The SymPy project (https://github.com/sympy/sympy) makes heavy use of caching to speed up the creation of symbols and expressions.  If available, we make use of the fastcache library (https://github.com/pbrady/fastcache) - an lru_cache written in C.  Otherwise we use lru_cache provided by functools.  When testing with 3.5, we started to observe `KeyError`s coming from the new lru_cache implementation in _functoolsmodule.c.  There was some discussion on this on the SymPy mailing list here https://groups.google.com/forum/#!topic/sympy/AnwYTJGRBB4.

Here's an example of the failure in the SymPy test suite (Note that this failure does not occur if we use the lru_cache in 3.4 or fastcache or if we use the lru_cache in 3.5 with no size limit)

$ /usr/local/opt/python-3.5/bin/python3
Python 3.5.0 (default, Oct  1 2015, 21:51:43) 
[GCC 5.1.1 20150618 (Red Hat 5.1.1-4)] on linux
Type "help", "copyright", "credits" or "license" for more information.
>>> from sympy import test
>>> test("test_frame.py")
============================= test process starts ==============================
executable:         /usr/local/opt/python-3.5/bin/python3  (3.5.0-final-0) [CPython]
architecture:       64-bit
cache:              yes
ground types:       python 
random seed:        7625225
hash randomization: on (PYTHONHASHSEED=2787593824)

sympy/physics/vector/tests/test_frame.py[4] .EE.                          [FAIL]

________________________________________________________________________________
____________ sympy/physics/vector/tests/test_frame.py:test_ang_vel _____________
  File "/home/peter/.local/lib/python3.5/site-packages/sympy/physics/vector/tests/test_frame.py", line 66, in test_ang_vel
    B = A.orientnew('B', 'Axis', [q2, A.x])
  File "/home/peter/.local/lib/python3.5/site-packages/sympy/physics/vector/frame.py", line 640, in orientnew
    newframe.orient(self, rot_type, amounts, rot_order)
  File "/home/peter/.local/lib/python3.5/site-packages/sympy/physics/vector/frame.py", line 523, in orient
    [-axis[1], axis[0], 0]]) * sin(theta) + axis * axis.T)
  File "/home/peter/.local/lib/python3.5/site-packages/sympy/core/cache.py", line 93, in wrapper
    retval = cfunc(*args, **kwargs)
  File "/home/peter/.local/lib/python3.5/site-packages/sympy/core/function.py", line 375, in __new__
    result = super(Function, cls).__new__(cls, *args, **options)
  File "/home/peter/.local/lib/python3.5/site-packages/sympy/core/cache.py", line 93, in wrapper
    retval = cfunc(*args, **kwargs)
  File "/home/peter/.local/lib/python3.5/site-packages/sympy/core/function.py", line 199, in __new__
    evaluated = cls.eval(*args)
  File "/home/peter/.local/lib/python3.5/site-packages/sympy/functions/elementary/trigonometric.py", line 459, in eval
    if arg.could_extract_minus_sign():
  File "/home/peter/.local/lib/python3.5/site-packages/sympy/core/expr.py", line 2047, in could_extract_minus_sign
    negative_self = -self
  File "/home/peter/.local/lib/python3.5/site-packages/sympy/core/expr.py", line 113, in __neg__
    return Mul(S.NegativeOne, self)
  File "/home/peter/.local/lib/python3.5/site-packages/sympy/core/cache.py", line 93, in wrapper
    retval = cfunc(*args, **kwargs)
KeyError: (<class 'sympy.core.mul.Mul'>, 1, B_y, sin(2*q(t)), <class 'sympy.core.assumptions.ManagedProperties'>, <class 'sympy.core.numbers.One'>, <class 'sympy.physics.vector.frame.CoordinateSym'>, sin)
________________________________________________________________________________
______________ sympy/physics/vector/tests/test_frame.py:test_dcm _______________
  File "/home/peter/.local/lib/python3.5/site-packages/sympy/physics/vector/tests/test_frame.py", line 131, in test_dcm
    B = A.orientnew('B', 'Axis', [q2, A.x])
  File "/home/peter/.local/lib/python3.5/site-packages/sympy/physics/vector/frame.py", line 640, in orientnew
    newframe.orient(self, rot_type, amounts, rot_order)
  File "/home/peter/.local/lib/python3.5/site-packages/sympy/physics/vector/frame.py", line 523, in orient
    [-axis[1], axis[0], 0]]) * sin(theta) + axis * axis.T)
  File "/home/peter/.local/lib/python3.5/site-packages/sympy/core/cache.py", line 93, in wrapper
    retval = cfunc(*args, **kwargs)
  File "/home/peter/.local/lib/python3.5/site-packages/sympy/core/function.py", line 375, in __new__
    result = super(Function, cls).__new__(cls, *args, **options)
  File "/home/peter/.local/lib/python3.5/site-packages/sympy/core/cache.py", line 93, in wrapper
    retval = cfunc(*args, **kwargs)
  File "/home/peter/.local/lib/python3.5/site-packages/sympy/core/function.py", line 199, in __new__
    evaluated = cls.eval(*args)
  File "/home/peter/.local/lib/python3.5/site-packages/sympy/functions/elementary/trigonometric.py", line 459, in eval
    if arg.could_extract_minus_sign():
  File "/home/peter/.local/lib/python3.5/site-packages/sympy/core/expr.py", line 2050, in could_extract_minus_sign
    (negative_self).extract_multiplicatively(-1) is not None)
  File "/home/peter/.local/lib/python3.5/site-packages/sympy/core/expr.py", line 1859, in extract_multiplicatively
    quotient = self / c
  File "/home/peter/.local/lib/python3.5/site-packages/sympy/core/decorators.py", line 77, in __sympifyit_wrapper
    return func(a, b)
  File "/home/peter/.local/lib/python3.5/site-packages/sympy/core/decorators.py", line 118, in binary_op_wrapper
    return func(self, other)
  File "/home/peter/.local/lib/python3.5/site-packages/sympy/core/expr.py", line 161, in __div__
    return Mul(self, Pow(other, S.NegativeOne))
  File "/home/peter/.local/lib/python3.5/site-packages/sympy/core/cache.py", line 93, in wrapper
    retval = cfunc(*args, **kwargs)
KeyError: (<class 'sympy.core.mul.Mul'>, B_x, -cos(2*q(t)) + 1, <class 'sympy.core.assumptions.ManagedProperties'>, <class 'sympy.physics.vector.frame.CoordinateSym'>, <class 'sympy.core.add.Add'>)

=========== tests finished: 2 passed, 2 exceptions, in 1.57 seconds ============
DO *NOT* COMMIT!
False
>>> 



Sadly, I don't have a simplified test case to reproduce the error.  I would be happy to try some debugging suggestions.

Thanks,
Peter.

Issue24483 fixes this issue.

I don't know what was the actual cause of the error and if issue24483 fixed it or mask.

As a sanity check I removed the stored hashvalue in Raymond Hettinger's backported lru_cache (which we use to support 2.6-3.2) and end up with errors as well.  So it seems like 24483 is the appropriate fix to restore the old behavior.  Thanks for looking into this.

Does this mean that some SymPy object is giving different hash values on successive calls to hash()? We definitely need to look into this on the SymPy side.

Seems this issue can be closed now.

Yes. The patch you suggested restores the old behavior of the lru_cache.