perf: use str.replace in Pointer escape/unescape

[berendkleinhaneveld]

Summary

Replaces the regex-based escape/unescape helpers in patchdiff/pointer.py with str.replace. For single-character substitutions, str.replace is ~3× faster than a compiled regex, and Pointer.__str__ calls escape(str(t)) for every token in the pointer — so every to_json / to_str_paths call benefits.

The four module-level re.compile objects (slash_re, tilde_re, tilde0_re, tilde1_re) and import re are removed. grep confirms no other file in the repo imports those names.

Micro-bench (reproducer from plan)

import timeit
from patchdiff.pointer import escape, unescape
print('escape  100k:', timeit.timeit(lambda: escape('somekey_42'),  number=100_000) * 1000, 'ms')
print('unescape 100k:', timeit.timeit(lambda: unescape('somekey_42'), number=100_000) * 1000, 'ms')

	escape 100k	unescape 100k
master	23.1 ms	23.5 ms
this branch	6.8 ms	7.5 ms
speedup	3.4×	3.1×

(Target was ≥2× on escape.)

Macro-bench

uv run pytest benchmarks/benchmark.py --benchmark-only -k "pointer" — pointer-append and pointer-evaluate groups unchanged, as expected (escape only runs on serialization, not append/evaluate):

test_pointer_append                  180.88 ns
test_pointer_evaluate_deep_list      1.28 us
test_pointer_evaluate_deep_dict      2.21 us

Correctness

• Full existing test suite passes unchanged (269 tests).
• Added test_escape_unescape_roundtrip in tests/test_pointer.py covering: '', 'plain', 'has/slash', 'has~tilde', '~/mix', '///', '~~~', and the tricky '~01' (which must escape to '~001', not be confused with the RFC 6901 ~0 + '1' sequence). For each token the test asserts both escape(t) matches RFC 6901 and unescape(escape(t)) == t.

RFC 6901 escape order (~ → ~0 before / → ~1 on escape; reverse on unescape) is preserved.

Test plan

• uv run pytest tests/test_pointer.py -v
• uv run pytest -x -q
• uv run pytest benchmarks/benchmark.py --benchmark-only -k "pointer"
• uv run ruff check
• uv run ruff format --check

🤖 Generated with Claude Code