milseman
commented
10 months ago This is built on top of https://github.com/apple/swift-experimental-string-processing/pull/706, which has many improvements.
This refactoring, compared to just https://github.com/apple/swift-experimental-string-processing/pull/706, has the following perf and regressions. They could be ARC-related, haven't explored exactly what's going on here:
=== Regressions ======================================================================
- CompilerMessages_All_Scalar 73.7ms 69.9ms 3.77ms 5.4%
- EmailRFCNoMatches_All_Scalar 126ms 122ms 3.74ms 3.1%
- CompilerMessages_All 90.1ms 86.7ms 3.42ms 3.9%
- SubtractionCCC_All 23ms 20.9ms 2.07ms 9.9%
- DiceRollsInText_All 42.9ms 41.1ms 1.82ms 4.4%
- BasicCCC_All 7.32ms 5.93ms 1.38ms 23.3%
- EmojiRegex_All_Scalar 41.9ms 40.5ms 1.37ms 3.4%
- BasicCCC_All_Scalar 7.29ms 5.93ms 1.36ms 22.9%
- AnchoredNotFound_All 14.8ms 13.5ms 1.33ms 9.8%
- DiceRollsInText_All_Scalar 39.9ms 38.6ms 1.31ms 3.4%
- BasicRangeCCC_All 7.44ms 6.38ms 1.07ms 16.7%
- EmailLookaheadNoMatches_All 24ms 23ms 984µs 4.3%
- InvertedCCC_All_Scalar 17.6ms 16.7ms 926µs 5.5%
- Numbers_All 6.73ms 5.8ms 923µs 15.9%
- CaseInsensitiveCCC_All 7.23ms 6.34ms 893µs 14.1%
- BasicRangeCCC_All_Scalar 7.29ms 6.4ms 884µs 13.8%
- CaseInsensitiveCCC_All_Scalar 7.22ms 6.35ms 871µs 13.7%
- EmailRFC_All 64.8ms 64ms 820µs 1.3%
- EmailLookaheadNoMatches_All_Scalar 21.4ms 20.7ms 735µs 3.6%
- NotFound_All_Scalar 6.82ms 6.18ms 643µs 10.4%
- NotFound_All 7.67ms 7.2ms 468µs 6.5%
- MACAddress 2.81ms 2.38ms 430µs 18.0%
- EmailRFC_All_Scalar 45.4ms 45ms 425µs 0.9%
- IntersectionCCC_All 22.2ms 21.8ms 415µs 1.9%
- SubtractionCCC_All_Scalar 21.4ms 21ms 384µs 1.8%
- AnchoredNotFound_First 9.13ms 8.78ms 358µs 4.1%
- Numbers_All_Scalar 5.66ms 5.34ms 319µs 6.0%
- IntersectionCCC_All_Scalar 22.2ms 21.9ms 318µs 1.5%
- Words_All 12.7ms 12.4ms 316µs 2.5%
- LiteralSearchNotFound_All 6.71ms 6.44ms 271µs 4.2%
- EmailLookahead_All 19.6ms 19.3ms 263µs 1.4%
- HangulSyllable_All 6.92ms 6.66ms 261µs 3.9%
- MACAddress_Scalar 2.49ms 2.25ms 232µs 10.3%
- IPv6Address 2.78ms 2.56ms 223µs 8.7%
- IPv6Address_Scalar 2.61ms 2.39ms 216µs 9.0%
- EmailLookahead_All_Scalar 17.3ms 17.1ms 205µs 1.2%
- LiteralSearch_All 6.89ms 6.7ms 195µs 2.9%
- IPv4Address_Scalar 2.24ms 2.06ms 186µs 9.0%
- HangulSyllable_All_Scalar 6.11ms 5.92ms 185µs 3.1%
- LiteralSearchNotFound_All_Scalar 5.68ms 5.51ms 172µs 3.1%
- GraphemeBreakNoCap_All_Scalar 2.99ms 2.83ms 161µs 5.7%
- GraphemeBreakNoCap_All 3.1ms 2.95ms 149µs 5.0%
- EmailBuiltinCharacterClass_All_Scalar 10.3ms 10.2ms 147µs 1.4%
- ReluctantQuantWithTerminal_Whole 5.41ms 5.27ms 147µs 2.8%
- LiteralSearch_All_Scalar 5.87ms 5.73ms 145µs 2.5%
- ReluctantQuantWithTerminal_Whole_Scalar 5.34ms 5.21ms 123µs 2.4%
- Css_All_Scalar 2.99ms 2.88ms 117µs 4.1%
- IPv4Address 2.32ms 2.22ms 105µs 4.7%
- EagarQuantWithTerminal_Whole_Scalar 520µs 441µs 79.6µs 18.1%
- Css_All 3.31ms 3.23ms 76.2µs 2.4%
- DiceNotation 4.58ms 4.51ms 75.1µs 1.7%
- EagarQuantWithTerminal_Whole 520µs 451µs 69.2µs 15.3%
- DiceNotation_Scalar 4.33ms 4.26ms 65.5µs 1.5%
- Lines_All_Scalar 1.73ms 1.67ms 64µs 3.8%
- Lines_All 1.77ms 1.72ms 53.1µs 3.1%
=== Improvements =====================================================================
- EmailLookaheadList_Scalar 3.59ms 3.96ms -370µs -9.3%
- EmailBuiltinCharacterClass_All 10.5ms 10.7ms -166µs -1.6%
- AnchoredNotFound_First_Scalar 5.47ms 5.62ms -153µs -2.7%
- EmailLookaheadList 3.81ms 3.87ms -54.8µs -1.4%
Refactor off of mutating methods
Refactors mutating methods into string methods for easier unit testing and parity-checking via assertions. Prepares for more efficient implementations.
Doing so creates many small-to-medium regressions, unfortunately, so this should only be done in conjunction with more refactorings and improvements.