<regex>: Improve search performance for regexes with initial + quantifiers (#5509)

muellerj2 · StephanTLavavej · web-flow · commit 2391e5eb104b · 2025-05-16T22:26:10.000-07:00
Co-authored-by: Stephan T. Lavavej &lt;stl@microsoft.com&gt;
diff --git a/benchmarks/CMakeLists.txt b/benchmarks/CMakeLists.txt
@@ -116,6 +116,7 @@ add_benchmark(nth_element src/nth_element.cpp)
 add_benchmark(path_lexically_normal src/path_lexically_normal.cpp)
 add_benchmark(priority_queue_push_range src/priority_queue_push_range.cpp)
 add_benchmark(random_integer_generation src/random_integer_generation.cpp)
+add_benchmark(regex_search src/regex_search.cpp)
 add_benchmark(remove src/remove.cpp)
 add_benchmark(replace src/replace.cpp)
 add_benchmark(reverse src/reverse.cpp)
diff --git a/benchmarks/src/regex_search.cpp b/benchmarks/src/regex_search.cpp
@@ -0,0 +1,38 @@
+// Copyright (c) Microsoft Corporation.
+// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
+
+#include <benchmark/benchmark.h>
+#include <regex>
+#include <string>
+
+#include "lorem.hpp"
+
+using namespace std;
+
+void bm_lorem_search(benchmark::State& state, const char* pattern) {
+    string repeated_lorem{lorem_ipsum};
+    for (long long i = 0; i < state.range(); ++i) {
+        repeated_lorem += repeated_lorem;
+    }
+    regex re{pattern};
+
+    for (auto _ : state) {
+        benchmark::DoNotOptimize(repeated_lorem);
+        const char* pos = repeated_lorem.data();
+        const char* end = repeated_lorem.data() + repeated_lorem.size();
+        cmatch match;
+        for (; regex_search(pos, end, match, re); ++pos) {
+            benchmark::DoNotOptimize(match);
+            pos = match[0].second;
+            if (pos == end) {
+                break;
+            }
+        }
+    }
+}
+
+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);
+
+BENCHMARK_MAIN();
diff --git a/stl/inc/regex b/stl/inc/regex
@@ -4128,6 +4128,15 @@ _BidIt _Matcher2<_BidIt, _Elem, _RxTraits, _It, _Alloc>::_Skip(_BidIt _First_arg
                 break;
             }
 
+        case _N_rep:
+            {
+                _Node_rep* _Node = static_cast<_Node_rep*>(_Nx);
+                if (_Node->_Min == 0) {
+                    return _First_arg;
+                }
+                break;
+            }
+
         case _N_begin:
             break;
 
@@ -4139,7 +4148,6 @@ _BidIt _Matcher2<_BidIt, _Elem, _RxTraits, _It, _Alloc>::_Skip(_BidIt _First_arg
         case _N_neg_assert:
         case _N_back:
         case _N_endif:
-        case _N_rep:
         case _N_end_rep:
         default:
             return _First_arg;
diff --git a/tests/std/include/test_regex_support.hpp b/tests/std/include/test_regex_support.hpp
@@ -282,8 +282,9 @@ class test_regex {
                             submatches_success = false;
                             break;
                         }
-                    } else if (!actual_capture.matched || actual_capture.first != (mr[0].first + expected_capture.first)
-                               || actual_capture.second != (mr[0].first + expected_capture.second)) {
+                    } else if (!actual_capture.matched
+                               || actual_capture.first != (subject.begin() + expected_capture.first)
+                               || actual_capture.second != (subject.begin() + expected_capture.second)) {
                         submatches_success = false;
                         break;
                     }
@@ -297,7 +298,8 @@ class test_regex {
                     for (const auto& expected_capture : capture_groups) {
                         std::string capture = "(unmatched)";
                         if (expected_capture.first != -1) {
-                            capture.assign(mr[0].first + expected_capture.first, mr[0].first + expected_capture.second);
+                            capture.assign(
+                                subject.begin() + expected_capture.first, subject.begin() + expected_capture.second);
                         }
                         printf(R"(%s"%s" [%td %td])", initial ? "" : ", ", capture.c_str(), expected_capture.first,
                             expected_capture.second);
@@ -313,8 +315,8 @@ class test_regex {
                         std::ptrdiff_t last        = -1;
                         if (actual_capture.matched) {
                             capture = actual_capture.str();
-                            first   = actual_capture.first - mr[0].first;
-                            last    = actual_capture.second - mr[0].first;
+                            first   = actual_capture.first - subject.begin();
+                            last    = actual_capture.second - subject.begin();
                         }
                         printf(R"(%s"%s" [%td %td])", initial ? "" : ", ", capture.c_str(), first, last);
                         initial = false;
diff --git a/tests/std/tests/VSO_0000000_regex_use/test.cpp b/tests/std/tests/VSO_0000000_regex_use/test.cpp
@@ -1889,6 +1889,54 @@ void test_gh_5490() {
     }
 }
 
+void test_gh_5509() {
+    // GH-5509 extended the matcher's skip optimization
+    // to regexes starting with a loop with at least one repetition,
+    // speeding up searches for such regexes
+
+    {
+        test_regex char_plus_regex(&g_regexTester, "(a+)");
+        char_plus_regex.should_search_match_capture_groups("blwerofaaweraf", "aa", match_default, {{7, 9}});
+        char_plus_regex.should_search_fail("blwerofwerf");
+    }
+
+    {
+        test_regex charclass_plus_regex(&g_regexTester, "([fa]+)");
+        charclass_plus_regex.should_search_match_capture_groups("blwerofaaweraf", "faa", match_default, {{6, 9}});
+        charclass_plus_regex.should_search_fail("blwerower");
+    }
+
+    {
+        test_regex string_plus_regex(&g_regexTester, "((?:aw)+)");
+        string_plus_regex.should_search_match_capture_groups("blwerofaawaweraf", "awaw", match_default, {{8, 12}});
+        string_plus_regex.should_search_fail("blwerofaerwaf");
+    }
+
+    {
+        test_regex anchored_string_plus_regex(&g_regexTester, "((?:^aw)+)");
+        anchored_string_plus_regex.should_search_match_capture_groups(
+            "blwerofa\nawaweraf", "aw", match_default, {{9, 11}});
+        anchored_string_plus_regex.should_search_fail("blwerof\naerwaf");
+    }
+
+    {
+        test_regex anchored_string_plus_regex(&g_regexTester, "((?:$\naw)+)");
+        anchored_string_plus_regex.should_search_match_capture_groups(
+            "blwerofa\nawaweraf", "\naw", match_default, {{8, 11}});
+        anchored_string_plus_regex.should_search_fail("blwerof\naerwaf");
+    }
+
+    {
+        test_regex string_star_string_regex(&g_regexTester, "((?:aw)*fa)");
+        string_star_string_regex.should_search_match_capture_groups(
+            "blwerofaawawfaeraf", "fa", match_default, {{6, 8}});
+        string_star_string_regex.should_search_match_capture_groups(
+            "blweroawawfaeraf", "awawfa", match_default, {{6, 12}});
+        string_star_string_regex.should_search_match("blwerofaerwaf", "fa");
+        string_star_string_regex.should_search_fail("blweroerwaf");
+    }
+}
+
 int main() {
     test_dev10_449367_case_insensitivity_should_work();
     test_dev11_462743_regex_collate_should_not_disable_regex_icase();
@@ -1936,6 +1984,7 @@ int main() {
     test_gh_5374();
     test_gh_5377();
     test_gh_5490();
+    test_gh_5509();
 
     return g_regexTester.result();
 }
