Closed
Description
Bug report
Bug summary
It seems that log-scale plotting fails whenever 1e-323 is used as a tick.
Code for reproduction
import matplotlib.pyplot as plt
plt.semilogy([1, 2], [1, 2e-323])
plt.show()or
import matplotlib.pyplot as plt
plt.semilogy([1, 2, 3], [1, 2e-323, 0])
plt.show()2e-323 can be replaced with 3e-323, 4e-323, ..., 10e-323 (replacing it with 1e-323 or 11e-323 makes it work).
Actual outcome
# long traceback
File "/some_path/matplotlib/lib/matplotlib/ticker.py", line 1157, in __call__
coeff = round(x / b ** exponent)
OverflowError: cannot convert float infinity to integerExpected outcome
A plot that is similar to when 2e-323 is replaced with 1e-323 or 11e-323.
Debugger result
The debugger shows that the issue is in LogFormatter.__call__ when x == 1e-323:
def __call__(self, x, pos=None):
# docstring inherited
if x == 0.0: # Symlog
return '0'
x = abs(x) # x = 1e-323
b = self._base # b = 10.0
# only label the decades
fx = math.log(x) / math.log(b) # fx = -323.0051853474518
is_x_decade = is_close_to_int(fx) # is_x_decade = False
exponent = round(fx) if is_x_decade else np.floor(fx) # exponent = -324.0
coeff = round(x / b ** exponent) # b ** exponent == 0.0Matplotlib version
- Operating system: Linux
- Matplotlib version: current master branch
- Matplotlib backend: TkAgg
- Python version: 3.7.8
Installed from source (using pip). The same issue appears also in 3.3.2.