bryancall commented on code in PR #13670: URL: https://github.com/apache/trafficserver/pull/13670#discussion_r3997679512
########## tools/benchmark/benchmark_Regex.cc: ########## @@ -0,0 +1,686 @@ +/** @file + + Benchmarks for the tsutil Regex wrapper: time per operation and the number and size + of heap allocations each operation makes. + + The allocation half matters as much as the timing half. PCRE2 routes every allocation + it makes for a compile or a match through the callbacks the wrapper installs, and those + call the system allocator, so counting calls to malloc across a region counts exactly + what the wrapper caused. Under the just-in-time engine a match should reach the system + allocator zero times, because the match data comes out of the caller's own buffer. The + interpreter is the exception: it allocates a backtracking frames vector through the same + allocator, so a match that runs interpreted does show up in the count. + + Interposing malloc is only wired up on Linux, where defining these symbols in the + executable is enough. Elsewhere the counters stay at zero and the report says so, so a + run on another platform still gives timings without quietly reporting zero allocations + as a result. + + @section license License + + Licensed to the Apache Software Foundation (ASF) under one + or more contributor license agreements. See the NOTICE file + distributed with this work for additional information + regarding copyright ownership. The ASF licenses this file + to you under the Apache License, Version 2.0 (the + "License"); you may not use this file except in compliance + with the License. You may obtain a copy of the License at + + http://www.apache.org/licenses/LICENSE-2.0 + + Unless required by applicable law or agreed to in writing, software + distributed under the License is distributed on an "AS IS" BASIS, + WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + See the License for the specific language governing permissions and + limitations under the License. + */ + +#include <cstddef> +#include <cstdint> +#include <cstdio> +#include <cstdlib> +#include <cstring> +#include <string> +#include <string_view> +#include <vector> + +#define CATCH_CONFIG_ENABLE_BENCHMARKING +#include <catch2/catch_test_macros.hpp> +#include <catch2/benchmark/catch_benchmark.hpp> + +#include "tsutil/Regex.h" + +#define PCRE2_CODE_UNIT_WIDTH 8 +#include <pcre2.h> + +// --------------------------------------------------------------------------- +// Allocation counting +// --------------------------------------------------------------------------- + +namespace +{ +struct AllocStats { + unsigned long calls = 0; + unsigned long bytes = 0; +}; + +// Counting is per thread so a benchmark that spawns threads does not race the counters. +// These benchmarks are single threaded; the qualifier is here so the numbers stay honest +// if one is added later. +thread_local AllocStats alloc_stats; +thread_local bool alloc_counting = false; + +class CountAllocations +{ +public: + CountAllocations() + { + alloc_stats = AllocStats{}; + alloc_counting = true; + } + ~CountAllocations() { alloc_counting = false; } + + AllocStats + stats() const + { + return alloc_stats; + } +}; + +#if defined(__linux__) +constexpr bool ALLOC_COUNTING_AVAILABLE = true; +#else +constexpr bool ALLOC_COUNTING_AVAILABLE = false; +#endif + +} // namespace + +#if defined(__linux__) +#include <dlfcn.h> + +// Interpose the system allocator. Defining these in the executable takes precedence over +// libc for every caller in the process, which is what makes the count cover PCRE2's own +// allocations as well as the wrapper's. +namespace +{ +using malloc_fn = void *(*)(size_t); +using free_fn = void (*)(void *); +using calloc_fn = void *(*)(size_t, size_t); +using realloc_fn = void *(*)(void *, size_t); + +malloc_fn real_malloc = nullptr; +free_fn real_free = nullptr; +calloc_fn real_calloc = nullptr; +realloc_fn real_realloc = nullptr; + +// dlsym() itself can allocate while the real pointers are still being resolved. Hand +// those few allocations out of a static buffer rather than recursing. +// +// Each block is preceded by a header holding its size, so a realloc of one can copy the +// old contents rather than silently returning uninitialised storage. The header is one +// max_align_t wide so the pointer handed back keeps the alignment malloc promises. +constexpr size_t BOOTSTRAP_HEADER = alignof(std::max_align_t); +static_assert(BOOTSTRAP_HEADER >= sizeof(size_t), "the bootstrap header must hold a size"); + +alignas(std::max_align_t) char bootstrap_buffer[16384]; +size_t bootstrap_used = 0; +bool resolving = false; + +bool +from_bootstrap(void *p) +{ + return p >= static_cast<void *>(bootstrap_buffer) && p < static_cast<void *>(bootstrap_buffer + sizeof(bootstrap_buffer)); +} Review Comment: Fixed in 58dcc8857f. from_bootstrap() now compares addresses through uintptr_t rather than ordering pointers into unrelated objects. Agreed a wrong answer there is not harmless: a false positive makes free() leak the block and makes realloc() read a header that was never written. _🤖 Addressed by [Claude Code](https://claude.com/claude-code)_ ########## tools/benchmark/CMakeLists.txt: ########## @@ -57,3 +57,11 @@ target_include_directories(benchmark_HuffmanDecode PRIVATE ${CMAKE_SOURCE_DIR}/l add_executable(benchmark_ascii_tolower benchmark_ascii_tolower.cc) target_link_libraries(benchmark_ascii_tolower PRIVATE Catch2::Catch2WithMain ts::tscore) + +add_executable(benchmark_Regex benchmark_Regex.cc) +target_link_libraries(benchmark_Regex PRIVATE Catch2::Catch2WithMain ts::tsutil) +if(CMAKE_SYSTEM_NAME STREQUAL "Linux") + # The allocation counters interpose the system allocator, which needs the real + # symbols from the next object in the search order. + target_link_libraries(benchmark_Regex PRIVATE ${CMAKE_DL_LIBS}) Review Comment: Made the change in 58dcc8857f, but the stated reason does not hold and I would rather be accurate about why. The target builds today on GCC 10.2.1 in the CentOS 7 CI container, GCC 16.2.1, and clang 22.1.8, all on Linux with RTLD_NEXT and no _GNU_SOURCE of its own, because g++ and clang++ define _GNU_SOURCE for C++ on Linux themselves. It does not fail to compile. It is still worth stating: the file needs a GNU extension and the definition was arriving implicitly from the compiler. The target now defines it. _🤖 Addressed by [Claude Code](https://claude.com/claude-code)_ -- This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. To unsubscribe, e-mail: [email protected] For queries about this service, please contact Infrastructure at: [email protected]
