<regex>: Speed up skip optimization for default regex_traits in collate mode (#5672)

Co-authored-by: Stephan T. Lavavej <[email protected]>

Author: muellerj2

benchmarks/src/regex_search.cpp

@@ -8,13 +8,14 @@
 #include "lorem.hpp"
 
 using namespace std;
+using namespace regex_constants;
 
-void bm_lorem_search(benchmark::State& state, const char* pattern) {
+void bm_lorem_search(benchmark::State& state, const char* pattern, syntax_option_type syntax = ECMAScript) {
     string repeated_lorem{lorem_ipsum};
     for (long long i = 0; i < state.range(); ++i) {
         repeated_lorem += repeated_lorem;
     }
-    regex re{pattern};
+    regex re{pattern, syntax};
 
     for (auto _ : state) {
         benchmark::DoNotOptimize(repeated_lorem);
@@ -33,6 +34,7 @@ void bm_lorem_search(benchmark::State& state, const char* pattern) {
 
 BENCHMARK_CAPTURE(bm_lorem_search, "^bibe", "^bibe")->Arg(2)->Arg(3)->Arg(4);
 BENCHMARK_CAPTURE(bm_lorem_search, "bibe", "bibe")->Arg(2)->Arg(3)->Arg(4);
+BENCHMARK_CAPTURE(bm_lorem_search, "bibe".collate, "bibe", regex_constants::collate)->Arg(2)->Arg(3)->Arg(4);
 BENCHMARK_CAPTURE(bm_lorem_search, "(bibe)", "(bibe)")->Arg(2)->Arg(3)->Arg(4);
 BENCHMARK_CAPTURE(bm_lorem_search, "(bibe)+", "(bibe)+")->Arg(2)->Arg(3)->Arg(4);
 BENCHMARK_CAPTURE(bm_lorem_search, "(?:bibe)+", "(?:bibe)+")->Arg(2)->Arg(3)->Arg(4);

stl/inc/regex

@@ -3690,6 +3690,8 @@ _BidIt1 _Compare_translate_both(_BidIt1 _Begin1, _BidIt1 _End1, _BidIt2 _Begin2,
     // compare character ranges, translating characters in both ranges according to syntax options
     if (_Sflags & regex_constants::icase) {
         return _STD _Cmp_chrange(_Begin1, _End1, _Begin2, _End2, _Cmp_icase<_RxTraits>{_Traits});
+    } else if constexpr (_Is_any_of_v<_RxTraits, regex_traits<char>, regex_traits<wchar_t>>) {
+        return _STD _Cmp_chrange(_Begin1, _End1, _Begin2, _End2, equal_to<typename _RxTraits::char_type>{});
     } else if (_Sflags & regex_constants::collate) {
         return _STD _Cmp_chrange(_Begin1, _End1, _Begin2, _End2, _Cmp_collate<_RxTraits>{_Traits});
     } else {
@@ -3703,6 +3705,8 @@ _BidIt1 _Compare_translate_left(_BidIt1 _Begin1, _BidIt1 _End1, _BidIt2 _Begin2,
     // compare character ranges, translating characters in the left range according to syntax options
     if (_Sflags & regex_constants::icase) {
         return _STD _Cmp_chrange(_Begin1, _End1, _Begin2, _End2, _Cmp_icase_translateleft<_RxTraits>{_Traits});
+    } else if constexpr (_Is_any_of_v<_RxTraits, regex_traits<char>, regex_traits<wchar_t>>) {
+        return _STD _Cmp_chrange(_Begin1, _End1, _Begin2, _End2, equal_to<typename _RxTraits::char_type>{});
     } else if (_Sflags & regex_constants::collate) {
         return _STD _Cmp_chrange(_Begin1, _End1, _Begin2, _End2, _Cmp_collate_translateleft<_RxTraits>{_Traits});
     } else {
@@ -3717,6 +3721,8 @@ _BidIt1 _Search_translate_left(_BidIt1 _Begin1, _BidIt1 _End1, _BidIt2 _Begin2,
     // after translating characters in the left sequence according to syntax options
     if (_Sflags & regex_constants::icase) {
         return _STD search(_Begin1, _End1, _Begin2, _End2, _Cmp_icase_translateleft<_RxTraits>{_Traits});
+    } else if constexpr (_Is_any_of_v<_RxTraits, regex_traits<char>, regex_traits<wchar_t>>) {
+        return _STD search(_Begin1, _End1, _Begin2, _End2, equal_to<typename _RxTraits::char_type>{});
     } else if (_Sflags & regex_constants::collate) {
         return _STD search(_Begin1, _End1, _Begin2, _End2, _Cmp_collate_translateleft<_RxTraits>{_Traits});
     } else {

tests/std/tests/VSO_0000000_regex_use/test.cpp

@@ -2103,6 +2103,32 @@ void test_gh_5576() {
         match_default, "AbGweEXfFlXlLLlffflEXlF");
 }
 
+void test_gh_5672() {
+    // GH-5672: Speed up skip optimization for default `regex_traits` in `collate` mode
+    // The PR added a faster branch in the skip optimization when matching in collate mode
+    // for default `regex_traits<char>` and `regex_traits<wchar_t>`.
+    // The following tests check that searching still works correctly when the faster branch is engaged.
+    {
+        test_regex collating_re(&g_regexTester, "g", regex_constants::collate);
+
+        collating_re.should_search_match("abcdefghijklmnopqrstuvwxyz", "g");
+        collating_re.should_search_fail("ABCDEFGHIJKLMNOPQRSTUVWXYZ");
+        collating_re.should_search_match("zyxwvutsrqponmlkjihgfedcba", "g");
+        collating_re.should_search_fail("ZYXWVUTSRQPONMLKJIHGFEDCBA");
+        collating_re.should_search_fail("zyxwvutsrqponmlkjihedcba");
+    }
+
+    {
+        test_wregex collating_re(&g_regexTester, L"g", regex_constants::collate);
+
+        collating_re.should_search_match(L"abcdefghijklmnopqrstuvwxyz", L"g");
+        collating_re.should_search_fail(L"ABCDEFGHIJKLMNOPQRSTUVWXYZ");
+        collating_re.should_search_match(L"zyxwvutsrqponmlkjihgfedcba", L"g");
+        collating_re.should_search_fail(L"ZYXWVUTSRQPONMLKJIHGFEDCBA");
+        collating_re.should_search_fail(L"zyxwvutsrqponmlkjihedcba");
+    }
+}
+
 int main() {
     test_dev10_449367_case_insensitivity_should_work();
     test_dev11_462743_regex_collate_should_not_disable_regex_icase();
@@ -2153,6 +2179,7 @@ int main() {
     test_gh_5490();
     test_gh_5509();
     test_gh_5576();
+    test_gh_5672();
 
     return g_regexTester.result();
 }