jax-ml/jax · error · TypeError

len() of unsized object

Error message

len() of unsized object

What it means

len() of an EArray with ndim==0 has no first-axis length, matching NumPy's 'len() of unsized object' TypeError for 0-d arrays. EArray.__len__ returns shape[0], which does not exist for scalars.

Source

Thrown at jax/_src/earray.py:70

    return EArray(self.aval, self._data.copy())

  def __repr__(self):
    return 'E' + repr(self._data)

  def __iter__(self):
    if self.ndim == 0: raise TypeError('iteration over a 0-d array')
    raise NotImplementedError

  # forward to aval
  shape = property(lambda self: self.aval.shape)
  dtype = property(lambda self: self.aval.dtype)

  # computed from shape and dtype
  ndim = property(lambda self: len(self.aval.shape))
  size = property(lambda self: math.prod(self.aval.shape))
  itemsize = property(lambda self: self.aval.dtype.itemsize)
  def __len__(self):
    if self.ndim == 0: raise TypeError('len() of unsized object')
    return self.shape[0]

  # forward to self._data
  devices = property(lambda self: self._data.devices)  # pyrefly: ignore[bad-override]
  _committed = property(lambda self: self._data._committed)
  is_fully_addressable = property(lambda self: self._data.is_fully_addressable)
  is_fully_replicated = property(lambda self: self._data.is_fully_replicated)
  delete = property(lambda self: self._data.delete)  # pyrefly: ignore[bad-override]
  is_deleted = property(lambda self: self._data.is_deleted)  # pyrefly: ignore[bad-override]
  on_device_size_in_bytes = property(lambda self: self._data.on_device_size_in_bytes)  # pyrefly: ignore[bad-override]
  unsafe_buffer_pointer = property(lambda self: self._data.unsafe_buffer_pointer)  # pyrefly: ignore[bad-override]

  # defer to extended dtype rules
  @property
  def sharding(self):
    phys_sharding = self._data.sharding
    return sharding_impls.logical_sharding(self.shape, self.dtype, phys_sharding)

View on GitHub (pinned to 1e1c6a8fc0)

Solutions

1. Guard with ndim: n = earr.shape[0] if earr.ndim > 0 else 1
2. Use size for element count: earr.size works for 0-d (returns 1)
3. Use np.ndim(earr)==0 checks in generic helpers before calling len

Example fix

# before
n = len(earr)

# after
n = earr.shape[0] if earr.ndim > 0 else 1

n = earr.shape[0] if getattr(earr, 'ndim', 0) > 0 else 1

def has_len(e) -> bool:
    return getattr(e, 'ndim', None) not in (0, None)

When it happens

Trigger: len(earr) where the EArray's aval shape is (); often inside generic container code (dataclasses, serializers) that calls len() on every value.

Common situations: Serialization or logging utilities that call len(x) to size containers; mixed scalar/array pipelines where a value is sometimes 0-d; etuple-driven symbolic expressions.

Related errors

• len() of unsized object
• iteration over a 0-d array
• len() of unsized object
• numpy masked arrays are not supported as direct inputs to JA
• Python int {value} too large to convert to int64

AI-assisted analysis of jax-ml/jax@1e1c6a8fc0 (2026-08-27). Data as JSON: /api/errors/8cb9a8cab183a382. Report an issue: GitHub.