diff --git a/compiler/dex/quick/quick_compiler.cc b/compiler/dex/quick/quick_compiler.cc
index 1cd742a..b5ecf9c 100644
--- a/compiler/dex/quick/quick_compiler.cc
+++ b/compiler/dex/quick/quick_compiler.cc
@@ -391,9 +391,9 @@
     Instruction::IGET_SHORT_QUICK,
     Instruction::INVOKE_LAMBDA,
     Instruction::UNUSED_F4,
-    Instruction::CAPTURE_VARIABLE,
+    Instruction::UNUSED_F5,
     Instruction::CREATE_LAMBDA,
-    Instruction::LIBERATE_VARIABLE,
+    Instruction::UNUSED_F7,
     Instruction::BOX_LAMBDA,
     Instruction::UNBOX_LAMBDA,
     Instruction::UNUSED_FA,
diff --git a/runtime/Android.mk b/runtime/Android.mk
index 059c4cd..995a1d5 100644
--- a/runtime/Android.mk
+++ b/runtime/Android.mk
@@ -104,7 +104,6 @@
   lambda/box_table.cc \
   lambda/closure.cc \
   lambda/closure_builder.cc \
-  lambda/leaking_allocator.cc \
   jni_internal.cc \
   jobject_comparator.cc \
   linear_alloc.cc \
diff --git a/runtime/dex_instruction-inl.h b/runtime/dex_instruction-inl.h
index e160a10..7344d13 100644
--- a/runtime/dex_instruction-inl.h
+++ b/runtime/dex_instruction-inl.h
@@ -454,8 +454,8 @@
   return FormatOf(Opcode()) == k25x;
 }
 
-// Copies all of the parameter registers into the arg array. Check the length with VRegB_25x()+2.
-inline void Instruction::GetAllArgs25x(uint32_t (&arg)[kMaxVarArgRegs25x]) const {
+// Copies all of the parameter registers into the arg array. Check the length with VRegB_25x()+1.
+inline void Instruction::GetAllArgs25x(uint32_t arg[kMaxVarArgRegs]) const {
   DCHECK_EQ(FormatOf(Opcode()), k25x);
 
   /*
@@ -500,21 +500,19 @@
    */
   switch (count) {
     case 4:
-      arg[5] = (Fetch16(0) >> 8) & 0x0f;  // vG
+      arg[4] = (Fetch16(0) >> 8) & 0x0f;  // vG
       FALLTHROUGH_INTENDED;
     case 3:
-      arg[4] = (reg_list >> 12) & 0x0f;  // vF
+      arg[3] = (reg_list >> 12) & 0x0f;  // vF
       FALLTHROUGH_INTENDED;
     case 2:
-      arg[3] = (reg_list >> 8) & 0x0f;  // vE
+      arg[2] = (reg_list >> 8) & 0x0f;  // vE
       FALLTHROUGH_INTENDED;
     case 1:
-      arg[2] = (reg_list >> 4) & 0x0f;  // vD
+      arg[1] = (reg_list >> 4) & 0x0f;  // vD
       FALLTHROUGH_INTENDED;
     default:  // case 0
-      // The required lambda 'this' is actually a pair, but the pair is implicit.
       arg[0] = VRegC_25x();  // vC
-      arg[1] = arg[0] + 1;   // vC + 1
       break;
   }
 }
diff --git a/runtime/dex_instruction.cc b/runtime/dex_instruction.cc
index 5250b0d..fc4df14 100644
--- a/runtime/dex_instruction.cc
+++ b/runtime/dex_instruction.cc
@@ -322,10 +322,10 @@
     }
     case k25x: {
       if (Opcode() == INVOKE_LAMBDA) {
-        uint32_t arg[kMaxVarArgRegs25x];
+        uint32_t arg[kMaxVarArgRegs];
         GetAllArgs25x(arg);
         const size_t num_extra_var_args = VRegB_25x();
-        DCHECK_LE(num_extra_var_args + 2, arraysize(arg));
+        DCHECK_LE(num_extra_var_args + 1, kMaxVarArgRegs);
 
         // invoke-lambda vC, {vD, vE, vF, vG}
         os << opcode << " v" << arg[0] << ", {";
@@ -333,7 +333,7 @@
           if (i != 0) {
             os << ", ";
           }
-          os << "v" << arg[i+2];  // Don't print the pair of vC registers. Pair is implicit.
+          os << "v" << arg[i+1];
         }
         os << "}";
         break;
diff --git a/runtime/dex_instruction.h b/runtime/dex_instruction.h
index 48a12e5..df2d379 100644
--- a/runtime/dex_instruction.h
+++ b/runtime/dex_instruction.h
@@ -180,11 +180,9 @@
     kVerifyVarArgRangeNonZero = 0x100000,
     kVerifyRuntimeOnly        = 0x200000,
     kVerifyError              = 0x400000,
-    kVerifyRegCString         = 0x800000,
   };
 
   static constexpr uint32_t kMaxVarArgRegs = 5;
-  static constexpr uint32_t kMaxVarArgRegs25x = 6;  // lambdas are 2 registers.
 
   // Returns the size (in 2 byte code units) of this instruction.
   size_t SizeInCodeUnits() const {
@@ -410,7 +408,7 @@
   void GetVarArgs(uint32_t args[kMaxVarArgRegs]) const {
     return GetVarArgs(args, Fetch16(0));
   }
-  void GetAllArgs25x(uint32_t (&args)[kMaxVarArgRegs25x]) const;
+  void GetAllArgs25x(uint32_t args[kMaxVarArgRegs]) const;
 
   // Returns the opcode field of the instruction. The given "inst_data" parameter must be the first
   // 16 bits of instruction.
@@ -538,7 +536,7 @@
 
   int GetVerifyTypeArgumentC() const {
     return (kInstructionVerifyFlags[Opcode()] & (kVerifyRegC | kVerifyRegCField |
-        kVerifyRegCNewArray | kVerifyRegCType | kVerifyRegCWide | kVerifyRegCString));
+        kVerifyRegCNewArray | kVerifyRegCType | kVerifyRegCWide));
   }
 
   int GetVerifyExtraFlags() const {
diff --git a/runtime/dex_instruction_list.h b/runtime/dex_instruction_list.h
index 9d7e0c4..a176772 100644
--- a/runtime/dex_instruction_list.h
+++ b/runtime/dex_instruction_list.h
@@ -263,10 +263,10 @@
   V(0xF2, IGET_SHORT_QUICK, "iget-short-quick", k22c, true, kIndexFieldOffset, kContinue | kThrow | kLoad | kRegCFieldOrConstant, kVerifyRegA | kVerifyRegB | kVerifyRuntimeOnly) \
   V(0xF3, INVOKE_LAMBDA, "invoke-lambda", k25x, false, kIndexNone, kContinue | kThrow | kInvoke | kExperimental, kVerifyRegC /*TODO: | kVerifyVarArg*/) \
   V(0xF4, UNUSED_F4, "unused-f4", k10x, false, kIndexUnknown, 0, kVerifyError) \
-  V(0xF5, CAPTURE_VARIABLE, "capture-variable", k21c, false, kIndexStringRef, kExperimental, kVerifyRegA | kVerifyRegBString) \
+  V(0xF5, UNUSED_F5, "unused-f5", k10x, false, kIndexUnknown, 0, kVerifyError) \
   /* TODO(iam): get rid of the unused 'false' column */ \
   V(0xF6, CREATE_LAMBDA, "create-lambda", k21c, false_UNUSED, kIndexMethodRef, kContinue | kThrow | kExperimental, kVerifyRegA | kVerifyRegBMethod) \
-  V(0xF7, LIBERATE_VARIABLE, "liberate-variable", k22c, false, kIndexStringRef, kExperimental, kVerifyRegA | kVerifyRegB | kVerifyRegCString) \
+  V(0xF7, UNUSED_F7, "unused-f7", k10x, false, kIndexUnknown, 0, kVerifyError) \
   V(0xF8, BOX_LAMBDA, "box-lambda", k22x, true, kIndexNone, kContinue | kExperimental, kVerifyRegA | kVerifyRegB) \
   V(0xF9, UNBOX_LAMBDA, "unbox-lambda", k22c, true, kIndexTypeRef, kContinue | kThrow | kExperimental, kVerifyRegA | kVerifyRegB | kVerifyRegCType) \
   V(0xFA, UNUSED_FA, "unused-fa", k10x, false, kIndexUnknown, 0, kVerifyError) \
diff --git a/runtime/interpreter/interpreter_common.cc b/runtime/interpreter/interpreter_common.cc
index c1af2ef..68d56f5 100644
--- a/runtime/interpreter/interpreter_common.cc
+++ b/runtime/interpreter/interpreter_common.cc
@@ -27,9 +27,6 @@
 namespace art {
 namespace interpreter {
 
-// All lambda closures have to be a consecutive pair of virtual registers.
-static constexpr size_t kLambdaVirtualRegisterWidth = 2;
-
 void ThrowNullPointerExceptionFromInterpreter() {
   ThrowNullPointerExceptionFromDexPC();
 }
@@ -486,16 +483,13 @@
 }
 
 // Separate declaration is required solely for the attributes.
-template <bool is_range,
-          bool do_assignability_check,
-          size_t kVarArgMax>
-    SHARED_REQUIRES(Locks::mutator_lock_)
+template<bool is_range, bool do_assignability_check> SHARED_REQUIRES(Locks::mutator_lock_)
 static inline bool DoCallCommon(ArtMethod* called_method,
                                 Thread* self,
                                 ShadowFrame& shadow_frame,
                                 JValue* result,
                                 uint16_t number_of_inputs,
-                                uint32_t (&arg)[kVarArgMax],
+                                uint32_t arg[Instruction::kMaxVarArgRegs],
                                 uint32_t vregC) ALWAYS_INLINE;
 
 SHARED_REQUIRES(Locks::mutator_lock_)
@@ -515,15 +509,13 @@
         Dbg::IsForcedInterpreterNeededForCalling(self, target);
 }
 
-template <bool is_range,
-          bool do_assignability_check,
-          size_t kVarArgMax>
+template<bool is_range, bool do_assignability_check>
 static inline bool DoCallCommon(ArtMethod* called_method,
                                 Thread* self,
                                 ShadowFrame& shadow_frame,
                                 JValue* result,
                                 uint16_t number_of_inputs,
-                                uint32_t (&arg)[kVarArgMax],
+                                uint32_t arg[Instruction::kMaxVarArgRegs],
                                 uint32_t vregC) {
   bool string_init = false;
   // Replace calls to String.<init> with equivalent StringFactory call.
@@ -568,10 +560,10 @@
     number_of_inputs--;
 
     // Rewrite the var-args, dropping the 0th argument ("this")
-    for (uint32_t i = 1; i < arraysize(arg); ++i) {
+    for (uint32_t i = 1; i < Instruction::kMaxVarArgRegs; ++i) {
       arg[i - 1] = arg[i];
     }
-    arg[arraysize(arg) - 1] = 0;
+    arg[Instruction::kMaxVarArgRegs - 1] = 0;
 
     // Rewrite the non-var-arg case
     vregC++;  // Skips the 0th vreg in the range ("this").
@@ -677,7 +669,7 @@
         AssignRegister(new_shadow_frame, shadow_frame, dest_reg, src_reg);
       }
     } else {
-      DCHECK_LE(number_of_inputs, arraysize(arg));
+      DCHECK_LE(number_of_inputs, Instruction::kMaxVarArgRegs);
 
       for (; arg_index < number_of_inputs; ++arg_index) {
         AssignRegister(new_shadow_frame, shadow_frame, first_dest_reg + arg_index, arg[arg_index]);
@@ -740,13 +732,12 @@
                   const Instruction* inst, uint16_t inst_data, JValue* result) {
   const uint4_t num_additional_registers = inst->VRegB_25x();
   // Argument word count.
-  const uint16_t number_of_inputs = num_additional_registers + kLambdaVirtualRegisterWidth;
-  // The lambda closure register is always present and is not encoded in the count.
-  // Furthermore, the lambda closure register is always wide, so it counts as 2 inputs.
+  const uint16_t number_of_inputs = num_additional_registers + 1;
+  // The first input register is always present and is not encoded in the count.
 
   // TODO: find a cleaner way to separate non-range and range information without duplicating
   //       code.
-  uint32_t arg[Instruction::kMaxVarArgRegs25x];  // only used in invoke-XXX.
+  uint32_t arg[Instruction::kMaxVarArgRegs];  // only used in invoke-XXX.
   uint32_t vregC = 0;   // only used in invoke-XXX-range.
   if (is_range) {
     vregC = inst->VRegC_3rc();
@@ -772,7 +763,7 @@
 
   // TODO: find a cleaner way to separate non-range and range information without duplicating
   //       code.
-  uint32_t arg[Instruction::kMaxVarArgRegs] = {};  // only used in invoke-XXX.
+  uint32_t arg[Instruction::kMaxVarArgRegs];  // only used in invoke-XXX.
   uint32_t vregC = 0;
   if (is_range) {
     vregC = inst->VRegC_3rc();
diff --git a/runtime/interpreter/interpreter_common.h b/runtime/interpreter/interpreter_common.h
index f57bddb..7398778 100644
--- a/runtime/interpreter/interpreter_common.h
+++ b/runtime/interpreter/interpreter_common.h
@@ -34,12 +34,7 @@
 #include "dex_instruction-inl.h"
 #include "entrypoints/entrypoint_utils-inl.h"
 #include "handle_scope-inl.h"
-#include "lambda/art_lambda_method.h"
 #include "lambda/box_table.h"
-#include "lambda/closure.h"
-#include "lambda/closure_builder-inl.h"
-#include "lambda/leaking_allocator.h"
-#include "lambda/shorty_field_type.h"
 #include "mirror/class-inl.h"
 #include "mirror/method.h"
 #include "mirror/object-inl.h"
@@ -138,44 +133,32 @@
   return success;
 }
 
-// Write out the 'Closure*' into vreg and vreg+1, as if it was a jlong.
+// Write out the 'ArtMethod*' into vreg and vreg+1
 static inline void WriteLambdaClosureIntoVRegs(ShadowFrame& shadow_frame,
-                                               const lambda::Closure* lambda_closure,
+                                               const ArtMethod& called_method,
                                                uint32_t vreg) {
   // Split the method into a lo and hi 32 bits so we can encode them into 2 virtual registers.
-  uint32_t closure_lo = static_cast<uint32_t>(reinterpret_cast<uintptr_t>(lambda_closure));
-  uint32_t closure_hi = static_cast<uint32_t>(reinterpret_cast<uint64_t>(lambda_closure)
+  uint32_t called_method_lo = static_cast<uint32_t>(reinterpret_cast<uintptr_t>(&called_method));
+  uint32_t called_method_hi = static_cast<uint32_t>(reinterpret_cast<uint64_t>(&called_method)
                                                     >> BitSizeOf<uint32_t>());
   // Use uint64_t instead of uintptr_t to allow shifting past the max on 32-bit.
   static_assert(sizeof(uint64_t) >= sizeof(uintptr_t), "Impossible");
 
-  DCHECK_NE(closure_lo | closure_hi, 0u);
+  DCHECK_NE(called_method_lo | called_method_hi, 0u);
 
-  shadow_frame.SetVReg(vreg, closure_lo);
-  shadow_frame.SetVReg(vreg + 1, closure_hi);
+  shadow_frame.SetVReg(vreg, called_method_lo);
+  shadow_frame.SetVReg(vreg + 1, called_method_hi);
 }
 
 // Handles create-lambda instructions.
 // Returns true on success, otherwise throws an exception and returns false.
 // (Exceptions are thrown by creating a new exception and then being put in the thread TLS)
 //
-// The closure must be allocated big enough to hold the data, and should not be
-// pre-initialized. It is initialized with the actual captured variables as a side-effect,
-// although this should be unimportant to the caller since this function also handles storing it to
-// the ShadowFrame.
-//
 // As a work-in-progress implementation, this shoves the ArtMethod object corresponding
 // to the target dex method index into the target register vA and vA + 1.
 template<bool do_access_check>
-static inline bool DoCreateLambda(Thread* self,
-                                  const Instruction* inst,
-                                  /*inout*/ShadowFrame& shadow_frame,
-                                  /*inout*/lambda::ClosureBuilder* closure_builder,
-                                  /*inout*/lambda::Closure* uninitialized_closure) {
-  DCHECK(closure_builder != nullptr);
-  DCHECK(uninitialized_closure != nullptr);
-  DCHECK_ALIGNED(uninitialized_closure, alignof(lambda::Closure));
-
+static inline bool DoCreateLambda(Thread* self, ShadowFrame& shadow_frame,
+                                  const Instruction* inst) {
   /*
    * create-lambda is opcode 0x21c
    * - vA is the target register where the closure will be stored into
@@ -188,69 +171,16 @@
   ArtMethod* const called_method = FindMethodFromCode<kStatic, do_access_check>(
       method_idx, &receiver, sf_method, self);
 
-  uint32_t vreg_dest_closure = inst->VRegA_21c();
+  uint32_t vregA = inst->VRegA_21c();
 
   if (UNLIKELY(!IsValidLambdaTargetOrThrow(called_method))) {
     CHECK(self->IsExceptionPending());
-    shadow_frame.SetVReg(vreg_dest_closure, 0u);
-    shadow_frame.SetVReg(vreg_dest_closure + 1, 0u);
+    shadow_frame.SetVReg(vregA, 0u);
+    shadow_frame.SetVReg(vregA + 1, 0u);
     return false;
   }
 
-  lambda::ArtLambdaMethod* initialized_lambda_method;
-  // Initialize the ArtLambdaMethod with the right data.
-  {
-    lambda::ArtLambdaMethod* uninitialized_lambda_method =
-        reinterpret_cast<lambda::ArtLambdaMethod*>(
-            lambda::LeakingAllocator::AllocateMemory(self, sizeof(lambda::ArtLambdaMethod)));
-
-    std::string captured_variables_shorty = closure_builder->GetCapturedVariableShortyTypes();
-    std::string captured_variables_long_type_desc;
-
-    // Synthesize a long type descriptor from the short one.
-    for (char shorty : captured_variables_shorty) {
-      lambda::ShortyFieldType shorty_field_type(shorty);
-      if (shorty_field_type.IsObject()) {
-        // Not the true type, but good enough until we implement verifier support.
-        captured_variables_long_type_desc += "Ljava/lang/Object;";
-        UNIMPLEMENTED(FATAL) << "create-lambda with an object captured variable";
-      } else if (shorty_field_type.IsLambda()) {
-        // Not the true type, but good enough until we implement verifier support.
-        captured_variables_long_type_desc += "Ljava/lang/Runnable;";
-        UNIMPLEMENTED(FATAL) << "create-lambda with a lambda captured variable";
-      } else {
-        // The primitive types have the same length shorty or not, so this is always correct.
-        DCHECK(shorty_field_type.IsPrimitive());
-        captured_variables_long_type_desc += shorty_field_type;
-      }
-    }
-
-    // Copy strings to dynamically allocated storage. This leaks, but that's ok. Fix it later.
-    // TODO: Strings need to come from the DexFile, so they won't need their own allocations.
-    char* captured_variables_type_desc = lambda::LeakingAllocator::MakeFlexibleInstance<char>(
-        self,
-        captured_variables_long_type_desc.size() + 1);
-    strcpy(captured_variables_type_desc, captured_variables_long_type_desc.c_str());
-    char* captured_variables_shorty_copy = lambda::LeakingAllocator::MakeFlexibleInstance<char>(
-        self,
-        captured_variables_shorty.size() + 1);
-    strcpy(captured_variables_shorty_copy, captured_variables_shorty.c_str());
-
-    new (uninitialized_lambda_method) lambda::ArtLambdaMethod(called_method,
-                                                              captured_variables_type_desc,
-                                                              captured_variables_shorty_copy,
-                                                              true);  // innate lambda
-    initialized_lambda_method = uninitialized_lambda_method;
-  }
-
-  // Write all the closure captured variables and the closure header into the closure.
-  lambda::Closure* initialized_closure;
-  {
-    initialized_closure =
-        closure_builder->CreateInPlace(uninitialized_closure, initialized_lambda_method);
-  }
-
-  WriteLambdaClosureIntoVRegs(/*inout*/shadow_frame, initialized_closure, vreg_dest_closure);
+  WriteLambdaClosureIntoVRegs(shadow_frame, *called_method, vregA);
   return true;
 }
 
@@ -259,11 +189,13 @@
 // Validates that the art method points to a valid lambda function, otherwise throws
 // an exception and returns null.
 // (Exceptions are thrown by creating a new exception and then being put in the thread TLS)
-static inline lambda::Closure* ReadLambdaClosureFromVRegsOrThrow(ShadowFrame& shadow_frame,
-                                                                 uint32_t vreg)
+static inline ArtMethod* ReadLambdaClosureFromVRegsOrThrow(ShadowFrame& shadow_frame,
+                                                           uint32_t vreg)
     SHARED_REQUIRES(Locks::mutator_lock_) {
-  // Lambda closures take up a consecutive pair of 2 virtual registers.
-  // On 32-bit the high bits are always 0.
+  // TODO(iam): Introduce a closure abstraction that will contain the captured variables
+  // instead of just an ArtMethod.
+  // This is temporarily using 2 vregs because a native ArtMethod can be up to 64-bit,
+  // but once proper variable capture is implemented it will only use 1 vreg.
   uint32_t vc_value_lo = shadow_frame.GetVReg(vreg);
   uint32_t vc_value_hi = shadow_frame.GetVReg(vreg + 1);
 
@@ -272,285 +204,17 @@
 
   // Use uint64_t instead of uintptr_t to allow left-shifting past the max on 32-bit.
   static_assert(sizeof(uint64_t) >= sizeof(uintptr_t), "Impossible");
-  lambda::Closure* const lambda_closure = reinterpret_cast<lambda::Closure*>(vc_value_ptr);
-  DCHECK_ALIGNED(lambda_closure, alignof(lambda::Closure));
+  ArtMethod* const called_method = reinterpret_cast<ArtMethod* const>(vc_value_ptr);
 
   // Guard against the user passing a null closure, which is odd but (sadly) semantically valid.
-  if (UNLIKELY(lambda_closure == nullptr)) {
+  if (UNLIKELY(called_method == nullptr)) {
     ThrowNullPointerExceptionFromInterpreter();
     return nullptr;
-  } else if (UNLIKELY(!IsValidLambdaTargetOrThrow(lambda_closure->GetTargetMethod()))) {
-    // Sanity check against data corruption.
+  } else if (UNLIKELY(!IsValidLambdaTargetOrThrow(called_method))) {
     return nullptr;
   }
 
-  return lambda_closure;
-}
-
-// Forward declaration for lock annotations. See below for documentation.
-template <bool do_access_check>
-static inline const char* GetStringDataByDexStringIndexOrThrow(ShadowFrame& shadow_frame,
-                                                               uint32_t string_idx)
-    SHARED_REQUIRES(Locks::mutator_lock_);
-
-// Find the c-string data corresponding to a dex file's string index.
-// Otherwise, returns null if not found and throws a VerifyError.
-//
-// Note that with do_access_check=false, we never return null because the verifier
-// must guard against invalid string indices.
-// (Exceptions are thrown by creating a new exception and then being put in the thread TLS)
-template <bool do_access_check>
-static inline const char* GetStringDataByDexStringIndexOrThrow(ShadowFrame& shadow_frame,
-                                                               uint32_t string_idx) {
-  ArtMethod* method = shadow_frame.GetMethod();
-  const DexFile* dex_file = method->GetDexFile();
-
-  mirror::Class* declaring_class = method->GetDeclaringClass();
-  if (!do_access_check) {
-    // MethodVerifier refuses methods with string_idx out of bounds.
-    DCHECK_LT(string_idx, declaring_class->GetDexCache()->NumStrings());
-  } else {
-    // Access checks enabled: perform string index bounds ourselves.
-    if (string_idx >= dex_file->GetHeader().string_ids_size_) {
-      ThrowVerifyError(declaring_class, "String index '%" PRIu32 "' out of bounds",
-                       string_idx);
-      return nullptr;
-    }
-  }
-
-  const char* type_string = dex_file->StringDataByIdx(string_idx);
-
-  if (UNLIKELY(type_string == nullptr)) {
-    CHECK_EQ(false, do_access_check)
-        << " verifier should've caught invalid string index " << string_idx;
-    CHECK_EQ(true, do_access_check)
-        << " string idx size check should've caught invalid string index " << string_idx;
-  }
-
-  return type_string;
-}
-
-// Handles capture-variable instructions.
-// Returns true on success, otherwise throws an exception and returns false.
-// (Exceptions are thrown by creating a new exception and then being put in the thread TLS)
-template<bool do_access_check>
-static inline bool DoCaptureVariable(Thread* self,
-                                     const Instruction* inst,
-                                     /*inout*/ShadowFrame& shadow_frame,
-                                     /*inout*/lambda::ClosureBuilder* closure_builder) {
-  DCHECK(closure_builder != nullptr);
-  using lambda::ShortyFieldType;
-  /*
-   * capture-variable is opcode 0xf6, fmt 0x21c
-   * - vA is the source register of the variable that will be captured
-   * - vB is the string ID of the variable's type that will be captured
-   */
-  const uint32_t source_vreg = inst->VRegA_21c();
-  const uint32_t string_idx = inst->VRegB_21c();
-  // TODO: this should be a proper [type id] instead of a [string ID] pointing to a type.
-
-  const char* type_string = GetStringDataByDexStringIndexOrThrow<do_access_check>(shadow_frame,
-                                                                                  string_idx);
-  if (UNLIKELY(type_string == nullptr)) {
-    CHECK(self->IsExceptionPending());
-    return false;
-  }
-
-  char type_first_letter = type_string[0];
-  ShortyFieldType shorty_type;
-  if (do_access_check &&
-      UNLIKELY(!ShortyFieldType::MaybeCreate(type_first_letter, /*out*/&shorty_type))) {  // NOLINT: [whitespace/comma] [3]
-    ThrowVerifyError(shadow_frame.GetMethod()->GetDeclaringClass(),
-                     "capture-variable vB must be a valid type");
-    return false;
-  } else {
-    // Already verified that the type is valid.
-    shorty_type = ShortyFieldType(type_first_letter);
-  }
-
-  const size_t captured_variable_count = closure_builder->GetCaptureCount();
-
-  // Note: types are specified explicitly so that the closure is packed tightly.
-  switch (shorty_type) {
-    case ShortyFieldType::kBoolean: {
-      uint32_t primitive_narrow_value = shadow_frame.GetVReg(source_vreg);
-      closure_builder->CaptureVariablePrimitive<bool>(primitive_narrow_value);
-      break;
-    }
-    case ShortyFieldType::kByte: {
-      uint32_t primitive_narrow_value = shadow_frame.GetVReg(source_vreg);
-      closure_builder->CaptureVariablePrimitive<int8_t>(primitive_narrow_value);
-      break;
-    }
-    case ShortyFieldType::kChar: {
-      uint32_t primitive_narrow_value = shadow_frame.GetVReg(source_vreg);
-      closure_builder->CaptureVariablePrimitive<uint16_t>(primitive_narrow_value);
-      break;
-    }
-    case ShortyFieldType::kShort: {
-      uint32_t primitive_narrow_value = shadow_frame.GetVReg(source_vreg);
-      closure_builder->CaptureVariablePrimitive<int16_t>(primitive_narrow_value);
-      break;
-    }
-    case ShortyFieldType::kInt: {
-      uint32_t primitive_narrow_value = shadow_frame.GetVReg(source_vreg);
-      closure_builder->CaptureVariablePrimitive<int32_t>(primitive_narrow_value);
-      break;
-    }
-    case ShortyFieldType::kDouble: {
-      closure_builder->CaptureVariablePrimitive(shadow_frame.GetVRegDouble(source_vreg));
-      break;
-    }
-    case ShortyFieldType::kFloat: {
-      closure_builder->CaptureVariablePrimitive(shadow_frame.GetVRegFloat(source_vreg));
-      break;
-    }
-    case ShortyFieldType::kLambda: {
-      UNIMPLEMENTED(FATAL) << " capture-variable with type kLambda";
-      // TODO: Capturing lambdas recursively will be done at a later time.
-      UNREACHABLE();
-    }
-    case ShortyFieldType::kLong: {
-      closure_builder->CaptureVariablePrimitive(shadow_frame.GetVRegLong(source_vreg));
-      break;
-    }
-    case ShortyFieldType::kObject: {
-      closure_builder->CaptureVariableObject(shadow_frame.GetVRegReference(source_vreg));
-      UNIMPLEMENTED(FATAL) << " capture-variable with type kObject";
-      // TODO: finish implementing this. disabled for now since we can't track lambda refs for GC.
-      UNREACHABLE();
-    }
-
-    default:
-      LOG(FATAL) << "Invalid shorty type value " << shorty_type;
-      UNREACHABLE();
-  }
-
-  DCHECK_EQ(captured_variable_count + 1, closure_builder->GetCaptureCount());
-
-  return true;
-}
-
-// Handles capture-variable instructions.
-// Returns true on success, otherwise throws an exception and returns false.
-// (Exceptions are thrown by creating a new exception and then being put in the thread TLS)
-template<bool do_access_check>
-static inline bool DoLiberateVariable(Thread* self,
-                                     const Instruction* inst,
-                                     size_t captured_variable_index,
-                                     /*inout*/ShadowFrame& shadow_frame) {
-  using lambda::ShortyFieldType;
-  /*
-   * liberate-variable is opcode 0xf7, fmt 0x22c
-   * - vA is the destination register
-   * - vB is the register with the lambda closure in it
-   * - vC is the string ID which needs to be a valid field type descriptor
-   */
-
-  const uint32_t dest_vreg = inst->VRegA_22c();
-  const uint32_t closure_vreg = inst->VRegB_22c();
-  const uint32_t string_idx = inst->VRegC_22c();
-  // TODO: this should be a proper [type id] instead of a [string ID] pointing to a type.
-
-
-  // Synthesize a long type descriptor from a shorty type descriptor list.
-  // TODO: Fix the dex encoding to contain the long and short type descriptors.
-  const char* type_string = GetStringDataByDexStringIndexOrThrow<do_access_check>(shadow_frame,
-                                                                                  string_idx);
-  if (UNLIKELY(do_access_check && type_string == nullptr)) {
-    CHECK(self->IsExceptionPending());
-    shadow_frame.SetVReg(dest_vreg, 0);
-    return false;
-  }
-
-  char type_first_letter = type_string[0];
-  ShortyFieldType shorty_type;
-  if (do_access_check &&
-      UNLIKELY(!ShortyFieldType::MaybeCreate(type_first_letter, /*out*/&shorty_type))) {  // NOLINT: [whitespace/comma] [3]
-    ThrowVerifyError(shadow_frame.GetMethod()->GetDeclaringClass(),
-                     "liberate-variable vC must be a valid type");
-    shadow_frame.SetVReg(dest_vreg, 0);
-    return false;
-  } else {
-    // Already verified that the type is valid.
-    shorty_type = ShortyFieldType(type_first_letter);
-  }
-
-  // Check for closure being null *after* the type check.
-  // This way we can access the type info in case we fail later, to know how many vregs to clear.
-  const lambda::Closure* lambda_closure =
-      ReadLambdaClosureFromVRegsOrThrow(/*inout*/shadow_frame, closure_vreg);
-
-  // Failed lambda target runtime check, an exception was raised.
-  if (UNLIKELY(lambda_closure == nullptr)) {
-    CHECK(self->IsExceptionPending());
-
-    // Clear the destination vreg(s) to be safe.
-    shadow_frame.SetVReg(dest_vreg, 0);
-    if (shorty_type.IsPrimitiveWide() || shorty_type.IsLambda()) {
-      shadow_frame.SetVReg(dest_vreg + 1, 0);
-    }
-    return false;
-  }
-
-  if (do_access_check &&
-      UNLIKELY(captured_variable_index >= lambda_closure->GetNumberOfCapturedVariables())) {
-    ThrowVerifyError(shadow_frame.GetMethod()->GetDeclaringClass(),
-                     "liberate-variable captured variable index %zu out of bounds",
-                     lambda_closure->GetNumberOfCapturedVariables());
-    // Clear the destination vreg(s) to be safe.
-    shadow_frame.SetVReg(dest_vreg, 0);
-    if (shorty_type.IsPrimitiveWide() || shorty_type.IsLambda()) {
-      shadow_frame.SetVReg(dest_vreg + 1, 0);
-    }
-    return false;
-  }
-
-  // Verify that the runtime type of the captured-variable matches the requested dex type.
-  if (do_access_check) {
-    ShortyFieldType actual_type = lambda_closure->GetCapturedShortyType(captured_variable_index);
-    if (actual_type != shorty_type) {
-      ThrowVerifyError(shadow_frame.GetMethod()->GetDeclaringClass(),
-                     "cannot liberate-variable of runtime type '%c' to dex type '%c'",
-                     static_cast<char>(actual_type),
-                     static_cast<char>(shorty_type));
-
-      shadow_frame.SetVReg(dest_vreg, 0);
-      if (shorty_type.IsPrimitiveWide() || shorty_type.IsLambda()) {
-        shadow_frame.SetVReg(dest_vreg + 1, 0);
-      }
-      return false;
-    }
-
-    if (actual_type.IsLambda() || actual_type.IsObject()) {
-      UNIMPLEMENTED(FATAL) << "liberate-variable type checks needs to "
-                           << "parse full type descriptor for objects and lambdas";
-    }
-  }
-
-  // Unpack the captured variable from the closure into the correct type, then save it to the vreg.
-  if (shorty_type.IsPrimitiveNarrow()) {
-    uint32_t primitive_narrow_value =
-        lambda_closure->GetCapturedPrimitiveNarrow(captured_variable_index);
-    shadow_frame.SetVReg(dest_vreg, primitive_narrow_value);
-  } else if (shorty_type.IsPrimitiveWide()) {
-      uint64_t primitive_wide_value =
-          lambda_closure->GetCapturedPrimitiveWide(captured_variable_index);
-      shadow_frame.SetVRegLong(dest_vreg, static_cast<int64_t>(primitive_wide_value));
-  } else if (shorty_type.IsObject()) {
-    mirror::Object* unpacked_object =
-        lambda_closure->GetCapturedObject(captured_variable_index);
-    shadow_frame.SetVRegReference(dest_vreg, unpacked_object);
-
-    UNIMPLEMENTED(FATAL) << "liberate-variable cannot unpack objects yet";
-  } else if (shorty_type.IsLambda()) {
-    UNIMPLEMENTED(FATAL) << "liberate-variable cannot unpack lambdas yet";
-  } else {
-    LOG(FATAL) << "unreachable";
-    UNREACHABLE();
-  }
-
-  return true;
+  return called_method;
 }
 
 template<bool do_access_check>
@@ -565,24 +229,22 @@
    *
    * - reading var-args for 0x25 gets us vD,vE,vF,vG (but not vB)
    */
-  uint32_t vreg_closure = inst->VRegC_25x();
-  const lambda::Closure* lambda_closure =
-      ReadLambdaClosureFromVRegsOrThrow(shadow_frame, vreg_closure);
+  uint32_t vC = inst->VRegC_25x();
+  ArtMethod* const called_method = ReadLambdaClosureFromVRegsOrThrow(shadow_frame, vC);
 
   // Failed lambda target runtime check, an exception was raised.
-  if (UNLIKELY(lambda_closure == nullptr)) {
+  if (UNLIKELY(called_method == nullptr)) {
     CHECK(self->IsExceptionPending());
     result->SetJ(0);
     return false;
   }
 
-  ArtMethod* const called_method = lambda_closure->GetTargetMethod();
   // Invoke a non-range lambda
   return DoLambdaCall<false, do_access_check>(called_method, self, shadow_frame, inst, inst_data,
                                               result);
 }
 
-// Handles invoke-XXX/range instructions (other than invoke-lambda[-range]).
+// Handles invoke-XXX/range instructions.
 // Returns true on success, otherwise throws an exception and returns false.
 template<InvokeType type, bool is_range, bool do_access_check>
 static inline bool DoInvoke(Thread* self, ShadowFrame& shadow_frame, const Instruction* inst,
@@ -859,17 +521,17 @@
   uint32_t vreg_target_object = inst->VRegA_22x(inst_data);
   uint32_t vreg_source_closure = inst->VRegB_22x();
 
-  lambda::Closure* lambda_closure = ReadLambdaClosureFromVRegsOrThrow(shadow_frame,
-                                                                      vreg_source_closure);
+  ArtMethod* closure_method = ReadLambdaClosureFromVRegsOrThrow(shadow_frame,
+                                                                vreg_source_closure);
 
   // Failed lambda target runtime check, an exception was raised.
-  if (UNLIKELY(lambda_closure == nullptr)) {
+  if (UNLIKELY(closure_method == nullptr)) {
     CHECK(self->IsExceptionPending());
     return false;
   }
 
   mirror::Object* closure_as_object =
-      Runtime::Current()->GetLambdaBoxTable()->BoxLambda(lambda_closure);
+      Runtime::Current()->GetLambdaBoxTable()->BoxLambda(closure_method);
 
   // Failed to box the lambda, an exception was raised.
   if (UNLIKELY(closure_as_object == nullptr)) {
@@ -902,16 +564,16 @@
     return false;
   }
 
-  lambda::Closure* unboxed_closure = nullptr;
+  ArtMethod* unboxed_closure = nullptr;
   // Raise an exception if unboxing fails.
   if (!Runtime::Current()->GetLambdaBoxTable()->UnboxLambda(boxed_closure_object,
-                                                            /*out*/&unboxed_closure)) {
+                                                            &unboxed_closure)) {
     CHECK(self->IsExceptionPending());
     return false;
   }
 
   DCHECK(unboxed_closure != nullptr);
-  WriteLambdaClosureIntoVRegs(/*inout*/shadow_frame, unboxed_closure, vreg_target_closure);
+  WriteLambdaClosureIntoVRegs(shadow_frame, *unboxed_closure, vreg_target_closure);
   return true;
 }
 
@@ -988,13 +650,10 @@
 #undef EXPLICIT_INSTANTIATION_DO_INVOKE_VIRTUAL_QUICK
 
 // Explicitly instantiate all DoCreateLambda functions.
-#define EXPLICIT_DO_CREATE_LAMBDA_DECL(_do_check)                                                 \
-template SHARED_REQUIRES(Locks::mutator_lock_)                                                    \
-bool DoCreateLambda<_do_check>(Thread* self,                                                      \
-                               const Instruction* inst,                                           \
-                               /*inout*/ShadowFrame& shadow_frame,                                \
-                               /*inout*/lambda::ClosureBuilder* closure_builder,                  \
-                               /*inout*/lambda::Closure* uninitialized_closure);
+#define EXPLICIT_DO_CREATE_LAMBDA_DECL(_do_check)                                    \
+template SHARED_REQUIRES(Locks::mutator_lock_)                                 \
+bool DoCreateLambda<_do_check>(Thread* self, ShadowFrame& shadow_frame,              \
+                        const Instruction* inst)
 
 EXPLICIT_DO_CREATE_LAMBDA_DECL(false);  // create-lambda
 EXPLICIT_DO_CREATE_LAMBDA_DECL(true);   // create-lambda
@@ -1030,29 +689,7 @@
 EXPLICIT_DO_UNBOX_LAMBDA_DECL(true);   // unbox-lambda
 #undef EXPLICIT_DO_BOX_LAMBDA_DECL
 
-// Explicitly instantiate all DoCaptureVariable functions.
-#define EXPLICIT_DO_CAPTURE_VARIABLE_DECL(_do_check)                                    \
-template SHARED_REQUIRES(Locks::mutator_lock_)                                          \
-bool DoCaptureVariable<_do_check>(Thread* self,                                         \
-                                  const Instruction* inst,                              \
-                                  ShadowFrame& shadow_frame,                            \
-                                  lambda::ClosureBuilder* closure_builder);
 
-EXPLICIT_DO_CAPTURE_VARIABLE_DECL(false);  // capture-variable
-EXPLICIT_DO_CAPTURE_VARIABLE_DECL(true);   // capture-variable
-#undef EXPLICIT_DO_CREATE_LAMBDA_DECL
-
-// Explicitly instantiate all DoLiberateVariable functions.
-#define EXPLICIT_DO_LIBERATE_VARIABLE_DECL(_do_check)                                   \
-template SHARED_REQUIRES(Locks::mutator_lock_)                                          \
-bool DoLiberateVariable<_do_check>(Thread* self,                                        \
-                                   const Instruction* inst,                             \
-                                   size_t captured_variable_index,                      \
-                                   ShadowFrame& shadow_frame);                          \
-
-EXPLICIT_DO_LIBERATE_VARIABLE_DECL(false);  // liberate-variable
-EXPLICIT_DO_LIBERATE_VARIABLE_DECL(true);   // liberate-variable
-#undef EXPLICIT_DO_LIBERATE_LAMBDA_DECL
 }  // namespace interpreter
 }  // namespace art
 
diff --git a/runtime/interpreter/interpreter_goto_table_impl.cc b/runtime/interpreter/interpreter_goto_table_impl.cc
index 9677d79..72e2ba0 100644
--- a/runtime/interpreter/interpreter_goto_table_impl.cc
+++ b/runtime/interpreter/interpreter_goto_table_impl.cc
@@ -17,13 +17,9 @@
 #if !defined(__clang__)
 // Clang 3.4 fails to build the goto interpreter implementation.
 
-
-#include "base/stl_util.h"  // MakeUnique
 #include "interpreter_common.h"
 #include "safe_math.h"
 
-#include <memory>  // std::unique_ptr
-
 namespace art {
 namespace interpreter {
 
@@ -183,9 +179,6 @@
     }
   }
 
-  std::unique_ptr<lambda::ClosureBuilder> lambda_closure_builder;
-  size_t lambda_captured_variable_index = 0;
-
   // Jump to first instruction.
   ADVANCE(0);
   UNREACHABLE_CODE_CHECK();
@@ -2419,20 +2412,7 @@
   HANDLE_INSTRUCTION_END();
 
   HANDLE_EXPERIMENTAL_INSTRUCTION_START(CREATE_LAMBDA) {
-    if (lambda_closure_builder == nullptr) {
-      // DoCreateLambda always needs a ClosureBuilder, even if it has 0 captured variables.
-      lambda_closure_builder = MakeUnique<lambda::ClosureBuilder>();
-    }
-
-    // TODO: these allocations should not leak, and the lambda method should not be local.
-    lambda::Closure* lambda_closure =
-        reinterpret_cast<lambda::Closure*>(alloca(lambda_closure_builder->GetSize()));
-    bool success = DoCreateLambda<do_access_check>(self,
-                                                   inst,
-                                                   /*inout*/shadow_frame,
-                                                   /*inout*/lambda_closure_builder.get(),
-                                                   /*inout*/lambda_closure);
-    lambda_closure_builder.reset(nullptr);  // reset state of variables captured
+    bool success = DoCreateLambda<true>(self, shadow_frame, inst);
     POSSIBLY_HANDLE_PENDING_EXCEPTION(!success, 2);
   }
   HANDLE_EXPERIMENTAL_INSTRUCTION_END();
@@ -2449,31 +2429,6 @@
   }
   HANDLE_EXPERIMENTAL_INSTRUCTION_END();
 
-  HANDLE_EXPERIMENTAL_INSTRUCTION_START(CAPTURE_VARIABLE) {
-    if (lambda_closure_builder == nullptr) {
-      lambda_closure_builder = MakeUnique<lambda::ClosureBuilder>();
-    }
-
-    bool success = DoCaptureVariable<do_access_check>(self,
-                                                      inst,
-                                                      /*inout*/shadow_frame,
-                                                      /*inout*/lambda_closure_builder.get());
-
-    POSSIBLY_HANDLE_PENDING_EXCEPTION(!success, 2);
-  }
-  HANDLE_EXPERIMENTAL_INSTRUCTION_END();
-
-  HANDLE_EXPERIMENTAL_INSTRUCTION_START(LIBERATE_VARIABLE) {
-    bool success = DoLiberateVariable<do_access_check>(self,
-                                                           inst,
-                                                           lambda_captured_variable_index,
-                                                           /*inout*/shadow_frame);
-    // Temporarily only allow sequences of 'liberate-variable, liberate-variable, ...'
-    lambda_captured_variable_index++;
-    POSSIBLY_HANDLE_PENDING_EXCEPTION(!success, 2);
-  }
-  HANDLE_EXPERIMENTAL_INSTRUCTION_END();
-
   HANDLE_INSTRUCTION_START(UNUSED_3E)
     UnexpectedOpcode(inst, shadow_frame);
   HANDLE_INSTRUCTION_END();
@@ -2510,6 +2465,14 @@
     UnexpectedOpcode(inst, shadow_frame);
   HANDLE_INSTRUCTION_END();
 
+  HANDLE_INSTRUCTION_START(UNUSED_F5)
+    UnexpectedOpcode(inst, shadow_frame);
+  HANDLE_INSTRUCTION_END();
+
+  HANDLE_INSTRUCTION_START(UNUSED_F7)
+    UnexpectedOpcode(inst, shadow_frame);
+  HANDLE_INSTRUCTION_END();
+
   HANDLE_INSTRUCTION_START(UNUSED_FA)
     UnexpectedOpcode(inst, shadow_frame);
   HANDLE_INSTRUCTION_END();
diff --git a/runtime/interpreter/interpreter_switch_impl.cc b/runtime/interpreter/interpreter_switch_impl.cc
index 083dfb5..b5cc11e 100644
--- a/runtime/interpreter/interpreter_switch_impl.cc
+++ b/runtime/interpreter/interpreter_switch_impl.cc
@@ -14,12 +14,9 @@
  * limitations under the License.
  */
 
-#include "base/stl_util.h"  // MakeUnique
 #include "interpreter_common.h"
 #include "safe_math.h"
 
-#include <memory>  // std::unique_ptr
-
 namespace art {
 namespace interpreter {
 
@@ -85,11 +82,6 @@
   const uint16_t* const insns = code_item->insns_;
   const Instruction* inst = Instruction::At(insns + dex_pc);
   uint16_t inst_data;
-
-  // TODO: collapse capture-variable+create-lambda into one opcode, then we won't need
-  // to keep this live for the scope of the entire function call.
-  std::unique_ptr<lambda::ClosureBuilder> lambda_closure_builder;
-  size_t lambda_captured_variable_index = 0;
   while (true) {
     dex_pc = inst->GetDexPc(insns);
     shadow_frame.SetDexPC(dex_pc);
@@ -2243,63 +2235,19 @@
         POSSIBLY_HANDLE_PENDING_EXCEPTION(!success, Next_2xx);
         break;
       }
-      case Instruction::CAPTURE_VARIABLE: {
-        if (!IsExperimentalInstructionEnabled(inst)) {
-          UnexpectedOpcode(inst, shadow_frame);
-        }
-
-        if (lambda_closure_builder == nullptr) {
-          lambda_closure_builder = MakeUnique<lambda::ClosureBuilder>();
-        }
-
-        PREAMBLE();
-        bool success = DoCaptureVariable<do_access_check>(self,
-                                                          inst,
-                                                          /*inout*/shadow_frame,
-                                                          /*inout*/lambda_closure_builder.get());
-        POSSIBLY_HANDLE_PENDING_EXCEPTION(!success, Next_2xx);
-        break;
-      }
       case Instruction::CREATE_LAMBDA: {
         if (!IsExperimentalInstructionEnabled(inst)) {
           UnexpectedOpcode(inst, shadow_frame);
         }
 
         PREAMBLE();
-
-        if (lambda_closure_builder == nullptr) {
-          // DoCreateLambda always needs a ClosureBuilder, even if it has 0 captured variables.
-          lambda_closure_builder = MakeUnique<lambda::ClosureBuilder>();
-        }
-
-        // TODO: these allocations should not leak, and the lambda method should not be local.
-        lambda::Closure* lambda_closure =
-            reinterpret_cast<lambda::Closure*>(alloca(lambda_closure_builder->GetSize()));
-        bool success = DoCreateLambda<do_access_check>(self,
-                                                       inst,
-                                                       /*inout*/shadow_frame,
-                                                       /*inout*/lambda_closure_builder.get(),
-                                                       /*inout*/lambda_closure);
-        lambda_closure_builder.reset(nullptr);  // reset state of variables captured
+        bool success = DoCreateLambda<do_access_check>(self, shadow_frame, inst);
         POSSIBLY_HANDLE_PENDING_EXCEPTION(!success, Next_2xx);
         break;
       }
-      case Instruction::LIBERATE_VARIABLE: {
-        if (!IsExperimentalInstructionEnabled(inst)) {
-          UnexpectedOpcode(inst, shadow_frame);
-        }
-
-        PREAMBLE();
-        bool success = DoLiberateVariable<do_access_check>(self,
-                                                           inst,
-                                                           lambda_captured_variable_index,
-                                                           /*inout*/shadow_frame);
-        // Temporarily only allow sequences of 'liberate-variable, liberate-variable, ...'
-        lambda_captured_variable_index++;
-        POSSIBLY_HANDLE_PENDING_EXCEPTION(!success, Next_2xx);
-        break;
-      }
-      case Instruction::UNUSED_F4: {
+      case Instruction::UNUSED_F4:
+      case Instruction::UNUSED_F5:
+      case Instruction::UNUSED_F7: {
         if (!IsExperimentalInstructionEnabled(inst)) {
           UnexpectedOpcode(inst, shadow_frame);
         }
diff --git a/runtime/lambda/art_lambda_method.h b/runtime/lambda/art_lambda_method.h
index ea13eb7..892d8c6 100644
--- a/runtime/lambda/art_lambda_method.h
+++ b/runtime/lambda/art_lambda_method.h
@@ -35,7 +35,7 @@
   // (Ownership of strings is retained by the caller and the lifetime should exceed this class).
   ArtLambdaMethod(ArtMethod* target_method,
                   const char* captured_variables_type_descriptor,
-                  const char* captured_variables_shorty,
+                  const char* captured_variables_shorty_,
                   bool innate_lambda = true);
 
   // Get the target method for this lambda that would be used by the invoke-lambda dex instruction.
diff --git a/runtime/lambda/box_table.cc b/runtime/lambda/box_table.cc
index 8eef10b..26575fd 100644
--- a/runtime/lambda/box_table.cc
+++ b/runtime/lambda/box_table.cc
@@ -18,8 +18,6 @@
 #include "base/mutex.h"
 #include "common_throws.h"
 #include "gc_root-inl.h"
-#include "lambda/closure.h"
-#include "lambda/leaking_allocator.h"
 #include "mirror/method.h"
 #include "mirror/object-inl.h"
 #include "thread.h"
@@ -28,53 +26,11 @@
 
 namespace art {
 namespace lambda {
-// Temporarily represent the lambda Closure as its raw bytes in an array.
-// TODO: Generate a proxy class for the closure when boxing the first time.
-using BoxedClosurePointerType = mirror::ByteArray*;
-
-static mirror::Class* GetBoxedClosureClass() SHARED_REQUIRES(Locks::mutator_lock_) {
-  return mirror::ByteArray::GetArrayClass();
-}
-
-namespace {
-  // Convenience functions to allocating/deleting box table copies of the closures.
-  struct ClosureAllocator {
-    // Deletes a Closure that was allocated through ::Allocate.
-    static void Delete(Closure* ptr) {
-      delete[] reinterpret_cast<char*>(ptr);
-    }
-
-    // Returns a well-aligned pointer to a newly allocated Closure on the 'new' heap.
-    static Closure* Allocate(size_t size) {
-      DCHECK_GE(size, sizeof(Closure));
-
-      // TODO: Maybe point to the interior of the boxed closure object after we add proxy support?
-      Closure* closure = reinterpret_cast<Closure*>(new char[size]);
-      DCHECK_ALIGNED(closure, alignof(Closure));
-      return closure;
-    }
-  };
-}  // namespace
 
 BoxTable::BoxTable()
   : allow_new_weaks_(true),
     new_weaks_condition_("lambda box table allowed weaks", *Locks::lambda_table_lock_) {}
 
-BoxTable::~BoxTable() {
-  // Free all the copies of our closures.
-  for (auto map_iterator = map_.begin(); map_iterator != map_.end(); ++map_iterator) {
-    std::pair<UnorderedMapKeyType, ValueType>& key_value_pair = *map_iterator;
-
-    Closure* closure = key_value_pair.first;
-
-    // Remove from the map first, so that it doesn't try to access dangling pointer.
-    map_iterator = map_.Erase(map_iterator);
-
-    // Safe to delete, no dangling pointers.
-    ClosureAllocator::Delete(closure);
-  }
-}
-
 mirror::Object* BoxTable::BoxLambda(const ClosureType& closure) {
   Thread* self = Thread::Current();
 
@@ -102,29 +58,22 @@
 
   // Release the lambda table lock here, so that thread suspension is allowed.
 
-  // Convert the Closure into a managed byte[] which will serve
+  // Convert the ArtMethod into a java.lang.reflect.Method which will serve
   // as the temporary 'boxed' version of the lambda. This is good enough
   // to check all the basic object identities that a boxed lambda must retain.
-  // It's also good enough to contain all the captured primitive variables.
 
   // TODO: Boxing an innate lambda (i.e. made with create-lambda) should make a proxy class
   // TODO: Boxing a learned lambda (i.e. made with unbox-lambda) should return the original object
-  BoxedClosurePointerType closure_as_array_object =
-      mirror::ByteArray::Alloc(self, closure->GetSize());
-
+  mirror::Method* method_as_object =
+      mirror::Method::CreateFromArtMethod(self, closure);
   // There are no thread suspension points after this, so we don't need to put it into a handle.
 
-  if (UNLIKELY(closure_as_array_object == nullptr)) {
+  if (UNLIKELY(method_as_object == nullptr)) {
     // Most likely an OOM has occurred.
     CHECK(self->IsExceptionPending());
     return nullptr;
   }
 
-  // Write the raw closure data into the byte[].
-  closure->CopyTo(closure_as_array_object->GetRawData(sizeof(uint8_t),  // component size
-                                                      0 /*index*/),     // index
-                  closure_as_array_object->GetLength());
-
   // The method has been successfully boxed into an object, now insert it into the hash map.
   {
     MutexLock mu(self, *Locks::lambda_table_lock_);
@@ -138,56 +87,38 @@
       return value.Read();
     }
 
-    // Otherwise we need to insert it into the hash map in this thread.
-
-    // Make a copy for the box table to keep, in case the closure gets collected from the stack.
-    // TODO: GC may need to sweep for roots in the box table's copy of the closure.
-    Closure* closure_table_copy = ClosureAllocator::Allocate(closure->GetSize());
-    closure->CopyTo(closure_table_copy, closure->GetSize());
-
-    // The closure_table_copy needs to be deleted by us manually when we erase it from the map.
-
-    // Actually insert into the table.
-    map_.Insert({closure_table_copy, ValueType(closure_as_array_object)});
+    // Otherwise we should insert it into the hash map in this thread.
+    map_.Insert(std::make_pair(closure, ValueType(method_as_object)));
   }
 
-  return closure_as_array_object;
+  return method_as_object;
 }
 
 bool BoxTable::UnboxLambda(mirror::Object* object, ClosureType* out_closure) {
   DCHECK(object != nullptr);
   *out_closure = nullptr;
 
-  Thread* self = Thread::Current();
-
   // Note that we do not need to access lambda_table_lock_ here
   // since we don't need to look at the map.
 
   mirror::Object* boxed_closure_object = object;
 
-  // Raise ClassCastException if object is not instanceof byte[]
-  if (UNLIKELY(!boxed_closure_object->InstanceOf(GetBoxedClosureClass()))) {
-    ThrowClassCastException(GetBoxedClosureClass(), boxed_closure_object->GetClass());
+  // Raise ClassCastException if object is not instanceof java.lang.reflect.Method
+  if (UNLIKELY(!boxed_closure_object->InstanceOf(mirror::Method::StaticClass()))) {
+    ThrowClassCastException(mirror::Method::StaticClass(), boxed_closure_object->GetClass());
     return false;
   }
 
   // TODO(iam): We must check that the closure object extends/implements the type
-  // specified in [type id]. This is not currently implemented since it's always a byte[].
+  // specified in [type id]. This is not currently implemented since it's always a Method.
 
   // If we got this far, the inputs are valid.
-  // Shuffle the byte[] back into a raw closure, then allocate it, copy, and return it.
-  BoxedClosurePointerType boxed_closure_as_array =
-      down_cast<BoxedClosurePointerType>(boxed_closure_object);
+  // Write out the java.lang.reflect.Method's embedded ArtMethod* into the vreg target.
+  mirror::AbstractMethod* boxed_closure_as_method =
+      down_cast<mirror::AbstractMethod*>(boxed_closure_object);
 
-  const int8_t* unaligned_interior_closure = boxed_closure_as_array->GetData();
-
-  // Allocate a copy that can "escape" and copy the closure data into that.
-  Closure* unboxed_closure =
-      LeakingAllocator::MakeFlexibleInstance<Closure>(self, boxed_closure_as_array->GetLength());
-  // TODO: don't just memcpy the closure, it's unsafe when we add references to the mix.
-  memcpy(unboxed_closure, unaligned_interior_closure, boxed_closure_as_array->GetLength());
-
-  DCHECK_EQ(unboxed_closure->GetSize(), static_cast<size_t>(boxed_closure_as_array->GetLength()));
+  ArtMethod* unboxed_closure = boxed_closure_as_method->GetArtMethod();
+  DCHECK(unboxed_closure != nullptr);
 
   *out_closure = unboxed_closure;
   return true;
@@ -196,7 +127,7 @@
 BoxTable::ValueType BoxTable::FindBoxedLambda(const ClosureType& closure) const {
   auto map_iterator = map_.Find(closure);
   if (map_iterator != map_.end()) {
-    const std::pair<UnorderedMapKeyType, ValueType>& key_value_pair = *map_iterator;
+    const std::pair<ClosureType, ValueType>& key_value_pair = *map_iterator;
     const ValueType& value = key_value_pair.second;
 
     DCHECK(!value.IsNull());  // Never store null boxes.
@@ -226,7 +157,7 @@
    */
   std::vector<ClosureType> remove_list;
   for (auto map_iterator = map_.begin(); map_iterator != map_.end(); ) {
-    std::pair<UnorderedMapKeyType, ValueType>& key_value_pair = *map_iterator;
+    std::pair<ClosureType, ValueType>& key_value_pair = *map_iterator;
 
     const ValueType& old_value = key_value_pair.second;
 
@@ -235,15 +166,10 @@
     mirror::Object* new_value = visitor->IsMarked(old_value_raw);
 
     if (new_value == nullptr) {
-      // The object has been swept away.
       const ClosureType& closure = key_value_pair.first;
-
+      // The object has been swept away.
       // Delete the entry from the map.
-      map_iterator = map_.Erase(map_iterator);
-
-      // Clean up the memory by deleting the closure.
-      ClosureAllocator::Delete(closure);
-
+      map_iterator = map_.Erase(map_.Find(closure));
     } else {
       // The object has been moved.
       // Update the map.
@@ -282,33 +208,16 @@
   new_weaks_condition_.Broadcast(self);
 }
 
-void BoxTable::EmptyFn::MakeEmpty(std::pair<UnorderedMapKeyType, ValueType>& item) const {
-  item.first = nullptr;
-
-  Locks::mutator_lock_->AssertSharedHeld(Thread::Current());
-  item.second = ValueType();  // Also clear the GC root.
-}
-
-bool BoxTable::EmptyFn::IsEmpty(const std::pair<UnorderedMapKeyType, ValueType>& item) const {
-  return item.first == nullptr;
-}
-
-bool BoxTable::EqualsFn::operator()(const UnorderedMapKeyType& lhs,
-                                    const UnorderedMapKeyType& rhs) const {
+bool BoxTable::EqualsFn::operator()(const ClosureType& lhs, const ClosureType& rhs) const {
   // Nothing needs this right now, but leave this assertion for later when
   // we need to look at the references inside of the closure.
-  Locks::mutator_lock_->AssertSharedHeld(Thread::Current());
+  if (kIsDebugBuild) {
+    Locks::mutator_lock_->AssertSharedHeld(Thread::Current());
+  }
 
-  return lhs->ReferenceEquals(rhs);
-}
-
-size_t BoxTable::HashFn::operator()(const UnorderedMapKeyType& key) const {
-  const lambda::Closure* closure = key;
-  DCHECK_ALIGNED(closure, alignof(lambda::Closure));
-
-  // Need to hold mutator_lock_ before calling into Closure::GetHashCode.
-  Locks::mutator_lock_->AssertSharedHeld(Thread::Current());
-  return closure->GetHashCode();
+  // TODO: Need rework to use read barriers once closures have references inside of them that can
+  // move. Until then, it's safe to just compare the data inside of it directly.
+  return lhs == rhs;
 }
 
 }  // namespace lambda
diff --git a/runtime/lambda/box_table.h b/runtime/lambda/box_table.h
index adb7332..9ffda66 100644
--- a/runtime/lambda/box_table.h
+++ b/runtime/lambda/box_table.h
@@ -34,7 +34,6 @@
 }  // namespace mirror
 
 namespace lambda {
-struct Closure;  // forward declaration
 
 /*
  * Store a table of boxed lambdas. This is required to maintain object referential equality
@@ -45,7 +44,7 @@
  */
 class BoxTable FINAL {
  public:
-  using ClosureType = art::lambda::Closure*;
+  using ClosureType = art::ArtMethod*;
 
   // Boxes a closure into an object. Returns null and throws an exception on failure.
   mirror::Object* BoxLambda(const ClosureType& closure)
@@ -73,9 +72,10 @@
       REQUIRES(!Locks::lambda_table_lock_);
 
   BoxTable();
-  ~BoxTable();
+  ~BoxTable() = default;
 
  private:
+  // FIXME: This needs to be a GcRoot.
   // Explanation:
   // - After all threads are suspended (exclusive mutator lock),
   //   the concurrent-copying GC can move objects from the "from" space to the "to" space.
@@ -97,30 +97,30 @@
   void BlockUntilWeaksAllowed()
       SHARED_REQUIRES(Locks::lambda_table_lock_);
 
-  // Wrap the Closure into a unique_ptr so that the HashMap can delete its memory automatically.
-  using UnorderedMapKeyType = ClosureType;
-
   // EmptyFn implementation for art::HashMap
   struct EmptyFn {
-    void MakeEmpty(std::pair<UnorderedMapKeyType, ValueType>& item) const
-        NO_THREAD_SAFETY_ANALYSIS;  // SHARED_REQUIRES(Locks::mutator_lock_)
-
-    bool IsEmpty(const std::pair<UnorderedMapKeyType, ValueType>& item) const;
+    void MakeEmpty(std::pair<ClosureType, ValueType>& item) const {
+      item.first = nullptr;
+    }
+    bool IsEmpty(const std::pair<ClosureType, ValueType>& item) const {
+      return item.first == nullptr;
+    }
   };
 
   // HashFn implementation for art::HashMap
   struct HashFn {
-    size_t operator()(const UnorderedMapKeyType& key) const
-        NO_THREAD_SAFETY_ANALYSIS;  // SHARED_REQUIRES(Locks::mutator_lock_)
+    size_t operator()(const ClosureType& key) const {
+      // TODO(iam): Rewrite hash function when ClosureType is no longer an ArtMethod*
+      return static_cast<size_t>(reinterpret_cast<uintptr_t>(key));
+    }
   };
 
   // EqualsFn implementation for art::HashMap
   struct EqualsFn {
-    bool operator()(const UnorderedMapKeyType& lhs, const UnorderedMapKeyType& rhs) const
-        NO_THREAD_SAFETY_ANALYSIS;  // SHARED_REQUIRES(Locks::mutator_lock_)
+    bool operator()(const ClosureType& lhs, const ClosureType& rhs) const;
   };
 
-  using UnorderedMap = art::HashMap<UnorderedMapKeyType,
+  using UnorderedMap = art::HashMap<ClosureType,
                                     ValueType,
                                     EmptyFn,
                                     HashFn,
diff --git a/runtime/lambda/closure.cc b/runtime/lambda/closure.cc
index 179e4ee..95a17c6 100644
--- a/runtime/lambda/closure.cc
+++ b/runtime/lambda/closure.cc
@@ -124,55 +124,6 @@
   memcpy(target, this, GetSize());
 }
 
-ArtMethod* Closure::GetTargetMethod() const {
-  return const_cast<ArtMethod*>(lambda_info_->GetArtMethod());
-}
-
-uint32_t Closure::GetHashCode() const {
-  // Start with a non-zero constant, a prime number.
-  uint32_t result = 17;
-
-  // Include the hash with the ArtMethod.
-  {
-    uintptr_t method = reinterpret_cast<uintptr_t>(GetTargetMethod());
-    result = 31 * result + Low32Bits(method);
-    if (sizeof(method) == sizeof(uint64_t)) {
-      result = 31 * result + High32Bits(method);
-    }
-  }
-
-  // Include a hash for each captured variable.
-  for (size_t i = 0; i < GetCapturedVariablesSize(); ++i) {
-    // TODO: not safe for GC-able values since the address can move and the hash code would change.
-    uint8_t captured_variable_raw_value;
-    CopyUnsafeAtOffset<uint8_t>(i, /*out*/&captured_variable_raw_value);  // NOLINT: [whitespace/comma] [3]
-
-    result = 31 * result + captured_variable_raw_value;
-  }
-
-  // TODO: Fix above loop to work for objects and lambdas.
-  static_assert(kClosureSupportsGarbageCollection == false,
-               "Need to update above loop to read the hash code from the "
-                "objects and lambdas recursively");
-
-  return result;
-}
-
-bool Closure::ReferenceEquals(const Closure* other) const {
-  DCHECK(other != nullptr);
-
-  // TODO: Need rework to use read barriers once closures have references inside of them that can
-  // move. Until then, it's safe to just compare the data inside of it directly.
-  static_assert(kClosureSupportsReferences == false,
-                "Unsafe to use memcmp in read barrier collector");
-
-  if (GetSize() != other->GetSize()) {
-    return false;
-  }
-
-  return memcmp(this, other, GetSize());
-}
-
 size_t Closure::GetNumberOfCapturedVariables() const {
   // TODO: refactor into art_lambda_method.h. Parsing should only be required here as a DCHECK.
   VariableInfo variable_info =
diff --git a/runtime/lambda/closure.h b/runtime/lambda/closure.h
index 31ff194..60d117e 100644
--- a/runtime/lambda/closure.h
+++ b/runtime/lambda/closure.h
@@ -49,19 +49,6 @@
   // The target_size must be at least as large as GetSize().
   void CopyTo(void* target, size_t target_size) const;
 
-  // Get the target method, i.e. the method that will be dispatched into with invoke-lambda.
-  ArtMethod* GetTargetMethod() const;
-
-  // Calculates the hash code. Value is recomputed each time.
-  uint32_t GetHashCode() const SHARED_REQUIRES(Locks::mutator_lock_);
-
-  // Is this the same closure as other? e.g. same target method, same variables captured.
-  //
-  // Determines whether the two Closures are interchangeable instances.
-  // Does *not* call Object#equals recursively. If two Closures compare ReferenceEquals true that
-  // means that they are interchangeable values (usually for the purpose of boxing/unboxing).
-  bool ReferenceEquals(const Closure* other) const SHARED_REQUIRES(Locks::mutator_lock_);
-
   // How many variables were captured?
   size_t GetNumberOfCapturedVariables() const;
 
diff --git a/runtime/lambda/closure_builder-inl.h b/runtime/lambda/closure_builder-inl.h
index 3cec21f..41a803b 100644
--- a/runtime/lambda/closure_builder-inl.h
+++ b/runtime/lambda/closure_builder-inl.h
@@ -35,8 +35,6 @@
 
   values_.push_back(value_storage);
   size_ += sizeof(T);
-
-  shorty_types_ += kShortyType;
 }
 
 }  // namespace lambda
diff --git a/runtime/lambda/closure_builder.cc b/runtime/lambda/closure_builder.cc
index 739e965..9c37db8 100644
--- a/runtime/lambda/closure_builder.cc
+++ b/runtime/lambda/closure_builder.cc
@@ -64,8 +64,6 @@
       UNIMPLEMENTED(FATAL) << "can't yet safely capture objects with read barrier";
     }
   }
-
-  shorty_types_ += ShortyFieldType::kObject;
 }
 
 void ClosureBuilder::CaptureVariableLambda(Closure* closure) {
@@ -80,8 +78,6 @@
 
   // A closure may be sized dynamically, so always query it for the true size.
   size_ += closure->GetSize();
-
-  shorty_types_ += ShortyFieldType::kLambda;
 }
 
 size_t ClosureBuilder::GetSize() const {
@@ -89,15 +85,9 @@
 }
 
 size_t ClosureBuilder::GetCaptureCount() const {
-  DCHECK_EQ(values_.size(), shorty_types_.size());
   return values_.size();
 }
 
-const std::string& ClosureBuilder::GetCapturedVariableShortyTypes() const {
-  DCHECK_EQ(values_.size(), shorty_types_.size());
-  return shorty_types_;
-}
-
 Closure* ClosureBuilder::CreateInPlace(void* memory, ArtLambdaMethod* target_method) const {
   DCHECK(memory != nullptr);
   DCHECK(target_method != nullptr);
@@ -148,14 +138,11 @@
                                    size_t variables_size) const {
   size_t total_size = header_size;
   const char* shorty_types = target_method->GetCapturedVariablesShortyTypeDescriptor();
-  DCHECK_STREQ(shorty_types, shorty_types_.c_str());
 
   size_t variables_offset = 0;
   size_t remaining_size = variables_size;
 
   const size_t shorty_count = target_method->GetNumberOfCapturedVariables();
-  DCHECK_EQ(shorty_count, GetCaptureCount());
-
   for (size_t i = 0; i < shorty_count; ++i) {
     ShortyFieldType shorty{shorty_types[i]};  // NOLINT [readability/braces] [4]
 
diff --git a/runtime/lambda/closure_builder.h b/runtime/lambda/closure_builder.h
index 23eb484..542e12a 100644
--- a/runtime/lambda/closure_builder.h
+++ b/runtime/lambda/closure_builder.h
@@ -40,12 +40,13 @@
 //
 // The mutator lock must be held for the duration of the lifetime of this object,
 // since it needs to temporarily store heap references into an internal list.
-class ClosureBuilder {
+class ClosureBuilder : ValueObject {
  public:
   using ShortyTypeEnum = decltype(ShortyFieldType::kByte);
 
+
   // Mark this primitive value to be captured as the specified type.
-  template <typename T, ShortyTypeEnum kShortyType = ShortyFieldTypeSelectEnum<T>::value>
+  template <typename T, ShortyTypeEnum kShortyType>
   void CaptureVariablePrimitive(T value);
 
   // Mark this object reference to be captured.
@@ -62,9 +63,6 @@
   // Returns how many variables have been captured so far.
   size_t GetCaptureCount() const;
 
-  // Get the list of captured variables' shorty field types.
-  const std::string& GetCapturedVariableShortyTypes() const;
-
   // Creates a closure in-place and writes out the data into 'memory'.
   // Memory must be at least 'GetSize' bytes large.
   // All previously marked data to be captured is now written out.
@@ -95,7 +93,6 @@
   size_t size_ = kInitialSize;
   bool is_dynamic_size_ = false;
   std::vector<ShortyFieldTypeTraits::MaxType> values_;
-  std::string shorty_types_;
 };
 
 }  // namespace lambda
diff --git a/runtime/lambda/leaking_allocator.cc b/runtime/lambda/leaking_allocator.cc
deleted file mode 100644
index 4910732..0000000
--- a/runtime/lambda/leaking_allocator.cc
+++ /dev/null
@@ -1,30 +0,0 @@
-/*
- * Copyright (C) 2015 The Android Open Source Project
- *
- * Licensed under the Apache License, Version 2.0 (the "License");
- * you may not use this file except in compliance with the License.
- * You may obtain a copy of the License at
- *
- *      http://www.apache.org/licenses/LICENSE-2.0
- *
- * Unless required by applicable law or agreed to in writing, software
- * distributed under the License is distributed on an "AS IS" BASIS,
- * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
- * See the License for the specific language governing permissions and
- * limitations under the License.
- */
-
-#include "lambda/leaking_allocator.h"
-#include "linear_alloc.h"
-#include "runtime.h"
-
-namespace art {
-namespace lambda {
-
-void* LeakingAllocator::AllocateMemory(Thread* self, size_t byte_size) {
-  // TODO: use GetAllocatorForClassLoader to allocate lambda ArtMethod data.
-  return Runtime::Current()->GetLinearAlloc()->Alloc(self, byte_size);
-}
-
-}  // namespace lambda
-}  // namespace art
diff --git a/runtime/lambda/leaking_allocator.h b/runtime/lambda/leaking_allocator.h
deleted file mode 100644
index c3222d0..0000000
--- a/runtime/lambda/leaking_allocator.h
+++ /dev/null
@@ -1,55 +0,0 @@
-/*
- * Copyright (C) 2015 The Android Open Source Project
- *
- * Licensed under the Apache License, Version 2.0 (the "License");
- * you may not use this file except in compliance with the License.
- * You may obtain a copy of the License at
- *
- *      http://www.apache.org/licenses/LICENSE-2.0
- *
- * Unless required by applicable law or agreed to in writing, software
- * distributed under the License is distributed on an "AS IS" BASIS,
- * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
- * See the License for the specific language governing permissions and
- * limitations under the License.
- */
-#ifndef ART_RUNTIME_LAMBDA_LEAKING_ALLOCATOR_H_
-#define ART_RUNTIME_LAMBDA_LEAKING_ALLOCATOR_H_
-
-#include <utility>  // std::forward
-
-namespace art {
-class Thread;  // forward declaration
-
-namespace lambda {
-
-// Temporary class to centralize all the leaking allocations.
-// Allocations made through this class are never freed, but it is a placeholder
-// that means that the calling code needs to be rewritten to properly:
-//
-// (a) Have a lifetime scoped to some other entity.
-// (b) Not be allocated over and over again if it was already allocated once (immutable data).
-//
-// TODO: do all of the above a/b for each callsite, and delete this class.
-class LeakingAllocator {
- public:
-  // Allocate byte_size bytes worth of memory. Never freed.
-  static void* AllocateMemory(Thread* self, size_t byte_size);
-
-  // Make a new instance of T, flexibly sized, in-place at newly allocated memory. Never freed.
-  template <typename T, typename... Args>
-  static T* MakeFlexibleInstance(Thread* self, size_t byte_size, Args&&... args) {
-    return new (AllocateMemory(self, byte_size)) T(std::forward<Args>(args)...);
-  }
-
-  // Make a new instance of T in-place at newly allocated memory. Never freed.
-  template <typename T, typename... Args>
-  static T* MakeInstance(Thread* self, Args&&... args) {
-    return new (AllocateMemory(self, sizeof(T))) T(std::forward<Args>(args)...);
-  }
-};
-
-}  // namespace lambda
-}  // namespace art
-
-#endif  // ART_RUNTIME_LAMBDA_LEAKING_ALLOCATOR_H_
diff --git a/runtime/verifier/method_verifier.cc b/runtime/verifier/method_verifier.cc
index 43c0b41..3d4f04c 100644
--- a/runtime/verifier/method_verifier.cc
+++ b/runtime/verifier/method_verifier.cc
@@ -1007,9 +1007,6 @@
     case Instruction::kVerifyRegCWide:
       result = result && CheckWideRegisterIndex(inst->VRegC());
       break;
-    case Instruction::kVerifyRegCString:
-      result = result && CheckStringIndex(inst->VRegC());
-      break;
   }
   switch (inst->GetVerifyExtraFlags()) {
     case Instruction::kVerifyArrayData:
@@ -3151,13 +3148,6 @@
       Fail(VERIFY_ERROR_FORCE_INTERPRETER);  // TODO(iam): implement invoke-lambda verification
       break;
     }
-    case Instruction::CAPTURE_VARIABLE: {
-      // Don't bother verifying, instead the interpreter will take the slow path with access checks.
-      // If the code would've normally hard-failed, then the interpreter will throw the
-      // appropriate verification errors at runtime.
-      Fail(VERIFY_ERROR_FORCE_INTERPRETER);  // TODO(iam): implement capture-variable verification
-      break;
-    }
     case Instruction::CREATE_LAMBDA: {
       // Don't bother verifying, instead the interpreter will take the slow path with access checks.
       // If the code would've normally hard-failed, then the interpreter will throw the
@@ -3165,15 +3155,10 @@
       Fail(VERIFY_ERROR_FORCE_INTERPRETER);  // TODO(iam): implement create-lambda verification
       break;
     }
-    case Instruction::LIBERATE_VARIABLE: {
-      // Don't bother verifying, instead the interpreter will take the slow path with access checks.
-      // If the code would've normally hard-failed, then the interpreter will throw the
-      // appropriate verification errors at runtime.
-      Fail(VERIFY_ERROR_FORCE_INTERPRETER);  // TODO(iam): implement liberate-variable verification
-      break;
-    }
 
-    case Instruction::UNUSED_F4: {
+    case Instruction::UNUSED_F4:
+    case Instruction::UNUSED_F5:
+    case Instruction::UNUSED_F7: {
       DCHECK(false);  // TODO(iam): Implement opcodes for lambdas
       // Conservatively fail verification on release builds.
       Fail(VERIFY_ERROR_BAD_CLASS_HARD) << "Unexpected opcode " << inst->DumpString(dex_file_);
diff --git a/test/955-lambda-smali/expected.txt b/test/955-lambda-smali/expected.txt
index 16381e4..3637099 100644
--- a/test/955-lambda-smali/expected.txt
+++ b/test/955-lambda-smali/expected.txt
@@ -16,13 +16,3 @@
 (MoveResult) testF success
 (MoveResult) testD success
 (MoveResult) testL success
-(CaptureVariables) (0-args, 1 captured variable 'Z'): value is true
-(CaptureVariables) (0-args, 1 captured variable 'B'): value is R
-(CaptureVariables) (0-args, 1 captured variable 'C'): value is ∂
-(CaptureVariables) (0-args, 1 captured variable 'S'): value is 1000
-(CaptureVariables) (0-args, 1 captured variable 'I'): value is 12345678
-(CaptureVariables) (0-args, 1 captured variable 'J'): value is 3287471278325742
-(CaptureVariables) (0-args, 1 captured variable 'F'): value is Infinity
-(CaptureVariables) (0-args, 1 captured variable 'D'): value is -Infinity
-(CaptureVariables) (0-args, 8 captured variable 'ZBCSIJFD'): value is true,R,∂,1000,12345678,3287471278325742,Infinity,-Infinity
-(CaptureVariables) Caught NPE
diff --git a/test/955-lambda-smali/smali/BoxUnbox.smali b/test/955-lambda-smali/smali/BoxUnbox.smali
index 915de2d..108b5fa 100644
--- a/test/955-lambda-smali/smali/BoxUnbox.smali
+++ b/test/955-lambda-smali/smali/BoxUnbox.smali
@@ -1,3 +1,4 @@
+#
 #  Copyright (C) 2015 The Android Open Source Project
 #
 #  Licensed under the Apache License, Version 2.0 (the "License");
@@ -35,8 +36,8 @@
 .end method
 
 #TODO: should use a closure type instead of ArtMethod.
-.method public static doHelloWorld(J)V
-    .registers 4 # 1 wide parameters, 2 locals
+.method public static doHelloWorld(Ljava/lang/reflect/ArtMethod;)V
+    .registers 3 # 1 parameters, 2 locals
 
     const-string v0, "(BoxUnbox) Hello boxing world! (0-args, no closure)"
 
@@ -50,9 +51,9 @@
 .method private static testBox()V
     .registers 3
 
-    create-lambda v0, LBoxUnbox;->doHelloWorld(J)V
+    create-lambda v0, LBoxUnbox;->doHelloWorld(Ljava/lang/reflect/ArtMethod;)V
     box-lambda v2, v0 # v2 = box(v0)
-    unbox-lambda v0, v2, J # v0 = unbox(v2)
+    unbox-lambda v0, v2, Ljava/lang/reflect/ArtMethod; # v0 = unbox(v2)
     invoke-lambda v0, {}
 
     return-void
@@ -62,7 +63,7 @@
 .method private static testBoxEquality()V
    .registers 6 # 0 parameters, 6 locals
 
-    create-lambda v0, LBoxUnbox;->doHelloWorld(J)V
+    create-lambda v0, LBoxUnbox;->doHelloWorld(Ljava/lang/reflect/ArtMethod;)V
     box-lambda v2, v0 # v2 = box(v0)
     box-lambda v3, v0 # v3 = box(v0)
 
@@ -94,7 +95,7 @@
     const v0, 0  # v0 = null
     const v1, 0  # v1 = null
 :start
-    unbox-lambda v2, v0, J
+    unbox-lambda v2, v0, Ljava/lang/reflect/ArtMethod;
     # attempting to unbox a null lambda will throw NPE
 :end
     return-void
@@ -139,7 +140,7 @@
     const-string v0, "This is not a boxed lambda"
 :start
     # TODO: use \FunctionalType; here instead
-    unbox-lambda v2, v0, J
+    unbox-lambda v2, v0, Ljava/lang/reflect/ArtMethod;
     # can't use a string, expects a lambda object here. throws ClassCastException.
 :end
     return-void
diff --git a/test/955-lambda-smali/smali/Main.smali b/test/955-lambda-smali/smali/Main.smali
index 9892d61..5d2aabb 100644
--- a/test/955-lambda-smali/smali/Main.smali
+++ b/test/955-lambda-smali/smali/Main.smali
@@ -24,7 +24,6 @@
     invoke-static {}, LTrivialHelloWorld;->run()V
     invoke-static {}, LBoxUnbox;->run()V
     invoke-static {}, LMoveResult;->run()V
-    invoke-static {}, LCaptureVariables;->run()V
 
 # TODO: add tests when verification fails
 
diff --git a/test/955-lambda-smali/smali/MoveResult.smali b/test/955-lambda-smali/smali/MoveResult.smali
index 52f7ba3..1725da3 100644
--- a/test/955-lambda-smali/smali/MoveResult.smali
+++ b/test/955-lambda-smali/smali/MoveResult.smali
@@ -41,7 +41,7 @@
 .method public static testZ()V
     .registers 6
 
-    create-lambda v0, LMoveResult;->lambdaZ(J)Z
+    create-lambda v0, LMoveResult;->lambdaZ(Ljava/lang/reflect/ArtMethod;)Z
     invoke-lambda v0, {}
     move-result v2
     const v3, 1
@@ -61,7 +61,7 @@
 .end method
 
 # Lambda target for testZ. Always returns "true".
-.method public static lambdaZ(J)Z
+.method public static lambdaZ(Ljava/lang/reflect/ArtMethod;)Z
     .registers 3
 
     const v0, 1
@@ -73,7 +73,7 @@
 .method public static testB()V
     .registers 6
 
-    create-lambda v0, LMoveResult;->lambdaB(J)B
+    create-lambda v0, LMoveResult;->lambdaB(Ljava/lang/reflect/ArtMethod;)B
     invoke-lambda v0, {}
     move-result v2
     const v3, 15
@@ -93,7 +93,7 @@
 .end method
 
 # Lambda target for testB. Always returns "15".
-.method public static lambdaB(J)B
+.method public static lambdaB(Ljava/lang/reflect/ArtMethod;)B
     .registers 3 # 1 parameters, 2 locals
 
     const v0, 15
@@ -105,7 +105,7 @@
 .method public static testS()V
     .registers 6
 
-    create-lambda v0, LMoveResult;->lambdaS(J)S
+    create-lambda v0, LMoveResult;->lambdaS(Ljava/lang/reflect/ArtMethod;)S
     invoke-lambda v0, {}
     move-result v2
     const/16 v3, 31000
@@ -125,7 +125,7 @@
 .end method
 
 # Lambda target for testS. Always returns "31000".
-.method public static lambdaS(J)S
+.method public static lambdaS(Ljava/lang/reflect/ArtMethod;)S
     .registers 3
 
     const/16 v0, 31000
@@ -137,7 +137,7 @@
 .method public static testI()V
     .registers 6
 
-    create-lambda v0, LMoveResult;->lambdaI(J)I
+    create-lambda v0, LMoveResult;->lambdaI(Ljava/lang/reflect/ArtMethod;)I
     invoke-lambda v0, {}
     move-result v2
     const v3, 128000
@@ -157,7 +157,7 @@
 .end method
 
 # Lambda target for testI. Always returns "128000".
-.method public static lambdaI(J)I
+.method public static lambdaI(Ljava/lang/reflect/ArtMethod;)I
     .registers 3
 
     const v0, 128000
@@ -167,9 +167,9 @@
 
 # Test that chars are returned correctly via move-result.
 .method public static testC()V
-    .registers 7
+    .registers 6
 
-    create-lambda v0, LMoveResult;->lambdaC(J)C
+    create-lambda v0, LMoveResult;->lambdaC(Ljava/lang/reflect/ArtMethod;)C
     invoke-lambda v0, {}
     move-result v2
     const v3, 65535
@@ -189,7 +189,7 @@
 .end method
 
 # Lambda target for testC. Always returns "65535".
-.method public static lambdaC(J)C
+.method public static lambdaC(Ljava/lang/reflect/ArtMethod;)C
     .registers 3
 
     const v0, 65535
@@ -199,12 +199,12 @@
 
 # Test that longs are returned correctly via move-result.
 .method public static testJ()V
-    .registers 9
+    .registers 8
 
-    create-lambda v0, LMoveResult;->lambdaJ(J)J
+    create-lambda v0, LMoveResult;->lambdaJ(Ljava/lang/reflect/ArtMethod;)J
     invoke-lambda v0, {}
     move-result v2
-    const-wide v4, 0xdeadf00dc0ffeeL
+    const-wide v4, 0xdeadf00dc0ffee
 
     if-ne v4, v2, :is_not_equal
     const-string v6, "(MoveResult) testJ success"
@@ -220,11 +220,11 @@
 
 .end method
 
-# Lambda target for testC. Always returns "0xdeadf00dc0ffeeL".
-.method public static lambdaJ(J)J
-    .registers 5
+# Lambda target for testC. Always returns "0xdeadf00dc0ffee".
+.method public static lambdaJ(Ljava/lang/reflect/ArtMethod;)J
+    .registers 4
 
-    const-wide v0, 0xdeadf00dc0ffeeL
+    const-wide v0, 0xdeadf00dc0ffee
     return-wide v0
 
 .end method
@@ -233,7 +233,7 @@
 .method public static testF()V
     .registers 6
 
-    create-lambda v0, LMoveResult;->lambdaF(J)F
+    create-lambda v0, LMoveResult;->lambdaF(Ljava/lang/reflect/ArtMethod;)F
     invoke-lambda v0, {}
     move-result v2
     const v3, infinityf
@@ -253,8 +253,8 @@
 .end method
 
 # Lambda target for testF. Always returns "infinityf".
-.method public static lambdaF(J)F
-    .registers 4
+.method public static lambdaF(Ljava/lang/reflect/ArtMethod;)F
+    .registers 3
 
     const v0, infinityf
     return v0
@@ -265,10 +265,10 @@
 .method public static testD()V
     .registers 8
 
-    create-lambda v0, LMoveResult;->lambdaD(J)D
+    create-lambda v0, LMoveResult;->lambdaD(Ljava/lang/reflect/ArtMethod;)D
     invoke-lambda v0, {}
     move-result-wide v2
-    const-wide v4, -infinity
+    const-wide v4, infinity
 
     if-ne v4, v2, :is_not_equal
     const-string v6, "(MoveResult) testD success"
@@ -285,10 +285,10 @@
 .end method
 
 # Lambda target for testD. Always returns "infinity".
-.method public static lambdaD(J)D
-    .registers 5
+.method public static lambdaD(Ljava/lang/reflect/ArtMethod;)D
+    .registers 4
 
-    const-wide v0, -infinity
+    const-wide v0, infinity # 123.456789
     return-wide v0
 
 .end method
@@ -298,7 +298,7 @@
 .method public static testL()V
     .registers 8
 
-    create-lambda v0, LMoveResult;->lambdaL(J)Ljava/lang/String;
+    create-lambda v0, LMoveResult;->lambdaL(Ljava/lang/reflect/ArtMethod;)Ljava/lang/String;
     invoke-lambda v0, {}
     move-result-object v2
     const-string v4, "Interned string"
@@ -319,8 +319,8 @@
 .end method
 
 # Lambda target for testL. Always returns "Interned string" (string).
-.method public static lambdaL(J)Ljava/lang/String;
-    .registers 5
+.method public static lambdaL(Ljava/lang/reflect/ArtMethod;)Ljava/lang/String;
+    .registers 4
 
     const-string v0, "Interned string"
     return-object v0
diff --git a/test/955-lambda-smali/smali/TrivialHelloWorld.smali b/test/955-lambda-smali/smali/TrivialHelloWorld.smali
index 3444b13..38ee95a 100644
--- a/test/955-lambda-smali/smali/TrivialHelloWorld.smali
+++ b/test/955-lambda-smali/smali/TrivialHelloWorld.smali
@@ -25,12 +25,12 @@
 .method public static run()V
 .registers 8
     # Trivial 0-arg hello world
-    create-lambda v0, LTrivialHelloWorld;->doHelloWorld(J)V
+    create-lambda v0, LTrivialHelloWorld;->doHelloWorld(Ljava/lang/reflect/ArtMethod;)V
     # TODO: create-lambda should not write to both v0 and v1
     invoke-lambda v0, {}
 
     # Slightly more interesting 4-arg hello world
-    create-lambda v2, doHelloWorldArgs(JLjava/lang/String;Ljava/lang/String;Ljava/lang/String;Ljava/lang/String;)V
+    create-lambda v2, doHelloWorldArgs(Ljava/lang/reflect/ArtMethod;Ljava/lang/String;Ljava/lang/String;Ljava/lang/String;Ljava/lang/String;)V
     # TODO: create-lambda should not write to both v2 and v3
     const-string v4, "A"
     const-string v5, "B"
@@ -43,9 +43,9 @@
     return-void
 .end method
 
-#TODO: should use a closure type instead of jlong. 
-.method public static doHelloWorld(J)V
-    .registers 5 # 1 wide parameters, 3 locals
+#TODO: should use a closure type instead of ArtMethod.
+.method public static doHelloWorld(Ljava/lang/reflect/ArtMethod;)V
+    .registers 3 # 1 parameters, 2 locals
 
     const-string v0, "Hello world! (0-args, no closure)"
 
@@ -55,17 +55,17 @@
     return-void
 .end method
 
-#TODO: should use a closure type instead of jlong. 
-.method public static doHelloWorldArgs(JLjava/lang/String;Ljava/lang/String;Ljava/lang/String;Ljava/lang/String;)V
-    .registers 9 # 1 wide parameter, 4 narrow parameters, 3 locals
+#TODO: should use a closure type instead of ArtMethod.
+.method public static doHelloWorldArgs(Ljava/lang/reflect/ArtMethod;Ljava/lang/String;Ljava/lang/String;Ljava/lang/String;Ljava/lang/String;)V
+    .registers 7 # 5 parameters, 2 locals
 
     const-string v0, " Hello world! (4-args, no closure)"
     sget-object v1, Ljava/lang/System;->out:Ljava/io/PrintStream;
 
+    invoke-virtual {v1, p1}, Ljava/io/PrintStream;->print(Ljava/lang/String;)V
     invoke-virtual {v1, p2}, Ljava/io/PrintStream;->print(Ljava/lang/String;)V
     invoke-virtual {v1, p3}, Ljava/io/PrintStream;->print(Ljava/lang/String;)V
     invoke-virtual {v1, p4}, Ljava/io/PrintStream;->print(Ljava/lang/String;)V
-    invoke-virtual {v1, p5}, Ljava/io/PrintStream;->print(Ljava/lang/String;)V
 
     invoke-virtual {v1, v0}, Ljava/io/PrintStream;->println(Ljava/lang/String;)V