diff --git a/std/assembly/array.ts b/std/assembly/array.ts
index 5e453f05..3811d5a6 100644
--- a/std/assembly/array.ts
+++ b/std/assembly/array.ts
@@ -1,34 +1,20 @@
-// import {
-//   MAX_BLENGTH,
-//   HEADER_SIZE,
-//   allocateUnsafe,
-//   reallocateUnsafe,
-//   LOAD,
-//   STORE
-// } from "./internal/arraybuffer";
-
 import {
   ALLOC,
   REALLOC,
   REGISTER,
   LINK,
-  ArrayBufferView
+  ArrayBufferView,
+  FREE
 } from "./runtime";
 
 import {
   ArrayBuffer
 } from "./arraybuffer";
 
-// import {
-//   allocateUnsafe as allocateUnsafeString,
-//   freeUnsafe as freeUnsafeString,
-//   copyUnsafe as copyUnsafeString
-// } from "./internal/string";
-
 import {
   COMPARATOR,
   SORT
-} from "./internal/sort";
+} from "./util/sort";
 
 import {
   itoa,
@@ -36,7 +22,7 @@ import {
   itoa_stream,
   dtoa_stream,
   MAX_DOUBLE_LENGTH
-} from "./internal/number";
+} from "./util/number";
 
 import {
   isArray as builtin_isArray
@@ -372,186 +358,233 @@ export class Array<T> extends ArrayBufferView {
   }
 
   reverse(): Array<T> {
-    var buffer = this.buffer_;
+    var base = this.dataStart;
     for (let front = 0, back = this.length_ - 1; front < back; ++front, --back) {
-      let temp = LOAD<T>(buffer, front);
-      STORE<T>(buffer, front, LOAD<T>(buffer, back));
-      STORE<T>(buffer, back, temp);
+      let temp: T = load<T>(base, front);
+      store<T>(base + (<usize>front << alignof<T>()), load<T>(base + (<usize>back << alignof<T>())));
+      store<T>(base + (<usize>back  << alignof<T>()), temp);
     }
     return this;
   }
 
   sort(comparator: (a: T, b: T) => i32 = COMPARATOR<T>()): this {
-    // TODO remove this when flow will allow trackcing null
+    // TODO remove this when flow will allow tracking null
     assert(comparator); // The comparison function must be a function
 
     var length = this.length_;
     if (length <= 1) return this;
-    var buffer = this.buffer_;
+    var base = this.dataStart;
     if (length == 2) {
-      let a = LOAD<T>(buffer, 1); // a = arr[1]
-      let b = LOAD<T>(buffer, 0); // b = arr[0]
+      let a: T = load<T>(base, sizeof<T>()); // a = arr[1]
+      let b: T = load<T>(base); // b = arr[0]
       if (comparator(a, b) < 0) {
-        STORE<T>(buffer, 1, b);   // arr[1] = b;
-        STORE<T>(buffer, 0, a);   // arr[0] = a;
+        store<T>(base, b, sizeof<T>()); // arr[1] = b;
+        store<T>(base, a); // arr[0] = a;
       }
       return this;
     }
-    SORT<T>(buffer, 0, length, comparator);
+    SORT<T>(base, length, comparator);
     return this;
   }
 
+  // FIXME: refactor into multiple functions?
   join(separator: string = ","): string {
     var lastIndex = this.length_ - 1;
     if (lastIndex < 0) return "";
     var result = "";
     var value: T;
-    var buffer = this.buffer_;
+    var base = this.dataStart;
+    // var buffer = this.buffer_;
     var sepLen = separator.length;
     var hasSeparator = sepLen != 0;
     if (value instanceof bool) {
-      if (!lastIndex) return select<string>("true", "false", LOAD<T,bool>(buffer, 0));
+      if (!lastIndex) return select("true", "false", load<T>(base));
 
       let valueLen = 5; // max possible length of element len("false")
       let estLen = (valueLen + sepLen) * lastIndex + valueLen;
-      let result = allocateUnsafeString(estLen);
+      let result = ALLOC(estLen << 1);
       let offset = 0;
       for (let i = 0; i < lastIndex; ++i) {
-        value = LOAD<T,bool>(buffer, i);
+        value = load<T>(base + i);
         valueLen = 4 + <i32>(!value);
-        copyUnsafeString(result, offset, select<string>("true", "false", value), 0, valueLen);
+        memory.copy(
+          result + (<usize>offset << 1),
+          changetype<usize>(select("true", "false", value)),
+          <usize>valueLen << 1
+        );
         offset += valueLen;
         if (hasSeparator) {
-          copyUnsafeString(result, offset, changetype<String>(separator), 0, sepLen);
+          memory.copy(
+            result + (<usize>offset << 1),
+            changetype<usize>(separator),
+            <usize>sepLen << 1
+          );
           offset += sepLen;
         }
       }
-      value = LOAD<T,bool>(buffer, lastIndex);
+      value = load<T>(base + <usize>lastIndex);
       valueLen = 4 + <i32>(!value);
-      copyUnsafeString(result, offset, select<string>("true", "false", value), 0, valueLen);
+      memory.copy(
+        result + (<usize>offset << 1),
+        changetype<usize>(select("true", "false", value)),
+        valueLen << 1
+      );
       offset += valueLen;
 
-      let out = result;
       if (estLen > offset) {
-        out = result.substring(0, offset);
-        freeUnsafeString(result);
+        let trimmed = changetype<string>(result).substring(0, offset);
+        FREE(result);
+        return trimmed; // registered in .substring
       }
-      return out;
+      return REGISTER<string>(result);
     } else if (isInteger<T>()) {
-      if (!lastIndex) return changetype<string>(itoa<T>(LOAD<T>(buffer, 0)));
+      if (!lastIndex) return changetype<string>(itoa<T>(load<T>(base)));
 
       const valueLen = (sizeof<T>() <= 4 ? 10 : 20) + <i32>isSigned<T>();
       let estLen = (valueLen + sepLen) * lastIndex + valueLen;
-      let result = allocateUnsafeString(estLen);
+      let result = ALLOC(estLen << 1);
       let offset = 0;
       for (let i = 0; i < lastIndex; ++i) {
-        value = LOAD<T>(buffer, i);
-        offset += itoa_stream<T>(changetype<usize>(result), offset, value);
+        value = load<T>(base + (<usize>i << alignof<T>()));
+        offset += itoa_stream<T>(result, offset, value);
         if (hasSeparator) {
-          copyUnsafeString(result, offset, separator, 0, sepLen);
+          memory.copy(
+            result + (<usize>offset << 1),
+            changetype<usize>(separator),
+            <usize>sepLen << 1
+          );
           offset += sepLen;
         }
       }
-      value = LOAD<T>(buffer, lastIndex);
-      offset += itoa_stream<T>(changetype<usize>(result), offset, value);
-      let out = result;
+      value = load<T>(base + (<usize>lastIndex << alignof<T>()));
+      offset += itoa_stream<T>(result, offset, value);
       if (estLen > offset) {
-        out = result.substring(0, offset);
-        freeUnsafeString(result);
+        let trimmed = changetype<string>(result).substring(0, offset);
+        FREE(result);
+        return trimmed; // registered in .substring
       }
-      return out;
+      return REGISTER<string>(result);
     } else if (isFloat<T>()) {
-      if (!lastIndex) return changetype<string>(dtoa(LOAD<T,f64>(buffer, 0)));
+      if (!lastIndex) return changetype<string>(dtoa(load<T>(base)));
 
       const valueLen = MAX_DOUBLE_LENGTH;
       let estLen = (valueLen + sepLen) * lastIndex + valueLen;
-      let result = allocateUnsafeString(estLen);
+      let result = ALLOC(estLen << 1);
       let offset = 0;
       for (let i = 0; i < lastIndex; ++i) {
-        value = LOAD<T,f64>(buffer, i);
-        offset += dtoa_stream(changetype<usize>(result), offset, value);
+        value = load<T>(base + (<usize>i << alignof<T>()));
+        offset += dtoa_stream(result, offset, value);
         if (hasSeparator) {
-          copyUnsafeString(result, offset, separator, 0, sepLen);
+          memory.copy(
+            result + (<usize>offset << 1),
+            changetype<usize>(separator),
+            <usize>sepLen << 1
+          );
           offset += sepLen;
         }
       }
-      value = LOAD<T,f64>(buffer, lastIndex);
-      offset += dtoa_stream(changetype<usize>(result), offset, value);
-      let out = result;
+      value = load<T>(base + (<usize>lastIndex << alignof<T>()));
+      offset += dtoa_stream(result, offset, value);
       if (estLen > offset) {
-        out = result.substring(0, offset);
-        freeUnsafeString(result);
+        let trimmed = changetype<string>(result).substring(0, offset);
+        FREE(result);
+        return trimmed; // registered in .substring
       }
-      return out;
+      return REGISTER<string>(result);
     } else if (isString<T>()) {
-      if (!lastIndex) return LOAD<string>(buffer, 0);
+      if (!lastIndex) return load<string>(base);
 
       let estLen = 0;
       for (let i = 0, len = lastIndex + 1; i < len; ++i) {
-        estLen += LOAD<string>(buffer, i).length;
+        estLen += load<string>(base + (<usize>i << alignof<T>())).length;
       }
       let offset = 0;
-      let result = allocateUnsafeString(estLen + sepLen * lastIndex);
+      let result = ALLOC((estLen + sepLen * lastIndex) << 1);
       for (let i = 0; i < lastIndex; ++i) {
-        value = LOAD<string>(buffer, i);
+        value = load<string>(base + (<usize>i << alignof<T>()));
         if (value) {
-          let valueLen = value.length;                          // tslint:disable-line:no-unsafe-any
-          copyUnsafeString(result, offset, value, 0, valueLen); // tslint:disable-line:no-unsafe-any
-          offset += valueLen;                                   // tslint:disable-line:no-unsafe-any
+          let valueLen = changetype<string>(value).length;
+          memory.copy(
+            result + (<usize>offset << 1),
+            changetype<usize>(value),
+            <usize>valueLen << 1
+          );
+          offset += valueLen;
         }
         if (hasSeparator) {
-          copyUnsafeString(result, offset, separator, 0, sepLen);
+          memory.copy(
+            result + (<usize>offset << 1),
+            changetype<usize>(separator),
+            <usize>sepLen << 1
+          );
           offset += sepLen;
         }
       }
-      value = LOAD<string>(buffer, lastIndex);
+      value = load<string>(base + (<usize>lastIndex << alignof<T>()));
       if (value) {
-        let valueLen = value.length;                          // tslint:disable-line:no-unsafe-any
-        copyUnsafeString(result, offset, value, 0, valueLen); // tslint:disable-line:no-unsafe-any
+        let valueLen = changetype<string>(value).length;
+        memory.copy(
+          result + (<usize>offset << 1),
+          changetype<usize>(value),
+          <usize>valueLen << 1
+        );
       }
-      return result;
+      return REGISTER<string>(result);
     } else if (isArray<T>()) {
       if (!lastIndex) {
-        value = LOAD<T>(buffer, 0);
-        return value ? value.join(separator) : ""; // tslint:disable-line:no-unsafe-any
+        value = load<T>(base);
+        return value ? value.join(separator) : "";
       }
       for (let i = 0; i < lastIndex; ++i) {
-        value = LOAD<T>(buffer, i);
-        if (value) result += value.join(separator); // tslint:disable-line:no-unsafe-any
+        value = load<T>(base + (<usize>i << alignof<T>()));
+        if (value) result += value.join(separator);
         if (hasSeparator) result += separator;
       }
-      value = LOAD<T>(buffer, lastIndex);
-      if (value) result += value.join(separator); // tslint:disable-line:no-unsafe-any
-      return result;
+      value = load<T>(base + (<usize>lastIndex << alignof<T>()));
+      if (value) result += value.join(separator);
+      return result; // registered by concatenation (FIXME: lots of garbage)
     } else if (isReference<T>()) { // References
       if (!lastIndex) return "[object Object]";
       const valueLen = 15; // max possible length of element len("[object Object]")
       let estLen = (valueLen + sepLen) * lastIndex + valueLen;
-      let result = allocateUnsafeString(estLen);
+      let result = ALLOC(estLen << 1);
       let offset = 0;
       for (let i = 0; i < lastIndex; ++i) {
-        value = LOAD<T>(buffer, i);
+        value = load<T>(base + (<usize>i << alignof<T>()));
         if (value) {
-          copyUnsafeString(result, offset, changetype<String>("[object Object]"), 0, valueLen);
+          memory.copy(
+            result + (<usize>offset << 1),
+            changetype<usize>("[object Object]"),
+            <usize>valueLen << 1
+          );
           offset += valueLen;
         }
         if (hasSeparator) {
-          copyUnsafeString(result, offset, changetype<String>(separator), 0, sepLen);
+          memory.copy(
+            result + (<usize>offset << 1),
+            changetype<usize>(separator),
+            <usize>sepLen << 1
+          );
           offset += sepLen;
         }
       }
-      if (LOAD<T>(buffer, lastIndex)) {
-        copyUnsafeString(result, offset, changetype<String>("[object Object]"), 0, valueLen);
+      if (load<T>(base + (<usize>lastIndex << alignof<T>()))) {
+        memory.copy(
+          result + (<usize>offset << 1),
+          changetype<usize>("[object Object]"),
+          <usize>valueLen << 1
+        );
         offset += valueLen;
       }
-      let out = result;
       if (estLen > offset) {
-        out = result.substring(0, offset);
-        freeUnsafeString(result);
+        let out = changetype<string>(result).substring(0, offset);
+        FREE(result);
+        return out; // registered in .substring
       }
-      return out;
+      return REGISTER<string>(result);
     } else {
-      assert(false); // Unsupported generic typename
+      ERROR("unspported type");
+      assert(false);
     }
   }
 
@@ -560,16 +593,16 @@ export class Array<T> extends ArrayBufferView {
     return this.join();
   }
 
-  private __gc(): void {
-    var buffer = this.buffer_;
-    __gc_mark(changetype<usize>(buffer)); // tslint:disable-line
-    if (isManaged<T>()) {
-      let offset: usize = 0;
-      let end = <usize>this.length_ << alignof<usize>();
-      while (offset < end) {
-        __gc_mark(load<usize>(changetype<usize>(buffer) + offset, HEADER_SIZE)); // tslint:disable-line
-        offset += sizeof<usize>();
-      }
-    }
-  }
+  // private __gc(): void {
+  //   var buffer = this.buffer_;
+  //   __gc_mark(changetype<usize>(buffer)); // tslint:disable-line
+  //   if (isManaged<T>()) {
+  //     let offset: usize = 0;
+  //     let end = <usize>this.length_ << alignof<usize>();
+  //     while (offset < end) {
+  //       __gc_mark(load<usize>(changetype<usize>(buffer) + offset, HEADER_SIZE)); // tslint:disable-line
+  //       offset += sizeof<usize>();
+  //     }
+  //   }
+  // }
 }
diff --git a/std/assembly/internal/arraybuffer.ts b/std/assembly/internal/arraybuffer.ts
deleted file mode 100644
index 4545d8e0..00000000
--- a/std/assembly/internal/arraybuffer.ts
+++ /dev/null
@@ -1,84 +0,0 @@
-import {
-  AL_MASK,
-  MAX_SIZE_32
- } from "./allocator";
-
-/** Size of an ArrayBuffer header. */
-@inline export const HEADER_SIZE: usize = (offsetof<ArrayBuffer>() + AL_MASK) & ~AL_MASK;
-/** Maximum byte length of an ArrayBuffer. */
-@inline export const MAX_BLENGTH: i32 = <i32>MAX_SIZE_32 - HEADER_SIZE;
-
-function computeSize(byteLength: i32): usize {
-  // round up to power of 2, with HEADER_SIZE=8:
-  // 0            -> 2^3  = 8
-  // 1..8         -> 2^4  = 16
-  // 9..24        -> 2^5  = 32
-  // ...
-  // MAX_LENGTH   -> 2^30 = 0x40000000 (MAX_SIZE_32)
-  return <usize>1 << <usize>(<u32>32 - clz<u32>(byteLength + HEADER_SIZE - 1));
-}
-
-// Low-level utility
-
-function __gc(ref: usize): void {}
-
-export function allocateUnsafe(byteLength: i32): ArrayBuffer {
-  assert(<u32>byteLength <= <u32>MAX_BLENGTH);
-  var buffer: usize;
-  if (isManaged<ArrayBuffer>()) {
-    buffer = __gc_allocate(computeSize(byteLength), __gc); // tslint:disable-line
-  } else {
-    buffer = memory.allocate(computeSize(byteLength));
-  }
-  store<i32>(buffer, byteLength, offsetof<ArrayBuffer>("byteLength"));
-  return changetype<ArrayBuffer>(buffer);
-}
-
-export function reallocateUnsafe(buffer: ArrayBuffer, newByteLength: i32): ArrayBuffer {
-  var oldByteLength = buffer.byteLength;
-  if (newByteLength > oldByteLength) {
-    assert(newByteLength <= MAX_BLENGTH);
-    if (newByteLength <= <i32>(computeSize(oldByteLength) - HEADER_SIZE)) { // fast path: zero out additional space
-      store<i32>(changetype<usize>(buffer), newByteLength, offsetof<ArrayBuffer>("byteLength"));
-    } else { // slow path: copy to new buffer
-      let newBuffer = allocateUnsafe(newByteLength);
-      memory.copy(
-        changetype<usize>(newBuffer) + HEADER_SIZE,
-        changetype<usize>(buffer) + HEADER_SIZE,
-        <usize>oldByteLength
-      );
-      if (!isManaged<ArrayBuffer>()) {
-        memory.free(changetype<usize>(buffer));
-      }
-      buffer = newBuffer;
-    }
-    memory.fill(
-      changetype<usize>(buffer) + HEADER_SIZE + <usize>oldByteLength,
-      0,
-      <usize>(newByteLength - oldByteLength)
-    );
-  } else if (newByteLength < oldByteLength) { // fast path: override size
-    // TBD: worth to copy and release if size is significantly less than before?
-    assert(newByteLength >= 0);
-    store<i32>(changetype<usize>(buffer), newByteLength, offsetof<ArrayBuffer>("byteLength"));
-  }
-  return buffer;
-}
-
-// The helpers below use two different types in order to emit loads and stores that load respectively
-// store one type to/from memory while returning/taking the desired output/input type. This allows to
-// emit instructions like
-//
-// * `i32.load8` ^= `<i32>load<i8>(...)` that reads an i8 but returns an i32, or
-// * `i64.load32_s` ^= `<i64>load<i32>(...)`) that reads a 32-bit as a 64-bit integer
-//
-// without having to emit an additional instruction for conversion purposes. The second parameter
-// can be omitted for references and other loads and stores that simply return the exact type.
-
-@inline export function LOAD<T,TOut = T>(buffer: ArrayBuffer, index: i32, byteOffset: i32 = 0): TOut {
-  return <TOut>load<T>(changetype<usize>(buffer) + (<usize>index << alignof<T>()) + <usize>byteOffset, HEADER_SIZE);
-}
-
-@inline export function STORE<T,TIn = T>(buffer: ArrayBuffer, index: i32, value: TIn, byteOffset: i32 = 0): void {
-  store<T>(changetype<usize>(buffer) + (<usize>index << alignof<T>()) + <usize>byteOffset, value, HEADER_SIZE);
-}
diff --git a/std/assembly/internal/string.ts b/std/assembly/internal/string.ts
deleted file mode 100644
index 2e4fd81f..00000000
--- a/std/assembly/internal/string.ts
+++ /dev/null
@@ -1,225 +0,0 @@
-export function compareImpl(str1: String, index1: usize, str2: String, index2: usize, len: usize): i32 {
-  var result: i32 = 0;
-  var ptr1 = changetype<usize>(str1) + (index1 << 1);
-  var ptr2 = changetype<usize>(str2) + (index2 << 1);
-  while (len && !(result = <i32>load<u16>(ptr1) - <i32>load<u16>(ptr2))) {
-    --len, ptr1 += 2, ptr2 += 2;
-  }
-  return result;
-}
-
-import { MAX_SIZE_32 } from "./allocator";
-import { String } from "../string";
-
-/** Size of a String header. */
-@inline export const HEADER_SIZE = (offsetof<String>() + 1) & ~1; // 2 byte aligned
-/** Maximum length of a String. */
-@inline export const MAX_LENGTH = (<i32>MAX_SIZE_32 - HEADER_SIZE) >>> 1;
-
-// Low-level utility
-
-function __gc(ref: usize): void {}
-
-export function allocateUnsafe(length: i32): String {
-  assert(length > 0 && length <= MAX_LENGTH);
-  var buffer: usize;
-  if (isManaged<String>()) {
-    buffer = __gc_allocate(HEADER_SIZE + (<usize>length << 1), __gc);  // tslint:disable-line
-  } else {
-    buffer = memory.allocate(HEADER_SIZE + (<usize>length << 1));
-  }
-  store<i32>(buffer, length);
-  return changetype<String>(buffer);
-}
-
-@inline
-export function freeUnsafe(buffer: String): void {
-  if (!isManaged<String>()) {
-    assert(buffer);
-    memory.free(changetype<usize>(buffer));
-  }
-}
-
-export function copyUnsafe(dest: String, destOffset: usize, src: String, srcOffset: usize, len: usize): void {
-  memory.copy(
-    changetype<usize>(dest) + (destOffset << 1) + HEADER_SIZE,
-    changetype<usize>(src)  + (srcOffset  << 1) + HEADER_SIZE,
-    len << 1
-  );
-}
-
-export function repeatUnsafe(dest: String, destOffset: usize, src: String, count: i32): void {
-  var length = src.length;
-  if (ASC_SHRINK_LEVEL > 1) {
-    let strLen = length << 1;
-    let to   = changetype<usize>(dest) + HEADER_SIZE + (destOffset << 1);
-    let from = changetype<usize>(src)  + HEADER_SIZE;
-    for (let i = 0, len = strLen * count; i < len; i += strLen) {
-      memory.copy(to + i, from, strLen);
-    }
-  } else {
-    switch (length) {
-      case 0: break;
-      case 1: {
-        let cc =  load<u16>(changetype<usize>(src), HEADER_SIZE);
-        let out = changetype<usize>(dest) + (destOffset << 1);
-        for (let i = 0; i < count; ++i) {
-          store<u16>(out + (i << 1), cc, HEADER_SIZE);
-        }
-        break;
-      }
-      case 2: {
-        let cc  = load<u32>(changetype<usize>(src), HEADER_SIZE);
-        let out = changetype<usize>(dest) + (destOffset << 1);
-        for (let i = 0; i < count; ++i) {
-          store<u32>(out + (i << 2), cc, HEADER_SIZE);
-        }
-        break;
-      }
-      case 3: {
-        let cc1 = load<u32>(changetype<usize>(src), HEADER_SIZE + 0);
-        let cc2 = load<u16>(changetype<usize>(src), HEADER_SIZE + 4);
-        let out = changetype<usize>(dest) + (destOffset << 1);
-        for (let i = 0; i < count; ++i) {
-          store<u32>(out + (i << 2), cc1, HEADER_SIZE + 0);
-          store<u16>(out + (i << 1), cc2, HEADER_SIZE + 4);
-        }
-        break;
-      }
-      case 4: {
-        let cc = load<u64>(changetype<usize>(src), HEADER_SIZE);
-        let out = changetype<usize>(dest) + (destOffset << 1);
-        for (let i = 0; i < count; ++i) {
-          store<u64>(out + (i << 3), cc, HEADER_SIZE);
-        }
-        break;
-      }
-      default: {
-        let strLen = length << 1;
-        let to   = changetype<usize>(dest) + HEADER_SIZE + (destOffset << 1);
-        let from = changetype<usize>(src)  + HEADER_SIZE;
-        for (let i = 0, len = strLen * count; i < len; i += strLen) {
-          memory.copy(to + i, from, strLen);
-        }
-        break;
-      }
-    }
-  }
-}
-
-// Helpers
-
-@inline export const enum CharCode {
-  PLUS = 0x2B,
-  MINUS = 0x2D,
-  DOT = 0x2E,
-  _0 = 0x30,
-  _1 = 0x31,
-  _2 = 0x32,
-  _3 = 0x33,
-  _4 = 0x34,
-  _5 = 0x35,
-  _6 = 0x36,
-  _7 = 0x37,
-  _8 = 0x38,
-  _9 = 0x39,
-  A = 0x41,
-  B = 0x42,
-  E = 0x45,
-  N = 0x4E,
-  O = 0x4F,
-  X = 0x58,
-  Z = 0x5a,
-  a = 0x61,
-  b = 0x62,
-  e = 0x65,
-  n = 0x6E,
-  o = 0x6F,
-  x = 0x78,
-  z = 0x7A
-}
-
-export function isWhiteSpaceOrLineTerminator(c: u16): bool {
-  switch (c) {
-    case 9:    // <TAB>
-    case 10:   // <LF>
-    case 13:   // <CR>
-    case 11:   // <VT>
-    case 12:   // <FF>
-    case 32:   // <SP>
-    case 160:  // <NBSP>
-    case 8232: // <LS>
-    case 8233: // <PS>
-    case 65279: return true; // <ZWNBSP>
-    default: return false;
-  }
-}
-
-/** Parses a string to an integer (usually), using the specified radix. */
-export function parse<T>(str: String, radix: i32 = 0): T {
-  var len: i32 = str.length;
-  if (!len) return <T>NaN;
-
-  var ptr = changetype<usize>(str) /* + HEAD -> offset */;
-  var code = <i32>load<u16>(ptr, HEADER_SIZE);
-
-  // determine sign
-  var sign: T;
-  if (code == CharCode.MINUS) {
-    if (!--len) return <T>NaN;
-    code = <i32>load<u16>(ptr += 2, HEADER_SIZE);
-    sign = -1;
-  } else if (code == CharCode.PLUS) {
-    if (!--len) return <T>NaN;
-    code = <i32>load<u16>(ptr += 2, HEADER_SIZE);
-    sign = 1;
-  } else {
-    sign = 1;
-  }
-
-  // determine radix
-  if (!radix) {
-    if (code == CharCode._0 && len > 2) {
-      switch (<i32>load<u16>(ptr + 2, HEADER_SIZE)) {
-        case CharCode.B:
-        case CharCode.b: {
-          ptr += 4; len -= 2;
-          radix = 2;
-          break;
-        }
-        case CharCode.O:
-        case CharCode.o: {
-          ptr += 4; len -= 2;
-          radix = 8;
-          break;
-        }
-        case CharCode.X:
-        case CharCode.x: {
-          ptr += 4; len -= 2;
-          radix = 16;
-          break;
-        }
-        default: radix = 10;
-      }
-    } else radix = 10;
-  } else if (radix < 2 || radix > 36) {
-    return <T>NaN;
-  }
-
-  // calculate value
-  var num: T = 0;
-  while (len--) {
-    code = <i32>load<u16>(ptr, HEADER_SIZE);
-    if (code >= CharCode._0 && code <= CharCode._9) {
-      code -= CharCode._0;
-    } else if (code >= CharCode.A && code <= CharCode.Z) {
-      code -= CharCode.A - 10;
-    } else if (code >= CharCode.a && code <= CharCode.z) {
-      code -= CharCode.a - 10;
-    } else break;
-    if (code >= radix) break;
-    num = (num * radix) + code;
-    ptr += 2;
-  }
-  return sign * num;
-}
diff --git a/std/assembly/internal/typedarray.ts b/std/assembly/internal/typedarray.ts
deleted file mode 100644
index 2364f325..00000000
--- a/std/assembly/internal/typedarray.ts
+++ /dev/null
@@ -1,206 +0,0 @@
-import {
-  ALLOC_RAW,
-  REGISTER,
-  LINK
-} from "../runtime";
-
-import {
-  ArrayBuffer
-} from "../arraybuffer";
-
-import {
-  SORT as SORT_IMPL
-} from "./sort";
-
-/** Typed array base class. Not a global object. */
-export abstract class TypedArray<T> {
-  [key: number]: T; // compatibility only
-
-  readonly buffer: ArrayBuffer;
-  readonly byteOffset: i32;
-  readonly byteLength: i32;
-
-  constructor(length: i32) {
-    const MAX_LENGTH = <u32>ArrayBuffer.MAX_BYTELENGTH / sizeof<T>();
-    if (<u32>length > MAX_LENGTH) throw new RangeError("Invalid typed array length");
-    var byteLength = length << alignof<T>();
-    this.buffer = new ArrayBuffer(byteLength);
-    this.byteOffset = 0;
-    this.byteLength = byteLength;
-  }
-
-  get length(): i32 {
-    return this.byteLength >>> alignof<T>();
-  }
-
-  // TODO: could compute load/store offset from index and emit an immediate -> make this a builtin?
-
-  @operator("[]") protected __get(index: i32): T {
-    if (<u32>index >= <u32>(this.byteLength >>> alignof<T>())) throw new Error("Index out of bounds");
-    return load<T>(changetype<usize>(this.buffer) + <usize>this.byteOffset, <usize>index << alignof<T>());
-  }
-
-  @inline @operator("{}") protected __unchecked_get(index: i32): T {
-    return load<T>(changetype<usize>(this.buffer) + <usize>this.byteOffset + (<usize>index << alignof<T>()));
-  }
-
-  @operator("[]=") protected __set(index: i32, value: native<T>): void {
-    if (<u32>index >= <u32>(this.byteLength >>> alignof<T>())) throw new Error("Index out of bounds");
-    store<T>(changetype<usize>(this.buffer) + <usize>this.byteOffset + (<usize>index << alignof<T>()), value);
-  }
-
-  @inline @operator("{}=") protected __unchecked_set(index: i32, value: native<T>): void {
-    store<T>(changetype<usize>(this.buffer) + <usize>this.byteOffset + (<usize>index << alignof<T>()), value);
-  }
-
-  // copyWithin(target: i32, start: i32, end: i32 = this.length): this
-}
-
-@inline export function FILL<TArray extends TypedArray<T>, T extends number>(
-  array: TArray,
-  value: native<T>,
-  start: i32,
-  end: i32
-): TArray {
-  var base = changetype<usize>(array.buffer) + <usize>array.byteOffset;
-  var len = array.length;
-  start = start < 0 ? max(len + start, 0) : min(start, len);
-  end   = end   < 0 ? max(len + end,   0) : min(end,   len);
-  if (sizeof<T>() == 1) {
-    if (start < end) memory.fill(base + <usize>start, <u8>value, <usize>(end - start));
-  } else {
-    for (; start < end; ++start) {
-      store<T>(base + (<usize>start << alignof()), value);
-    }
-  }
-  return array;
-}
-
-@inline export function SORT<TArray extends TypedArray<T>, T>(
-  array: TArray,
-  comparator: (a: T, b: T) => i32
-): TArray {
-  var length = array.length;
-  var offset = <usize>array.byteOffset;
-  if (length <= 1) return array;
-  var buffer = changetype<usize>(array.buffer);
-  if (length == 2) {
-    let a = load<T>(buffer + offset, sizeof<T>());
-    let b = load<T>(buffer + offset);
-    if (comparator(a, b) < 0) {
-      store<T>(buffer + offset, b, sizeof<T>());
-      store<T>(buffer + offset, a);
-    }
-    return array;
-  }
-  // TODO
-  // SORT_IMPL<T>(buffer, byteOffset, length, comparator);
-  return array;
-}
-
-@inline export function SUBARRAY<TArray extends TypedArray<T>, T>(
-  array: TArray,
-  begin: i32,
-  end: i32
-): TArray {
-  var buffer = changetype<usize>(array.buffer);
-  var length = <i32>array.length;
-  if (begin < 0) begin = max(length + begin, 0);
-  else begin = min(begin, length);
-  if (end < 0) end = max(length + end, begin);
-  else end = max(min(end, length), begin);
-  var out = ALLOC_RAW(offsetof<TArray>());
-  store<usize>(out, buffer, offsetof<TArray>("buffer"));
-  store<usize>(out, buffer + (<usize>begin << alignof<T>()), offsetof<TArray>("dataStart"));
-  store<usize>(out, buffer + (<usize>end   << alignof<T>()), offsetof<TArray>("dataStart"));
-  store<i32>(out, <i32>array.byteOffset + (begin << alignof<T>()), offsetof<TArray>("byteOffset"));
-  store<i32>(out, (end - begin) << alignof<T>(), offsetof<TArray>("byteLength"));
-  LINK(buffer, REGISTER<TArray>(out)); // register first, then link
-  return changetype<TArray>(out);
-}
-
-@inline export function REDUCE<TArray extends TypedArray<T>, T, TRet>(
-  array: TArray,
-  callbackfn: (accumulator: TRet, value: T, index: i32, array: TArray) => TRet,
-  initialValue: TRet
-): TRet {
-  var base = changetype<usize>(array.buffer) + <usize>array.byteOffset;
-  for (let i = 0, k = array.length; i < k; i++) {
-    initialValue = callbackfn(initialValue, load<T>(base + (<usize>i << alignof<T>())), i, array);
-  }
-  return initialValue;
-}
-
-@inline export function REDUCE_RIGHT<TArray extends TypedArray<T>, T, TRet>(
-  array: TArray,
-  callbackfn: (accumulator: TRet, value: T, index: i32, array: TArray) => TRet,
-  initialValue: TRet
-): TRet {
-  var base = changetype<usize>(array.buffer) + <usize>array.byteOffset;
-  for (let i = array.length - 1; i >= 0; i--) {
-    initialValue = callbackfn(initialValue, load<T>(base + (<usize>i << alignof<T>())), i, array);
-  }
-  return initialValue;
-}
-
-@inline export function MAP<TArray extends TypedArray<T>, T>(
-  array: TArray,
-  callbackfn: (value: T, index: i32, self: TArray) => T,
-): TArray {
-  var length = array.length;
-  var base = changetype<usize>(array.buffer) + <usize>array.byteOffset;
-  var result = instantiate<TArray>(length);
-  var resultBase = changetype<usize>(result.buffer); // assumes byteOffset = 0
-  for (let i = 0; i < length; i++) {
-    store<T>(
-      resultBase + (<usize>i << alignof<T>()),
-      callbackfn(load<T>(base + (<usize>i << alignof<T>())), i, array)
-    );
-  }
-
-  return result;
-}
-
-@inline export function FIND_INDEX<TArray extends TypedArray<T>, T>(
-  array: TArray,
-  callbackfn: (value: T, index: i32, array: TArray) => bool,
-): i32 {
-  var base = changetype<usize>(array.buffer) + <usize>array.byteOffset;
-  for (let i = 0, k = array.length; i < k; i++) {
-    if (callbackfn(load<T>(base + (<usize>i << alignof<T>())), i, array)) return i;
-  }
-  return -1;
-}
-
-@inline export function SOME<TArray extends TypedArray<T>, T>(
-  array: TArray,
-  callbackfn: (value: T, index: i32, array: TArray) => bool,
-): bool {
-  var base = changetype<usize>(array.buffer) + <usize>array.byteOffset;
-  for (let i = 0, k = array.length; i < k; i++) {
-    if (callbackfn(load<T>(base + (<usize>i << alignof<T>())), i, array)) return true;
-  }
-  return false;
-}
-
-@inline export function EVERY<TArray extends TypedArray<T>, T>(
-  array: TArray,
-  callbackfn: (value: T, index: i32, array: TArray) => bool,
-): bool {
-  var base = changetype<usize>(array.buffer) + <usize>array.byteOffset;
-  for (let i = 0, k = array.length; i < k; i++) {
-    if (callbackfn(load<T>(base + (<usize>i << alignof<T>())), i, array)) continue;
-    return false;
-  }
-  return true;
-}
-
-@inline export function FOREACH<TArray extends TypedArray<T>, T>(
-  array: TArray,
-  callbackfn: (value: T, index: i32, array: TArray) => void,
-): void {
-  var base = changetype<usize>(array.buffer) + <usize>array.byteOffset;
-  for (let i = 0, k = array.length; i < k; i++) {
-    callbackfn(load<T>(base + (<usize>i << alignof<T>())), i, array);
-  }
-}
diff --git a/std/assembly/string.ts b/std/assembly/string.ts
index 26c8368e..bc4a2e28 100644
--- a/std/assembly/string.ts
+++ b/std/assembly/string.ts
@@ -355,7 +355,8 @@ import {
       // split by chars
       length = min<isize>(length, <isize>limit);
       let result = new Array<String>(length);
-      let buffer = <ArrayBuffer>result.buffer_;
+      let buffer = unreachable(); // TODO
+      // let buffer = <ArrayBuffer>result.buffer_;
       for (let i: isize = 0; i < length; ++i) {
         let char = ALLOC(2);
         store<u16>(
diff --git a/std/assembly/internal/allocator.ts b/std/assembly/util/allocator.ts
similarity index 100%
rename from std/assembly/internal/allocator.ts
rename to std/assembly/util/allocator.ts
diff --git a/std/assembly/internal/hash.ts b/std/assembly/util/hash.ts
similarity index 86%
rename from std/assembly/internal/hash.ts
rename to std/assembly/util/hash.ts
index dd7c7403..f4dcea3e 100644
--- a/std/assembly/internal/hash.ts
+++ b/std/assembly/util/hash.ts
@@ -1,11 +1,4 @@
-import {
-  HEADER_SIZE
-} from "./string";
-
-/** Computes the 32-bit hash of a value of any type. */
-@inline
-export function HASH<T>(key: T): u32 {
-  // branch-level tree-shaking makes this a `(return (call ...))`
+@inline export function HASH<T>(key: T): u32 {
   if (isString(key)) {
     return hashStr(key);
   } else if (isReference<T>()) {
@@ -66,7 +59,7 @@ function hash64(key: u64): u32 {
 function hashStr(key: string): u32 {
   var v = FNV_OFFSET;
   for (let i: usize = 0, k: usize = key.length << 1; i < k; ++i) {
-    v = (v ^ <u32>load<u8>(changetype<usize>(key) + i, HEADER_SIZE)) * FNV_PRIME;
+    v = (v ^ <u32>load<u8>(changetype<usize>(key) + i)) * FNV_PRIME;
   }
   return v;
 }
diff --git a/std/assembly/internal/memory.ts b/std/assembly/util/memory.ts
similarity index 100%
rename from std/assembly/internal/memory.ts
rename to std/assembly/util/memory.ts
diff --git a/std/assembly/internal/number.ts b/std/assembly/util/number.ts
similarity index 99%
rename from std/assembly/internal/number.ts
rename to std/assembly/util/number.ts
index 0cc7ac21..01dff985 100644
--- a/std/assembly/internal/number.ts
+++ b/std/assembly/util/number.ts
@@ -8,10 +8,6 @@ import {
   FREE
 } from "../runtime";
 
-import {
-  LOAD
-} from "./arraybuffer";
-
 @inline export const MAX_DOUBLE_LENGTH = 28;
 
 @lazy @inline const POWERS10: u32[] = [
diff --git a/std/assembly/internal/sort.ts b/std/assembly/util/sort.ts
similarity index 93%
rename from std/assembly/internal/sort.ts
rename to std/assembly/util/sort.ts
index 587de405..db9e3048 100644
--- a/std/assembly/internal/sort.ts
+++ b/std/assembly/util/sort.ts
@@ -1,6 +1,4 @@
-/** Obtains the default comparator for the specified value type. */
-@inline
-export function COMPARATOR<T>(): (a: T, b: T) => i32 {
+@inline export function COMPARATOR<T>(): (a: T, b: T) => i32 {
   if (isInteger<T>()) {
     if (isSigned<T>() && sizeof<T>() <= 4) {
       return (a: T, b: T): i32 => (<i32>(a - b));
@@ -40,8 +38,7 @@ export function COMPARATOR<T>(): (a: T, b: T) => i32 {
   }
 }
 
-@inline
-export function SORT<T>(
+@inline export function SORT<T>(
   dataStart: usize,
   length: i32,
   comparator: (a: T, b: T) => i32
@@ -58,7 +55,6 @@ export function SORT<T>(
   }
 }
 
-/** Sorts an Array with the 'Insertion Sort' algorithm. */
 function insertionSort<T>(
   dataStart: usize,
   length: i32,
@@ -77,7 +73,6 @@ function insertionSort<T>(
   }
 }
 
-/** Sorts an Array with the 'Weak Heap Sort' algorithm. */
 function weakHeapSort<T>(
   dataStart: usize,
   length: i32,
diff --git a/std/assembly/util/string.ts b/std/assembly/util/string.ts
new file mode 100644
index 00000000..a3374134
--- /dev/null
+++ b/std/assembly/util/string.ts
@@ -0,0 +1,124 @@
+export function compareImpl(str1: string, index1: usize, str2: string, index2: usize, len: usize): i32 {
+  var result: i32 = 0;
+  var ptr1 = changetype<usize>(str1) + (index1 << 1);
+  var ptr2 = changetype<usize>(str2) + (index2 << 1);
+  while (len && !(result = <i32>load<u16>(ptr1) - <i32>load<u16>(ptr2))) {
+    --len, ptr1 += 2, ptr2 += 2;
+  }
+  return result;
+}
+
+@inline export const enum CharCode {
+  PLUS = 0x2B,
+  MINUS = 0x2D,
+  DOT = 0x2E,
+  _0 = 0x30,
+  _1 = 0x31,
+  _2 = 0x32,
+  _3 = 0x33,
+  _4 = 0x34,
+  _5 = 0x35,
+  _6 = 0x36,
+  _7 = 0x37,
+  _8 = 0x38,
+  _9 = 0x39,
+  A = 0x41,
+  B = 0x42,
+  E = 0x45,
+  N = 0x4E,
+  O = 0x4F,
+  X = 0x58,
+  Z = 0x5a,
+  a = 0x61,
+  b = 0x62,
+  e = 0x65,
+  n = 0x6E,
+  o = 0x6F,
+  x = 0x78,
+  z = 0x7A
+}
+
+export function isWhiteSpaceOrLineTerminator(c: u16): bool {
+  switch (c) {
+    case 9:    // <TAB>
+    case 10:   // <LF>
+    case 13:   // <CR>
+    case 11:   // <VT>
+    case 12:   // <FF>
+    case 32:   // <SP>
+    case 160:  // <NBSP>
+    case 8232: // <LS>
+    case 8233: // <PS>
+    case 65279: return true; // <ZWNBSP>
+    default: return false;
+  }
+}
+
+/** Parses a string to an integer (usually), using the specified radix. */
+export function parse<T>(str: string, radix: i32 = 0): T {
+  var len: i32 = str.length;
+  if (!len) return <T>NaN;
+
+  var ptr = changetype<usize>(str) /* + HEAD -> offset */;
+  var code = <i32>load<u16>(ptr, HEADER_SIZE);
+
+  // determine sign
+  var sign: T;
+  if (code == CharCode.MINUS) {
+    if (!--len) return <T>NaN;
+    code = <i32>load<u16>(ptr += 2, HEADER_SIZE);
+    sign = -1;
+  } else if (code == CharCode.PLUS) {
+    if (!--len) return <T>NaN;
+    code = <i32>load<u16>(ptr += 2, HEADER_SIZE);
+    sign = 1;
+  } else {
+    sign = 1;
+  }
+
+  // determine radix
+  if (!radix) {
+    if (code == CharCode._0 && len > 2) {
+      switch (<i32>load<u16>(ptr + 2, HEADER_SIZE)) {
+        case CharCode.B:
+        case CharCode.b: {
+          ptr += 4; len -= 2;
+          radix = 2;
+          break;
+        }
+        case CharCode.O:
+        case CharCode.o: {
+          ptr += 4; len -= 2;
+          radix = 8;
+          break;
+        }
+        case CharCode.X:
+        case CharCode.x: {
+          ptr += 4; len -= 2;
+          radix = 16;
+          break;
+        }
+        default: radix = 10;
+      }
+    } else radix = 10;
+  } else if (radix < 2 || radix > 36) {
+    return <T>NaN;
+  }
+
+  // calculate value
+  var num: T = 0;
+  while (len--) {
+    code = <i32>load<u16>(ptr, HEADER_SIZE);
+    if (code >= CharCode._0 && code <= CharCode._9) {
+      code -= CharCode._0;
+    } else if (code >= CharCode.A && code <= CharCode.Z) {
+      code -= CharCode.A - 10;
+    } else if (code >= CharCode.a && code <= CharCode.z) {
+      code -= CharCode.a - 10;
+    } else break;
+    if (code >= radix) break;
+    num = (num * radix) + code;
+    ptr += 2;
+  }
+  return sign * num;
+}