Fix tail(n, seq) returning whole sequence for n <= 0

toolz/itertoolz.py

@@ -328,12 +328,20 @@ def tail(n, seq):
     >>> tail(2, [10, 20, 30, 40, 50])
     [40, 50]
 
+    ``n`` must be non-negative; a negative ``n`` raises ``ValueError`` rather
+    than silently masking an upstream bug in the caller.
+
     See Also:
         drop
         take
     """
+    if n < 0:
+        raise ValueError("n must be non-negative")
     try:
-        return seq[-n:]
+        # seq[len(seq) - n:] is correct for 0 <= n <= len(seq); clamp the
+        # lower bound so n > len(seq) returns the whole sequence instead of
+        # wrapping to a negative index, and n == 0 returns the empty tail.
+        return seq[max(len(seq) - n, 0):]
     except (TypeError, KeyError):
         return tuple(collections.deque(seq, n))
 

toolz/tests/test_itertoolz.py

@@ -189,6 +189,20 @@ def test_tail():
     assert list(tail(3, 'ABCDE')) == list('CDE')
     assert list(tail(3, iter('ABCDE'))) == list('CDE')
     assert list(tail(2, (3, 2, 1))) == list((2, 1))
+    # n larger than the sequence returns the whole sequence.
+    assert list(tail(10, 'ABC')) == list('ABC')
+    assert list(tail(10, iter('ABC'))) == list('ABC')
+    # tail(0, seq) is the last zero elements: empty, consistently for any
+    # iterable. Regression for gh #626: seq[-0:] == seq[0:] wrongly returned
+    # the whole sequence for sliceable inputs, while the deque fallback
+    # returned empty, so the two input paths disagreed.
+    assert list(tail(0, 'ABCDE')) == []
+    assert list(tail(0, iter('ABCDE'))) == []
+    assert list(tail(0, (3, 2, 1))) == []
+    # A negative n is a programming error (a masked upstream bug), so it
+    # raises rather than silently returning empty, matching nth's contract.
+    assert raises(ValueError, lambda: tail(-2, 'ABCDE'))
+    assert raises(ValueError, lambda: tail(-2, iter('ABCDE')))
 
 
 def test_drop():