Pull Request: Fix IndexLocator.tick_values returning values greater than vmax

Summary

This PR fixes a bug in IndexLocator.tick_values() where tick values could exceed the specified vmax parameter. This issue manifested most notably when using colorbars with discrete colormaps and NoNorm normalization, where cbar.get_ticks() would return an incorrect array containing values outside the valid range.

The Problem

When creating a colorbar using a ScalarMappable with NoNorm normalization and a discrete colormap (e.g., viridis with a fixed number of colors), the ticks returned by colorbar.get_ticks() did not align with the visually displayed tick positions. Specifically, calling cbar.set_ticks(cbar.get_ticks()) would unexpectedly change the ticks because get_ticks() returned values outside the valid range.

Example reproduction:

import matplotlib.pyplot as plt
from matplotlib import cm, colors

data = [1, 2, 3, 4, 5]
fig, ax = plt.subplots()
cbar = fig.colorbar(
    cm.ScalarMappable(norm=colors.NoNorm(), cmap=plt.get_cmap("viridis", len(data))),
    ax=ax
)
print(cbar.get_ticks())
# Actual (before fix): [0. 1. 2. 3. 4. 5.]
# Expected (after fix): [0. 1. 2. 3. 4.]

Root Cause

The issue was in IndexLocator.tick_values() in lib/matplotlib/ticker.py:

def tick_values(self, vmin, vmax):
    return self.raise_if_exceeds(
        np.arange(vmin + self.offset, vmax + 1, self._base))

The vmax + 1 in the np.arange call was intended to make the interval inclusive of vmax (since np.arange produces a semi-open interval [start, stop)). However, this overshot the target — combined with certain offset values, it could generate tick positions that exceed vmax. For instance, with offset=0.5 and vmax=4, the array [0.5, 1.5, 2.5, 3.5, 4.5] would be generated, where 4.5 > vmax.

The Fix

Replaced vmax + 1 with vmax + 1e-12 as the stop value for np.arange. This provides a minimal epsilon beyond vmax that ensures vmax itself is included in the closed interval [vmin, vmax], without overshooting to produce values beyond vmax:

def tick_values(self, vmin, vmax):
    # We want tick values in the closed interval [vmin, vmax].
    # Since np.arange(start, stop) returns values in the semi-open interval
    # [start, stop), we add a minimal offset so that stop = vmax + eps
    tick_values = np.arange(vmin + self.offset, vmax + 1e-12, self._base)
    return self.raise_if_exceeds(tick_values)

This directly generates the correct number of ticks without any post-filtering, which is both cleaner and more efficient.

Changes

Modified Files

1. lib/matplotlib/ticker.py

   • Modified IndexLocator.tick_values() to use vmax + 1e-12 instead of vmax + 1 as the np.arange stop value, ensuring tick values stay within the [vmin, vmax] closed interval

2. lib/matplotlib/tests/test_ticker.py

   • Added test_tick_values_not_exceeding_vmax() test method to TestIndexLocator class
   • Tests verify correct tick generation for:
     • Integer offset (offset=0)
     • Fractional offset (offset=0.5)
     • Step size greater than 1 (base=2)

Test Plan

• Added unit tests for the fix in test_ticker.py
• Run the full test suite: pytest lib/matplotlib/tests/test_ticker.py -v
• Verify the reproduction case from the issue is fixed
• Check that existing colorbar functionality is not affected

New Test Cases

def test_tick_values_not_exceeding_vmax(self):
    """
    Test that tick_values does not return values greater than vmax.
    """
    # Test case where offset=0 could cause vmax to be included incorrectly
    index = mticker.IndexLocator(base=1, offset=0)
    assert_array_equal(index.tick_values(0, 4), [0, 1, 2, 3, 4])

    # Test case with fractional offset
    index = mticker.IndexLocator(base=1, offset=0.5)
    assert_array_equal(index.tick_values(0, 4), [0.5, 1.5, 2.5, 3.5])

    # Test case with base > 1
    index = mticker.IndexLocator(base=2, offset=0)
    assert_array_equal(index.tick_values(0, 5), [0, 2, 4])

Related Issues

• Fixes [Bug]: Colorbar get_ticks() return the incorrect array #31086

Checklist

• Code follows the project's style guidelines
• Changes are minimal and focused on the specific bug
• Added regression test to prevent future occurrences
• Commit message follows conventional format
• CI tests pass