diff --git a/osh/split.py b/osh/split.py
index 9296c820cf..b4ecb16e16 100644
--- a/osh/split.py
+++ b/osh/split.py
@@ -182,10 +182,10 @@ def SplitForWordEval(self, s, ifs=None):
         Also used by the explicit shSplit() function.
         """
         sp = self._GetSplitter(ifs=ifs)
-        spans = sp.Split(s, True)
+        #spans = sp.Split(s, True)
 
         # Note: pass allow_escape=False so \ isn't special
-        #spans = sp.Split(s, False)
+        spans = sp.Split(s, False)
 
         if 0:
             for span in spans:
diff --git a/osh/word_eval.py b/osh/word_eval.py
index ec659c962e..d5aeea6e29 100644
--- a/osh/word_eval.py
+++ b/osh/word_eval.py
@@ -48,7 +48,6 @@
     sh_lvalue,
     sh_lvalue_t,
 )
-from core import bash_impl
 from core import error
 from core import pyos
 from core import pyutil
@@ -504,6 +503,64 @@ def Eval(self, part):
         return result
 
 
+
+class FrameEvaluator(object):
+
+    def __init__(self, splitter):
+        # type: (SplitContext) -> None
+        self.frags = []  # type: List[str]
+        self.frag = []  # type: List[str]
+        self.splitter = splitter
+
+    def _Append(self, text):
+        # type: (str) -> None
+        self.frag.append(text)
+
+    def _NextFrag(self):
+        # type: () -> None
+        if len(self.frag):
+            self.frags.append("".join(self.frag))
+            self.frag = []
+
+    def _Split(self, piece, quoted, do_split, will_glob):
+        # type: (str, bool, bool, bool) -> None
+
+        # A word like /$exists/$notexists/ should eval to ['/someval//'] not ['/someval/', '/']
+        if len(piece) == 0 and do_split:
+            return
+
+        if do_split:
+            splits = self.splitter.SplitForWordEval(piece)
+            if len(splits) == 0:
+                self._NextFrag()
+                return
+
+            last = len(splits) - 1
+            for i, tosplit in enumerate(splits):
+                self._Split(tosplit, quoted, False, will_glob)
+
+                if i != last:
+                    self._NextFrag()
+
+        elif quoted and will_glob:
+            self._Append(glob_.GlobEscape(piece))
+
+        else:
+            self._Append(piece)
+
+    def Eval(self, frame, will_glob):
+        # type: (List[Piece], bool) -> List[str]
+        del self.frags[:]
+        del self.frag[:]
+
+        for piece in frame:
+            self._Split(piece.s, piece.quoted, piece.do_split, will_glob)
+
+        self._NextFrag()
+
+        return self.frags
+
+
 class AbstractWordEvaluator(StringWordEvaluator):
     """Abstract base class for word evaluators.
 
@@ -538,6 +595,8 @@ def __init__(
 
         self.globber = glob_.Globber(exec_opts)
 
+        self.frame_ev = FrameEvaluator(self.splitter)
+
     def CheckCircularDeps(self):
         # type: () -> None
         raise NotImplementedError()
@@ -790,15 +849,19 @@ def _Length(self, val, token):
 
             elif case(value_e.BashArray):
                 val = cast(value.BashArray, UP_val)
-                length = bash_impl.BashArray_Length(val)
-
-            elif case(value_e.BashAssoc):
-                val = cast(value.BashAssoc, UP_val)
-                length = bash_impl.BashAssoc_Length(val)
+                # There can be empty placeholder values in the array.
+                length = 0
+                for s in val.strs:
+                    if s is not None:
+                        length += 1
 
             elif case(value_e.SparseArray):
                 val = cast(value.SparseArray, UP_val)
-                length = bash_impl.SparseArray_Length(val)
+                length = len(val.d)
+
+            elif case(value_e.BashAssoc):
+                val = cast(value.BashAssoc, UP_val)
+                length = len(val.d)
 
             else:
                 raise error.TypeErr(
@@ -1974,7 +2037,12 @@ def _EvalWordFrame(self, frame, argv):
         all_quoted = True
         any_quoted = False
 
-        #log('--- frame %s', frame)
+        if 0:
+            log('---')
+            log('FRAME')
+            for i, piece in enumerate(frame):
+                log('(%d) %s', i, piece)
+            log('')
 
         for piece in frame:
             if len(piece.s):
@@ -1998,31 +2066,7 @@ def _EvalWordFrame(self, frame, argv):
             return
 
         will_glob = not self.exec_opts.noglob()
-
-        if 0:
-            log('---')
-            log('FRAME')
-            for i, piece in enumerate(frame):
-                log('(%d) %s', i, piece)
-            log('')
-
-        # Array of strings, some of which are BOTH IFS-escaped and GLOB escaped!
-        frags = []  # type: List[str]
-        for piece in frame:
-            if will_glob and piece.quoted:
-                frag = glob_.GlobEscape(piece.s)
-            else:
-                # If we have a literal \, then we turn it into \\\\.
-                # Splitting takes \\\\ -> \\
-                # Globbing takes \\ to \ if it doesn't match
-                frag = _BackslashEscape(piece.s)
-
-            if piece.do_split:
-                frag = _BackslashEscape(frag)
-            else:
-                frag = self.splitter.Escape(frag)
-
-            frags.append(frag)
+        frags = self.frame_ev.Eval(frame, will_glob)
 
         if 0:
             log('---')
@@ -2031,27 +2075,27 @@ def _EvalWordFrame(self, frame, argv):
                 log('(%d) %s', i, frag)
             log('')
 
-        flat = ''.join(frags)
-        #log('flat: %r', flat)
-
-        args = self.splitter.SplitForWordEval(flat)
-
-        # space=' '; argv $space"".  We have a quoted part, but we CANNOT elide.
-        # Add it back and don't bother globbing.
-        if len(args) == 0 and any_quoted:
-            argv.append('')
+        if not len(frags):
             return
 
-        #log('split args: %r', args)
-        for a in args:
-            if glob_.LooksLikeGlob(a):
-                n = self.globber.Expand(a, argv)
+        for frag in frags:
+            if glob_.LooksLikeGlob(frag):
+                n = self.globber.Expand(frag, argv)
                 if n < 0:
                     # TODO: location info, with span IDs carried through the frame
                     raise error.FailGlob('Pattern %r matched no files' % a,
                                          loc.Missing)
+            elif will_glob:
+                argv.append(glob_.GlobUnescape(frag))
             else:
-                argv.append(glob_.GlobUnescape(a))
+                argv.append(frag)
+
+        if 0:
+            log('---')
+            log('ARGV')
+            for i, arg in enumerate(argv):
+                log('(%d) %s', i, arg)
+            log('')
 
     def _EvalWordToArgv(self, w):
         # type: (CompoundWord) -> List[str]