Repository navigation
[Bug]: boxplot_stats raises IndexError when all values are masked #32508
Description
Bug summary
Bug report
matplotlib.cbook.boxplot_stats() raises an IndexError when given a masked array in which all values are masked.
Minimal reproducible example
import numpy as np
from matplotlib import cbook
data = np.ma.masked_array(
[10.0, 20.0],
mask=[True, True]
)
cbook.boxplot_stats(data)Expected behavior
The function should handle an all-masked dataset gracefully, following the existing empty-data behavior, instead of raising an exception.
Actual behavior
After the masked values are removed, the resulting array is empty. The subsequent percentile calculation raises an IndexError.
Additional context
Partially masked data appears to be covered by the existing test_boxplot_masked test, but an all-masked dataset does not appear to be covered.
I reproduced this behavior using the local Matplotlib checkout. The proposed fix has not yet been submitted for review.
Code for reproduction
import numpy as np
from matplotlib import cbook
data = np.ma.masked_array([10.0, 20.0], mask=[True, True])
cbook.boxplot_stats(data)Actual outcome
The code raises an IndexError when all elements of the masked array are masked.
Example error with NumPy 2.x:
IndexError: index -1 is out of bounds for axis 0 with size 0
The exact error message may vary depending on the NumPy version.
Expected outcome
The function should handle an array with all values masked gracefully, without raising an exception. It should return appropriate statistics for the empty data remaining after masked values are excluded, consistent with the function's existing empty-data behavior.
Additional information
No response
Operating system
No response
Matplotlib Version
3.11.1
Matplotlib Backend
No response
Python version
No response
Jupyter version
No response
Installation
None
Activity
lucaznch commentedon Oct 20, 2026
Thanks for the report. I'm new to git bisect but assuming I used it correctly it points to f8b3b27
rcomer commentedon Oct 20, 2026
Confirmed, the exception came in at f8b3b27, though before that the mask was just ignored so you got spurious statistics. Looks like we just need to move the lines that get the unmasked values above the line that checks for zero-length input.
mansi-tomar-sudo commentedon Oct 20, 2026
Thanks for confirming the issue and identifying the likely cause! I'd like to work on the fix and add a regression test for the all-masked input case. Could you please assign this issue to me? I'd also like to confirm that I'm welcome to submit a PR for this fix.
rcomer commentedon Oct 20, 2026
@mansi-tomar-sudo you are welcome to submit a pull request for any issue, but we do not assign them. Please see our contributors’ guide:
https://matplotlib.org/devdocs/devel/index.html
mansi-tomar-sudo commentedon Oct 25, 2026
Hi! I’ve submitted PR #32515 to address the all-masked input case in
boxplot_stats. I’ve updated the PR description and AI disclosure, fixed the formatting issues, and rerun the relevant pre-commit hooks. All GitHub checks are now passing.Could someone please take a look at the PR when time permits and let me know if any further changes are needed? Thank you!