OverflowError/TypeError with evalf ceiling(log(1)/log(2))

[Mygod]

Python 3.7.3 (default, Apr  3 2019, 19:16:38) 
[GCC 8.0.1 20180414 (experimental) [trunk revision 259383]] on linux
Type "help", "copyright", "credits" or "license" for more information.
>>> import sympy
>>> test = sympy.symbols('test')
>>> print(sympy.ceiling(sympy.log(test) / sympy.log(2)).evalf(subs={test: 1}))
Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
  File "/path/to/lib/python3.7/site-packages/sympy/core/evalf.py", line 1442, in evalf
    result = evalf(self, prec + 4, options)
  File "/path/to/lib/python3.7/site-packages/sympy/core/evalf.py", line 1309, in evalf
    r = rf(x, prec, options)
  File "/path/to/lib/python3.7/site-packages/sympy/core/evalf.py", line 410, in evalf_ceiling
    return get_integer_part(expr.args[0], 1, options)
  File "/path/to/lib/python3.7/site-packages/sympy/core/evalf.py", line 400, in get_integer_part
    re_, re_acc = calc_part(re(expr, evaluate=False), ire)
  File "/path/to/lib/python3.7/site-packages/sympy/core/evalf.py", line 356, in calc_part
    re_im, size, options)
  File "/path/to/lib/python3.7/site-packages/sympy/core/evalf.py", line 1309, in evalf
    r = rf(x, prec, options)
  File "/path/to/lib/python3.7/site-packages/sympy/core/evalf.py", line 251, in evalf_re
    return get_complex_part(expr.args[0], 0, prec, options)
  File "/path/to/lib/python3.7/site-packages/sympy/core/evalf.py", line 237, in get_complex_part
    res = evalf(expr, workprec, options)
  File "/path/to/lib/python3.7/site-packages/sympy/core/evalf.py", line 1309, in evalf
    r = rf(x, prec, options)
  File "/path/to/lib/python3.7/site-packages/sympy/core/evalf.py", line 561, in evalf_mul
    arg = evalf(arg, prec, options)
  File "/path/to/lib/python3.7/site-packages/sympy/core/evalf.py", line 1309, in evalf
    r = rf(x, prec, options)
  File "/path/to/lib/python3.7/site-packages/sympy/core/evalf.py", line 678, in evalf_pow
    re, im, re_acc, im_acc = evalf(base, prec + 5, options)
  File "/path/to/lib/python3.7/site-packages/sympy/core/evalf.py", line 1309, in evalf
    r = rf(x, prec, options)
  File "/path/to/lib/python3.7/site-packages/sympy/core/evalf.py", line 847, in evalf_log
    re = mpf_log(mpf_abs(xre), prec, rnd)
  File "/path/to/lib/python3.7/site-packages/mpmath/libmp/libelefun.py", line 689, in mpf_log
    return from_man_exp(exp*ln2_fixed(wp), -wp, prec, rnd)
  File "/path/to/lib/python3.7/site-packages/mpmath/libmp/libelefun.py", line 98, in g
    newprec = int(prec*1.05+10)
OverflowError: cannot convert float infinity to integer
>>> print(sympy.ceiling(sympy.log(test) / sympy.log(3)).evalf(subs={test: 1}))
Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
  File "/path/to/lib/python3.7/site-packages/sympy/core/evalf.py", line 1442, in evalf
    result = evalf(self, prec + 4, options)
  File "/path/to/lib/python3.7/site-packages/sympy/core/evalf.py", line 1309, in evalf
    r = rf(x, prec, options)
  File "/path/to/lib/python3.7/site-packages/sympy/core/evalf.py", line 410, in evalf_ceiling
    return get_integer_part(expr.args[0], 1, options)
  File "/path/to/lib/python3.7/site-packages/sympy/core/evalf.py", line 400, in get_integer_part
    re_, re_acc = calc_part(re(expr, evaluate=False), ire)
  File "/path/to/lib/python3.7/site-packages/sympy/core/evalf.py", line 356, in calc_part
    re_im, size, options)
  File "/path/to/lib/python3.7/site-packages/sympy/core/evalf.py", line 1309, in evalf
    r = rf(x, prec, options)
  File "/path/to/lib/python3.7/site-packages/sympy/core/evalf.py", line 251, in evalf_re
    return get_complex_part(expr.args[0], 0, prec, options)
  File "/path/to/lib/python3.7/site-packages/sympy/core/evalf.py", line 237, in get_complex_part
    res = evalf(expr, workprec, options)
  File "/path/to/lib/python3.7/site-packages/sympy/core/evalf.py", line 1309, in evalf
    r = rf(x, prec, options)
  File "/path/to/lib/python3.7/site-packages/sympy/core/evalf.py", line 561, in evalf_mul
    arg = evalf(arg, prec, options)
  File "/path/to/lib/python3.7/site-packages/sympy/core/evalf.py", line 1309, in evalf
    r = rf(x, prec, options)
  File "/path/to/lib/python3.7/site-packages/sympy/core/evalf.py", line 678, in evalf_pow
    re, im, re_acc, im_acc = evalf(base, prec + 5, options)
  File "/path/to/lib/python3.7/site-packages/sympy/core/evalf.py", line 1309, in evalf
    r = rf(x, prec, options)
  File "/path/to/lib/python3.7/site-packages/sympy/core/evalf.py", line 847, in evalf_log
    re = mpf_log(mpf_abs(xre), prec, rnd)
  File "/path/to/lib/python3.7/site-packages/mpmath/libmp/libelefun.py", line 736, in mpf_log
    m = -log_agm(to_fixed(x, wp), wp)
  File "/path/to/lib/python3.7/site-packages/mpmath/libmp/libmpf.py", line 518, in to_fixed
    else:           return man >> (-offset)
TypeError: unsupported operand type(s) for >>: 'int' and 'float'
>>> 

Using sympy 1.4. Using a variable seems to be required to reproduce this.

[oscarbenjamin]

It goes wrong here:

sympy/sympy/core/evalf.py

Lines 350 to 356 in 58e7820

	ire, iim, ire_acc, iim_acc = evalf(
	re_im - nint, 10, options) # don't need much precision
	assert not iim
	size = -fastlog(ire) + 2 # -ve b/c ire is less than 1
	if size > prec:
	ire, iim, ire_acc, iim_acc = evalf(
	re_im, size, options)

The reim variable is re(log(test)/log(2)) which is zero (if test=1). Then fastlog returns -oo which then give size=oo which is used as the prec argument in the recursive call to evalf. Perhaps size should be bounded at maxprec here.

[Mygod]

EDIT: Moved to #17642

[jksuom]

The reim variable is re(log(test)/log(2)) which is zero (if test=1).

re_im comes from fzero assigned to ire on this line:

sympy/sympy/core/evalf.py

Line 315 in 1ea2b20

ire, iim, ire_acc, iim_acc = evalf(expr, assumed_size, options)

It seems that the internal evalf is expected to return None instead of fzero. In this case, evalf_mul should have checked the result of log(1)*log(2)**-1.