Bug summary

In this simple test I make two scatter plots. The first one has values drawn from a uniform distribution between 0 and 1000. The second plot is the same except that values are now drawn between 0 and 1 only.

Finally, I set set_xscale("symlog") and set_yscale("symlog") on both plots. The result is that the first plot responds correctly, while the second plot does not change, and the data points remain in the same linear display.

Code for reproduction

# Modules #
import random
from matplotlib import pyplot

# New graph #
fig, axes = pyplot.subplots()

# Constants #
upper_bound = 1.0 # try it with 1000.0 instead and it works

# Do it #
x = [random.uniform(0.0, upper_bound) for x in range(1000)]
y = [random.uniform(0.0, upper_bound) for x in range(1000)]
pyplot.plot(x, y, 'o')
axes.set_xscale("symlog")
axes.set_yscale("symlog")

Actual outcome

With upper_bound = 1.0.

Expected outcome

With upper_bound = 1000.0.

Additional information

This is quite a serious bug as it silently produces broken or miss-leading visualizations which then bias the interpretation of critical scientific data.

Operating system

macOS

Matplotlib Version

3.6.2

Matplotlib Backend

MacOSX

Python version

Python 3.10.8

Jupyter version

No response

Installation

conda