Increase URL parsing speed by approximately 2 times

[AyumuKasuga]

urlparse rejects the ASCII control characters that are not permitted. To find these characters, it examined each character of the URL, and each character of each keyword component. It used a generator in Python. A profile of httpx2.URL(...) shows that this generator uses approximately 43 percent of the parse time.

The ASCII characters that are not printable are exactly the C0 control characters and DEL. One compiled regex search can replace the generator. The match object gives the character and also its position. The second pass to find the character is not necessary. The error messages and the character positions do not change.

This commit also makes two smaller changes to quote. urlparse calls this function five times for each URL.

• Call PERCENT_ENCODED_REGEX.finditer(...) and not re.finditer(PERCENT_ENCODED_REGEX, ...). The second form repeats the re._compile cache operation at each call.
• Call percent_encoded immediately if the string contains no '%' character. Most strings do not contain this character.

The measurements use 40 rounds. Each variant runs in its own process. The order of the two variants changes at each round. A change in the condition of the machine thus has an equal effect on both variants. The values are medians:

short           (25 characters)     4.61us ->   2.57us   1.79x
typical         (69 characters)     7.42us ->   3.64us   2.04x
percent-encoded (70 characters)     7.85us ->   4.91us   1.60x
long query      (1 KB)             30.31us ->  12.28us   2.47x

For a full client request cycle with MockTransport, the time decreases from 25.31us to 20.54us for a typical URL. For a URL that has a long query string, the time decreases from 48.90us to 29.13us. The two sets of samples do not overlap. In each case, the slowest of the 40 new samples is faster than the fastest of the 40 old samples.

A test compares the new character class with the previous condition for all 1114112 Unicode code points. The two sets of characters are equal. A second test compares the new implementation with the previous one for approximately 60000 fixed and random URLs and keyword component calls. The results and the exception messages are the same.

The new tests use the limits of the control character ranges, CR, the printable characters that are adjacent to the limits, and characters that are not printable and are not ASCII. The new benchmarks use a long URL and a URL that contains '%xx' escape sequences.

Summary

Checklist

• I understand that this PR may be closed in case there was no previous discussion. (This doesn't apply to typos!)
• I've added a test for each change that was introduced, and I tried as much as possible to make a single atomic change.
• I've updated the documentation accordingly.

[codspeed-hq]

Merging this PR will improve performance by 31.78%

⚠️ Different runtime environments detected

Some benchmarks with significant performance changes were compared across different runtime environments,
which may affect the accuracy of the results.

Open the report in CodSpeed to investigate

⚡ 4 improved benchmarks
✅ 13 untouched benchmarks
🆕 2 new benchmarks
⏩ 7 skipped benchmarks¹

Performance Changes

	Mode	Benchmark	BASE	HEAD	Efficiency
⚡	Simulation	test_bench_url_parse	144.6 µs	85.7 µs	+68.65%
⚡	Simulation	test_bench_url_join	229.3 µs	158.4 µs	+44.74%
⚡	Simulation	test_bench_request_multipart	285.4 µs	256.6 µs	+11.2%
⚡	Simulation	test_bench_request_json_post	500 µs	450 µs	+11.11%
🆕	Simulation	test_bench_url_parse_long_query	N/A	119.1 µs	N/A
🆕	Simulation	test_bench_url_parse_percent_encoded	N/A	93.9 µs	N/A

Tip

Curious why this is faster? Comment @codspeedbot explain why this is faster on this PR, or directly use the CodSpeed MCP with your agent.

---

_{Comparing AyumuKasuga:perf-urlparse-hot-path (085f98f) with main (829b93a)}

Footnotes

1. 7 benchmarks were skipped, so the baseline results were used instead. If they were deleted from the codebase, click here and archive them to remove them from the performance reports. ↩

[cubic-dev-ai]

No issues found across 4 files

_{Re-trigger cubic}