Fix slowdown in CategoricalIndex.__repr__ (rapidsai#16665)

Fixes: rapidsai#13297 

This PR fixes a slow-down in performing repr of a `CategoricalIndex` when there are too many unique values. There was no other choice to fix this in a better way by using public APIs, because all the public APIs seem to be performing categories validation even if `fastpath=True`.

Authors:
  - GALI PREM SAGAR (https://github.com/galipremsagar)

Approvers:
  - Matthew Murray (https://github.com/Matt711)

URL: rapidsai#16665

python/cudf/cudf/core/index.py

@@ -1443,7 +1443,21 @@ def __repr__(self):
                     output[:break_idx].replace("'", "") + output[break_idx:]
                 )
             else:
-                output = repr(preprocess.to_pandas())
+                # Too many non-unique categories will cause
+                # the output to take too long. In this case, we
+                # split the categories into data and categories
+                # and generate the repr separately and
+                # merge them.
+                pd_cats = pd.Categorical(
+                    preprocess.astype(preprocess.categories.dtype).to_pandas()
+                )
+                pd_preprocess = pd.CategoricalIndex(pd_cats)
+                data_repr = repr(pd_preprocess).split("\n")
+                pd_preprocess.dtype._categories = (
+                    preprocess.categories.to_pandas()
+                )
+                cats_repr = repr(pd_preprocess).split("\n")
+                output = "\n".join(data_repr[:-1] + cats_repr[-1:])
 
             output = output.replace("nan", str(cudf.NA))
         elif preprocess._values.nullable:

python/cudf/cudf/testing/_utils.py

@@ -1,6 +1,7 @@
 # Copyright (c) 2020-2024, NVIDIA CORPORATION.
 
 import itertools
+import signal
 import string
 from collections import abc
 from contextlib import contextmanager
@@ -368,3 +369,23 @@ def sv_to_udf_str_testing_lowering(context, builder, sig, args):
     return cast_string_view_to_udf_string(
         context, builder, sig.args[0], sig.return_type, args[0]
     )
+
+
+class cudf_timeout:
+    """
+    Context manager to raise a TimeoutError after a specified number of seconds.
+    """
+
+    def __init__(self, seconds, *, timeout_message=""):
+        self.seconds = int(seconds)
+        self.timeout_message = timeout_message
+
+    def _timeout_handler(self, signum, frame):
+        raise TimeoutError(self.timeout_message)
+
+    def __enter__(self):
+        signal.signal(signal.SIGALRM, self._timeout_handler)
+        signal.alarm(self.seconds)
+
+    def __exit__(self, type, value, traceback):
+        signal.alarm(0)

python/cudf/cudf/tests/test_repr.py

@@ -1480,3 +1480,14 @@ def test_interval_index_repr():
     gi = cudf.from_pandas(pi)
 
     assert repr(pi) == repr(gi)
+
+
+def test_large_unique_categories_repr():
+    # Unfortunately, this is a long running test (takes about 1 minute)
+    # and there is no way we can reduce the time
+    pi = pd.CategoricalIndex(range(100_000_000))
+    gi = cudf.CategoricalIndex(range(100_000_000))
+    expected_repr = repr(pi)
+    with utils.cudf_timeout(2, timeout_message="Failed to repr fast enough"):
+        actual_repr = repr(gi)
+    assert expected_repr == actual_repr