perf: translate for canonicalize_name

pypa/packaging#1030 · merged 2026-01-05 · base 52e57f2003 → head 3803ce341b · ran on Docker

BLOCK

At least one behaviour change looks like an unintended regression. The “faster” code does up to 5× the work on valid inputs, growing with input size (complexity regression).

Known regressions in this PR

✓unicode-lowercase · fixed by #1064 · verified · first found by Breakwater

Counterexamples · 3 groups of 8 classes: 8 likely bug

Output changed

known regression: unicode-lowercase

inputcanonicalize_name("İ")

basevalue: "i̇"

headvalue: "İ"

Likely bugvotes: 3 likely-bug

The PR aims only to speed up canonicalize_name while preserving the original semantics (lower‑casing per PEP 503). The new implementation only lower‑cases ASCII letters, so non‑ASCII characters like ‘İ’ are not lower‑cased, changing the output.

Emulates ``re.sub(r

perf: translate for canonicalize_name

Found by: Nemotron strategy search, Ultra's planned example, hostile mutation, search sample · also caught by: Plain Hypothesis, Mutations of the PR's own test strings, Nemotron plan only

ran onlocal Docker · image bw-pair:d09088ff31a7e098

Regression test verified passes on base, fails on head

def test_1030_p1_value_0():
    """Breakwater P1: base behaviour pinned (found by differential testing; label: likely-bug)."""
    from packaging.utils import canonicalize_name
    assert canonicalize_name('İ') == 'i̇'

Reproduce

uv run breakwater probe --repo https://github.com/pypa/packaging --base 52e57f20037f --head 3803ce341b59 --target packaging.utils:canonicalize_name --backend docker --input "İ"

2 near-duplicate classes same behaviour, inputs with extra traits

canonicalize_name("İ ") base value: "i̇ " · head value: "İ "unicode-lowercase test verified
canonicalize_name("İ\u0000") base value: "i̇\u0000" · head value: "İ\u0000"unicode-lowercase test verified

Exception behaviour changed

inputcanonicalize_name(["İ"])

baseraises TypeError

headraises AttributeError

Neither exception type subclasses the other.

Likely bugvotes: 3 likely-bug

The change introduces a different exception type (AttributeError vs TypeError) when processing the same input, which is not mentioned in the PR title, docs, or tests and therefore is an unintended regression.

perf: translate for canonicalize_name

Emulates ``re.sub(r

Found by: Nemotron strategy search, search sample · no baseline found it

ran onlocal Docker · image bw-pair:d09088ff31a7e098

Regression test verified passes on base, errors on head

def test_1030_p2_outcome_3():
    """Breakwater P2: base behaviour pinned (found by differential testing; label: likely-bug)."""
    import pytest
    from packaging.utils import canonicalize_name
    with pytest.raises(TypeError) as excinfo:
        canonicalize_name(['İ'])
    assert type(excinfo.value) is TypeError  # exact type: a subclass would be a behaviour change

Equality / ordering changed

acanonicalize_name("İ")

bcanonicalize_name("py3")

basea > b → False

heada > b → True

Likely bugvotes: 3 likely-bug

Ordering changes because the new code no longer lower‑cases Unicode letters, so comparison results differ from the original specification that required ``.lower()``.

Emulates ``re.sub(r

perf: translate for canonicalize_name

Found by: Ultra's planned example, relations pass · also caught by: Plain Hypothesis, Nemotron plan only

ran onlocal Docker · image bw-pair:d09088ff31a7e098

Regression test verified passes on base, fails on head

def test_1030_p3_gt_4():
    """Breakwater P3: base behaviour pinned (found by differential testing; label: likely-bug)."""
    from packaging.utils import canonicalize_name
    a = canonicalize_name('İ')
    b = canonicalize_name('py3')
    assert (a > b) is False

3 near-duplicate relation flipes same behaviour, inputs with extra traits

canonicalize_name("İ") vs canonicalize_name("py3") base a < b → True · head a < b → False test verified
canonicalize_name("ẞ") vs canonicalize_name("K") base a > b → True · head a > b → False test verified
canonicalize_name("ẞ") vs canonicalize_name("K") base a < b → False · head a < b → True test verified

Regression tests

8 of 8 classes have a generated pytest that pins the old behaviour. Each passed on the base commit and failed on the head commit in the sandbox before it was offered.

Download test_1030.py

Performance · canonicalize_name

Worst family with mostly valid inputs: 4.52× the instructions of base (many_underscores_and_dots).

Measured in backtest packaging-v2 (same commits, run bt-1030-e38d1c, image bw-pair:a959e7a1becfaee9). The semantic results above are from a later run; perf doesn't use the semantic planner, so it was not re-measured.

R1.2
R4
R1
F3
F1

Head/base cachegrind instructions per call, by input size (log-log). Each line is one input family Nemotron proposed; points above 1× are where the "faster" code does more work.

Family	Inputs	Ratio range	Slope base → head	Verdict
`R1.2`	Names starting/ending with multiple dots or underscoresclaimed speedup · expand:R1	0.22× – 2.28×	0.86 → 0.52	slower
`R4`	many_underscores_and_dotssuspected slowdown · archive	1.04× – 4.52×	0.56 → 0.78	complexity-regression
`R1`	Normal package names are short‑ish alphanumerics with only a few '_' or '.' characters. The new translate‑table + single‑pass replace handles this in linear timclaimed speedup · archive	0.07× – 1.66×	0.86 → 0.44	slower
`F3`	typical_package_namesclaimed speedup · plan	0.15× – 1.90×	0.74 → 0.48	slower
`F1`	typical package names containing mixed case, dots and underscoresclaimed speedup · plan	0.10× – 1.13×	0.94 → 0.59	slower

Receipt

Target	Backend	Image	Run	Audit	Adjudication
`packaging.utils:canonicalize_name`	local Docker	`bw-pair:d09088ff31a7e098`	`bt-1030-0d870d`	$0.027	$0.021

Nemotron cost on Nebius Token Factory. Backend: where the untrusted code ran. local Docker (an image cited by its content-addressed tag). Every image is content-addressed (repo at each commit + harness), so every card re-runs from the image and seed 0. Raw data: audit, adjudication.