🧪 test: replace wall-clock checks with a bench

tests/tokenizer/test_xml.py

@@ -1,7 +1,6 @@
 from __future__ import annotations
 
-import time
-from typing import TYPE_CHECKING, Final, cast
+from typing import Final, cast
 
 import pytest
 from bench.operations import INPUTS
@@ -19,9 +18,6 @@
     parse_xml,
 )
 
-if TYPE_CHECKING:
-    from collections.abc import Callable
-
 
 def elements(node: Element) -> list[Element]:
     return [child for child in node if isinstance(child, Element)]
@@ -698,30 +694,6 @@ def _plain_repeated(count: int) -> str:
     return "<r>" + f"<e {attrs}/>" * 2 + "</r>"  # the second element re-reads already-interned names
 
 
-def _min_parse_seconds(markup: str) -> float:
-    best = float("inf")
-    for _ in range(5):  # the floor over repeats drops scheduler noise, which only ever inflates a sample
-        start = time.perf_counter()
-        parse_xml(markup)
-        best = min(best, time.perf_counter() - start)
-    return best
-
-
-@pytest.mark.parametrize(
-    "build",
-    [
-        pytest.param(_namespaced_distinct, id="expanded-name"),
-        pytest.param(_plain_repeated, id="raw-name"),
-    ],
-)
-def test_xml_duplicate_detection_scales_linearly(build: Callable[[int], str]) -> None:
-    base: Final = _min_parse_seconds(build(10_000))
-    quadrupled: Final = _min_parse_seconds(build(40_000))
-    # Linear work quadruples with the input; the reverted nested scan is ~16x. 8x sits two-fold
-    # below the quadratic floor and two-fold above the linear one, so jitter cannot flip it.
-    assert quadrupled < base * 8
-
-
 @pytest.mark.parametrize("count", [pytest.param(2000, id="grows-and-rehashes")])
 def test_xml_many_distinct_namespaced_attributes_parse(count: int) -> None:
     root: Final = parse_xml(_namespaced_distinct(count)).find("r")
@@ -823,7 +795,7 @@ def test_round_trip_against_lxml() -> None:
     assert [node.tag.split(":")[-1] for node in (ours, *ours.descendants) if isinstance(node, Element)] == their_locals
 
 
-@pytest.mark.parametrize(("index", "count"), [(0, 128), (1, 1)], ids=["many", "single"])
+@pytest.mark.parametrize(("index", "count"), [(0, 128), (1, 1), (2, 1_000)], ids=["many", "single", "wide"])
 @pytest.mark.oracle
 def test_lxml_namespace_benchmark_output(index: int, count: int) -> None:
     etree: Final = pytest.importorskip("lxml.etree", exc_type=ImportError)

tests/url/test_idna.py

@@ -13,8 +13,8 @@
 from __future__ import annotations
 
 import re
-import time
 from pathlib import Path
+from typing import Final
 
 import pytest
 
@@ -347,13 +347,11 @@ def test_host_at_the_input_cap_still_encodes() -> None:
     assert _url_to_ascii("\u3400" * 16384).startswith("xn--")
 
 
-def test_oversize_host_is_rejected_before_the_quadratic_passes_run() -> None:
-    """A host past the cap raises at once, in bounded time, instead of driving the O(n^2) reorder and punycode loops."""
-    host = "\u0316\u0301" * 60000
-    start = time.process_time()
-    with pytest.raises(ValueError, match="exceeds the IDNA input limit of 16384"):
-        _url_to_ascii(host)
-    assert time.process_time() - start < 2.0
+def test_oversize_host_keeps_its_unicode_form() -> None:
+    # past the 16384-code-point cap the host skips the O(n^2) reorder and punycode passes and keeps its Unicode form;
+    # U+0316 then U+0301 alternate combining classes 220 and 230, the input an insertion-sort reorder is O(n^2) on
+    url: Final = "http://" + "\u0316\u0301" * 60000 + "/"
+    assert normalize_url(url) == url
 
 
 @pytest.mark.parametrize(

tools/bench/ci.py

@@ -240,6 +240,7 @@ def _book_html() -> str:
     "parse-xml-text-references": ("parse-xml-text", 1),
     "parse-xml-text-tiny": ("parse-xml-text", 2),
     "parse-xml-prefixes-single": ("parse-xml-prefixes", 1),
+    "parse-xml-prefixes-wide": ("parse-xml-prefixes", 2),
     "query-closest-distinct": ("query-closest", 1),
     "query-parents-distinct": ("query-parents", 1),
     "query-siblings-single": ("query-siblings", 1),

tools/bench/operations.py

@@ -1929,7 +1929,7 @@ def _encoding_result_cases() -> tuple[tuple[str, bytes], ...]:
             + " ".join(f'p{index}:value="x"' for index in range(count))
             + "/>",
         )
-        for count in (128, 1)
+        for count in (128, 1, 1_000)
     ),
     "parse-xml-names": lambda: (
         (