Move comment to where it belongs.

core/expr_eval.py

@@ -102,23 +102,7 @@ def _StringToInteger(s, word=None):


def _ValToArith(val, word=None):
  """Convert runtime.value to Python int or list.

  NOTE: We could have a runtime.arith_value and get rid of the isinstance()
  check.  But this means our array indexing is lazy, which I think is fine.

  PROBLEM: array[1000000]=1 could use up a lot of memory.

  But I think that is OK, at least until there's evidence that people want
  to use arrays that way.

  But what about hash tables?  That should be a separate type.

  Representation could be:

  ['1', 2, 3, None, None, '4', None]
  Then length counts the entries that are not None.
  """
  """Convert runtime.value to a Python int or list of strings."""
  assert isinstance(val, runtime.value), '%r %r' % (val, type(val))
  if val.tag == value_e.Undef:
    return 0

core/state.py

@@ -375,9 +375,13 @@ def SetVar(self, lval, value, new_flags, lookup_mode):
        try:
          strs[lval.index] = value.s
        except IndexError:
          # TODO: strict-array won't support this.  For Oil arrays.
          # Fill it in with None.  It could look like this:
          # ['1', 2, 3, None, None, '4', None]
          # Then ${#a[@]} counts the entries that are not None.
          #
          # TODO: strict-array for Oil arrays won't auto-fill.
          n = len(strs) - lval.index + 1
          strs.extend([None] * n)  # Fill it in with None
          strs.extend([None] * n)
          strs[lval.index] = value.s
      else:
        # TODO: