The behavior of log scales when dealing with non-positive values is a common gotcha issue.

This seems to only be documented at

https://matplotlib.org/api/scale_api.html#matplotlib.scale.LogScale and http://matplotlib.org/devdocs/gallery/pylab_examples/log_demo.html

The log-demo should be expanded to a tutorial with prose explaining

• npos clip vs mask behavior (fill_between incorrect with log y-axis and value 0 #8623 Addresses #8045 (set_yscale("log") inconsistent with semilogy()) #8056 and links there in)
• loglog vs semilog vs ax.set_scale vs plt.xscale/plt.yscale
• changing the base
• how ticking works with log scale (issues around http://matplotlib.org/users/dflt_style_changes.html?highlight=default%20changes#tick-label-formatting and issue around it (Minor ticks on log-scale colorbar are not cleared #8358 Default locator for log-scale messes up minor ticks sometimes #7200 ticklabels on log axes sometimes adds overwritten labels on top of existing ones  #8622 Issue #6967 - symlog + minor ticks = exception #8353 Incorrect display of minor ticks in Log Scale plot with large font size #8419 weird bug when using custom formatter; tick labels overlapping #8401 Log-scale ignores Formatter #8517 Colorbar with imshow(logNorm) shows unexpected minor ticks #8307)
• maybe fold in the symlog demo (https://matplotlib.org/examples/pylab_examples/symlog_demo.html)