I'd like a backtrace when my Windows application crashes. Boost stacktrace looks interesting, and I've played with it a bit in Linux; however, I'm getting a zero result from safe_dump_to()
in Windows!
I presume I've made a simple oversight, and I'm hoping someone can help.
I'm using the MSVS2017 compiler with ninja and CMake.
I've built Boost v1.67 built with these flags:
link=static address-model=64 variant=release threading=multi
I ran the tests in boost_1_67_0/libs/stacktrace/test/
to make sure I'd built the library correctly.
Here is my CMake file, note I used -DCMAKE_BUILD_TYPE=RelWithDebInfo
:
cmake_minimum_required(VERSION 3.0)
set( target_name stacktrace_test)
# boost
set(Boost_USE_STATIC_LIBS ON)
find_package( Boost REQUIRED COMPONENTS date_time filesystem thread
system stacktrace_windbg)
add_executable(${target_name} main.cpp)
set_property(TARGET ${target_name} PROPERTY CXX_STANDARD 14 )
target_compile_definitions(${target_name} PUBLIC
BOOST_ALL_NO_LIB=1 # disable pragama inclusion
BOOST_STACKTRACE_LINK=1
BOOST_STACKTRACE_USE_WINDBG=1
)
target_include_directories(${target_name} SYSTEM PUBLIC
${Boost_INCLUDE_DIRS}
)
target_link_libraries(${target_name}
${Boost_LIBRARIES} # boost
dbgeng
ole32
)
Here is my code, based on Boost stactrace's getting started page:
#include <iostream>
#include <signal.h> // ::signal, ::raise
#include <boost/stacktrace/stacktrace.hpp>
#include <boost/filesystem.hpp>
void handler(int signum)
{
::signal(signum, SIG_DFL);
size_t result = boost::stacktrace::safe_dump_to("./backtrace.dump");
std::cout << " " << result << std::endl;
::raise(SIGABRT);
}
void crash()
{
abort();
}
int main(int,char**)
{
::signal(SIGSEGV, &handler);
::signal(SIGABRT, &handler);
if(boost::filesystem::exists("backtrace.dump")) // existing stacktrace?
{
std::ifstream ifs("backtrace.dump");
boost::stacktrace::stacktrace st =
boost::stacktrace::stacktrace::from_dump(ifs);
std::cout << "Stacktrace from prior run:\n" << st << std::endl;
ifs.close(); // cleanup
boost::filesystem::remove("backtrace.dump");
return 0;
}
crash();
return 0;
}
And here is the output from my compilation:
C:\SRC\sandbox\boost\stacktrace\win-simple\Build_RWD>ninja -v
[1/2 50%]:C:\PROGRA~2\MICROS~1\2017\BUILDT~1\VC\Tools\MSVC\1413~1.261\bin\Hostx64\x64\cl.exe /nologo /TP -DBOOST_ALL_NO_LIB=1 -DBOOST_STACKTRACE_LINK=1 -DBOOST_STACKTRACE_USE_WINDBG=1 -IC:\3rd_libs\boost\boost_1_67_0 /DWIN32 /D_WINDOWS /W3 /GR /EHsc /MD /Zi /O2 /Ob1 /DNDEBUG -std:c++14 /showIncludes /FoCMakeFiles\stacktrace_test.dir\main.cpp.obj /FdCMakeFiles\stacktrace_test.dir\ /FS -c ..\main.cpp
[2/2 100%]:cmd.exe /C "cd . && "C:\Program Files\CMake\bin\cmake.exe" -E vs_link_exe --intdir=CMakeFiles\stacktrace_test.dir --manifests -- C:\PROGRA~2\MICROS~1\2017\BUILDT~1\VC\Tools\MSVC\1413~1.261\bin\Hostx64\x64\link.exe /nologo CMakeFiles\stacktrace_test.dir\main.cpp.obj /out:stacktrace_test.exe /implib:stacktrace_test.lib /pdb:stacktrace_test.pdb /version:0.0 /machine:x64 /debug /INCREMENTAL /subsystem:console C:\3rd_libs\boost\boost_1_67_0\stage\lib\libboost_date_time.lib C:\3rd_libs\boost\boost_1_67_0\stage\lib\libboost_filesystem.lib C:\3rd_libs\boost\boost_1_67_0\stage\lib\libboost_thread.lib C:\3rd_libs\boost\boost_1_67_0\stage\lib\libboost_system.lib C:\3rd_libs\boost\boost_1_67_0\stage\lib\libboost_stacktrace_windbg.lib C:\3rd_libs\boost\boost_1_67_0\stage\lib\libboost_chrono.lib C:\3rd_libs\boost\boost_1_67_0\stage\lib\libboost_atomic.lib dbgeng.lib ole32.lib kernel32.lib user32.lib gdi32.lib winspool.lib shell32.lib ole32.lib oleaut32.lib uuid.lib comdlg32.lib advapi32.lib && cd ."
LINK : stacktrace_test.exe not found or not built by the last incremental link; performing full link
I've reviewed the compilation flags and nothing jumps out at me.
Any help is appreciated! Thank you, SB
As it turns out,functionality in boost::stacktrace is disabled for windows as it's reported it may cause deadlocks. Re-enabling the code in stacktrace/detail/safe_dump_win.ipp
solved the problem for me.
Note that I took the github development version of safe_dump_win.ipp
, frame_msvc.ipp
, and frame_unwind.ipp
as my starting point.
In my limited testing I have not seen the deadlock. Should the software deadlock I will be in no worse shape than if it quietly crashed and, hopefully, I will get more traces than deadlocks!