BUG: `IntervalIndex.unique()` only contains the first interval if all interval borders are negative

[johannes-mueller]

Pandas version checks

• I have checked that this issue has not already been reported.

• I have confirmed this bug exists on the latest version of pandas.

• I have confirmed this bug exists on the main branch of pandas.

Reproducible Example

import pandas as pd

print(pd.__version__)

idx_pos = pd.IntervalIndex.from_tuples([(3, 4), (3, 4), (2, 3), (2, 3), (1, 2), (1, 2)])

print(idx_pos.unique())
assert idx_pos.unique().shape == (3,)  # succeeds

idx_neg = pd.IntervalIndex.from_tuples([(-4, -3), (-4, -3), (-3, -2), (-3, -2), (-2, -1), (-2, -1)])

print(idx_neg.unique())
assert idx_neg.unique().shape == (3,), f"Actual shape: {idx_neg.unique().shape}"

Issue Description

Output with current main:

3.0.0.dev0+2250.g13f7b8b7e3
IntervalIndex([(3, 4], (2, 3], (1, 2]], dtype='interval[int64, right]')
IntervalIndex([(-4, -3]], dtype='interval[int64, right]')
Traceback (most recent call last):
  File "/home/jmu3si/tmp/pd-demo.py", line 12, in <module>
    assert idx_neg.unique().shape == (3,), f"Actual shape: {idx_neg.unique().shape}"
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
AssertionError: Actual shape: (1,)

Only the interval (-4, 3] appears in the uniqued index.

A couple of other observations:

• The same result occurs with closed="left"
• Intervals that are not fully negative, e.g. (-2, 0] also appear in the uniqued index
• This does not seem to be a regression. I reproduced it all the way back to pandas-1.4.3

Expected Behavior

Expect correct unique index for index_neg to be

IntervalIndex([(-4, -3], (-3, -2], (-2, -1]], dtype='interval[int64, right]') as it correctly did with the positive interval index.

Installed Versions

INSTALLED VERSIONS

commit : 13f7b8b
python : 3.13.2
python-bits : 64
OS : Linux
OS-release : 6.8.0-60-generic
Version : #63~22.04.1-Ubuntu SMP PREEMPT_DYNAMIC Tue Apr 22 19:00:15 UTC 2
machine : x86_64
processor : x86_64
byteorder : little
LC_ALL : None
LANG : de_DE.UTF-8
LOCALE : de_DE.UTF-8

pandas : 3.0.0.dev0+2250.g13f7b8b7e3
numpy : 2.2.6
dateutil : 2.9.0.post0
pip : 25.1.1
Cython : 3.1.2
sphinx : 8.2.3
IPython : 9.4.0
adbc-driver-postgresql: None
adbc-driver-sqlite : None
bs4 : 4.13.4
bottleneck : 1.5.0
fastparquet : 2024.11.0
fsspec : 2025.7.0
html5lib : 1.1
hypothesis : 6.136.1
gcsfs : 2025.7.0
jinja2 : 3.1.6
lxml.etree : 6.0.0
matplotlib : 3.10.3
numba : 0.61.2
numexpr : 2.11.0
odfpy : None
openpyxl : 3.1.5
psycopg2 : 2.9.10
pymysql : 1.4.6
pyarrow : 21.0.0
pyiceberg : 0.9.1
pyreadstat : 1.3.0
pytest : 8.4.1
python-calamine : None
pytz : 2025.2
pyxlsb : 1.0.10
s3fs : 2025.7.0
scipy : 1.16.0
sqlalchemy : 2.0.41
tables : 3.10.2
tabulate : 0.9.0
xarray : 2025.7.1
xlrd : 2.0.2
xlsxwriter : 3.2.5
zstandard : 0.23.0
qtpy : None
pyqt5 : None

[khemkaran10]

I think the issue is with this logic:

pandas/pandas/core/arrays/interval.py

Lines 1992 to 2000 in 073710f

	def unique(self) -> IntervalArray:
	# No overload variant of "__getitem__" of "ExtensionArray" matches argument
	# type "Tuple[slice, int]"
	nc = unique(
	self._combined.view("complex128")[:, 0] # type: ignore[call-overload]
	)
	nc = nc[:, None]
	return self._from_combined(nc)

Test Script:

import pandas as pd
idx_pos = pd.IntervalIndex.from_tuples([(3, 4), (3, 4), (2, 3), (2, 3), (1, 2), (1, 2)])
print(idx_pos)

ia = idx_pos._data

print("\nCombined (ia._combined):")
print(ia._combined)

combined_view = ia._combined.view("complex128")
print("\ncombined_view (as complex128):")
print(combined_view)

print("idx_pos unique()")
print(idx_pos.unique())

print("-------------------------------------------------------------------")

idx_neg = pd.IntervalIndex.from_tuples([(-4, -3), (-4, -3), (-3, -2), (-3, -2), (-2, -1), (-2, -1)])
print(idx_neg)
ia = idx_neg._data

print("\nCombined (ia._combined):")
print(ia._combined)

combined_view = ia._combined.view("complex128")
print("\ncombined_view (as complex128):")
print(combined_view)
print("idx_neg unique()")
print(idx_neg.unique())

Output:

IntervalIndex([(3, 4], (3, 4], (2, 3], (2, 3], (1, 2], (1, 2]], dtype='interval[int64, right]')

Combined (ia._combined):
[[3 4]
 [3 4]
 [2 3]
 [2 3]
 [1 2]
 [1 2]]

combined_view (as complex128):
[[1.5e-323+2.0e-323j]
 [1.5e-323+2.0e-323j]
 [9.9e-324+1.5e-323j]
 [9.9e-324+1.5e-323j]
 [4.9e-324+9.9e-324j]
 [4.9e-324+9.9e-324j]]

idx_pos unique()
IntervalIndex([(3, 4], (2, 3], (1, 2]], dtype='interval[int64, right]')
-------------------------------------------------------------------
IntervalIndex([(-4, -3], (-4, -3], (-3, -2], (-3, -2], (-2, -1], (-2, -1]], dtype='interval[int64, right]')

Combined (ia._combined):
[[-4 -3]
 [-4 -3]
 [-3 -2]
 [-3 -2]
 [-2 -1]
 [-2 -1]]

combined_view (as complex128):
[[nan+nanj]
 [nan+nanj]
 [nan+nanj]
 [nan+nanj]
 [nan+nanj]
 [nan+nanj]]

idx_neg unique()
IntervalIndex([(-4, -3]], dtype='interval[int64, right]')

[rhshadrach]

Thanks for the report. Confirmed on main. Further investigations and PRs to fix are welcome.