic_bu
     CHECK_EQ(toString(result.errors[1]), "Type 'number' could not be converted into 'boolean'");
 }
 
+TEST_CASE_FIXTURE(Fixture, "attempt_to_call_an_intersection_of_tables")
+{
+    CheckResult result = check(R"(
+        local function f(t: { x: number } & { y: string })
+            t()
+        end
+    )");
+
+    LUAU_REQUIRE_ERROR_COUNT(1, result);
+
+    if (FFlag::DebugLuauDeferredConstraintResolution)
+        CHECK_EQ(toString(result.errors[0]), "Cannot call non-function {| x: number |} & {| y: string |}");
+    else
+        CHECK_EQ(toString(result.errors[0]), "Cannot call non-function {| x: number |}");
+}
+
 TEST_SUITE_END();

tools/faillist.txt

@@ -88,7 +88,6 @@ TableTests.oop_polymorphic
 TableTests.quantify_even_that_table_was_never_exported_at_all
 TableTests.quantify_metatables_of_metatables_of_table
 TableTests.reasonable_error_when_adding_a_nonexistent_property_to_an_array_like_table
-TableTests.result_is_bool_for_equality_operators_if_lhs_is_any
 TableTests.right_table_missing_key2
 TableTests.shared_selfs
 TableTests.shared_selfs_from_free_param
@@ -167,7 +166,6 @@ TypeInferOperators.CallOrOfFunctions
 TypeInferOperators.cli_38355_recursive_union
 TypeInferOperators.compound_assign_mismatch_metatable
 TypeInferOperators.disallow_string_and_types_without_metatables_from_arithmetic_binary_ops
-TypeInferOperators.luau-polyfill.String.slice
 TypeInferOperators.operator_eq_completely_incompatible
 TypeInferOperators.typecheck_overloaded_multiply_that_is_an_intersection
 TypeInferOperators.typecheck_overloaded_multiply_that_is_an_intersection_on_rhs