Home

Awesome

Cpptrace <!-- omit in toc -->

build test Quality Gate Status <br/> Community Discord Link <br/> Try on Compiler Explorer

Cpptrace is a simple and portable C++ stacktrace library supporting C++11 and greater on Linux, macOS, and Windows including MinGW and Cygwin environments. The goal: Make stack traces simple for once.

Cpptrace also has a C API, docs here.

Table of Contents <!-- omit in toc -->

30-Second Overview

Generating stack traces is as easy as:

#include <cpptrace/cpptrace.hpp>

void trace() {
    cpptrace::generate_trace().print();
}

Demo

Cpptrace can also retrieve function inlining information on optimized release builds:

Inlining

Cpptrace provides access to resolved stack traces as well as lightweight raw traces (just addresses) that can be resolved later:

const auto raw_trace = cpptrace::generate_raw_trace();
// then later
raw_trace.resolve().print();

Cpptrace provides a way to produce stack traces on arbitrary exceptions. More information on this system below.

#include <cpptrace/from_current.hpp>
void foo() {
    throw std::runtime_error("foo failed");
}
int main() {
    CPPTRACE_TRY {
        foo();
    } CPPTRACE_CATCH(const std::exception& e) {
        std::cerr<<"Exception: "<<e.what()<<std::endl;
        cpptrace::from_current_exception().print();
    }
}

from_current

There are a few extraneous frames at the top of the stack corresponding to internals of exception handling in the standard library. These are a small price to pay for stack traces on all exceptions.

Cpptrace also provides a handful of traced exception objects that store stack traces when thrown. This is useful when the exceptions might not be caught by CPPTRACE_CATCH:

#include <cpptrace/cpptrace.hpp>

void trace() {
    throw cpptrace::logic_error("This wasn't supposed to happen!");
}

Exception

Additional notable features:

Snippets

CMake FetchContent Usage

include(FetchContent)
FetchContent_Declare(
  cpptrace
  GIT_REPOSITORY https://github.com/jeremy-rifkin/cpptrace.git
  GIT_TAG        v0.7.4 # <HASH or TAG>
)
FetchContent_MakeAvailable(cpptrace)
target_link_libraries(your_target cpptrace::cpptrace)

# Needed for shared library builds on windows:  copy cpptrace.dll to the same directory as the
# executable for your_target
if(WIN32)
  add_custom_command(
    TARGET your_target POST_BUILD
    COMMAND ${CMAKE_COMMAND} -E copy_if_different
    $<TARGET_FILE:cpptrace::cpptrace>
    $<TARGET_FILE_DIR:your_target>
  )
endif()

Be sure to configure with -DCMAKE_BUILD_TYPE=Debug or -DCMAKE_BUILD_TYPE=RelWithDebInfo for symbols and line information.

On macOS it is recommended to generate a .dSYM file, see Platform Logistics below.

For other ways to use the library, such as through package managers, a system-wide installation, or on a platform without internet access see How to Include The Library below.

FAQ

What about C++23 <stacktrace>?

Some day C++23's <stacktrace> will be ubiquitous. And maybe one day the msvc implementation will be acceptable. The original motivation for cpptrace was to support projects using older C++ standards and as the library has grown its functionality has extended beyond the standard library's implementation.

Cpptrace provides functionality beyond what the standard library provides and what implementations provide, such as:

What does cpptrace have over other C++ stacktrace libraries?

Other C++ stacktrace libraries, such as boost stacktrace and backward-cpp, fall short when it comes to portability and ease of use. In testing, I found neither to provide adaquate coverage of various environments. Even when they can be made to work in an environment they require manual configuration from the end-user, possibly requiring manual installation of third-party dependencies. This is a highly undesirable burden to impose on users, especially when it is for a software package which just provides diagnostics as opposed to core functionality. Additionally, cpptrace provides support for resolving inlined calls by default for DWARF symbols (boost does not do this, backward-cpp can do this but only for some back-ends), better support for resolving full function signatures, and nicer API, among other features.

Prerequisites

[!IMPORTANT] Debug info (-g//Z7//Zi//DEBUG/-DBUILD_TYPE=Debug/-DBUILD_TYPE=RelWithDebInfo) is required for complete trace information.

Basic Usage

cpptrace::generate_trace() can be used to generate a stacktrace object at the current call site. Resolved frames can be accessed from this object with .frames and the trace can be printed with .print(). Cpptrace also provides a method to get light-weight raw traces with cpptrace::generate_raw_trace(), which are just vectors of program counters, which can be resolved at a later time.

namespace cpptrace

All functions are thread-safe unless otherwise noted.

Stack Traces

The core resolved stack trace object. Generate a trace with cpptrace::generate_trace() or cpptrace::stacktrace::current(). On top of a set of helper functions struct stacktrace allows direct access to frames as well as iterators.

cpptrace::stacktrace::print can be used to print a stacktrace. cpptrace::stacktrace::print_with_snippets can be used to print a stack trace with source code snippets.

namespace cpptrace {
    // Some type sufficient for an instruction pointer, currently always an alias to std::uintptr_t
    using frame_ptr = std::uintptr_t;

    struct stacktrace_frame {
        frame_ptr raw_address; // address in memory
        frame_ptr object_address; // address in the object file
        // nullable<T> represents a nullable integer. More docs later.
        nullable<std::uint32_t> line;
        nullable<std::uint32_t> column;
        std::string filename;
        std::string symbol;
        bool is_inline;
        bool operator==(const stacktrace_frame& other) const;
        bool operator!=(const stacktrace_frame& other) const;
        object_frame get_object_info() const; // object_address is stored but if the object_path is needed this can be used
        std::string to_string() const;
        /* operator<<(ostream, ..) and std::format support exist for this object */
    };

    struct stacktrace {
        std::vector<stacktrace_frame> frames;
        // here as a drop-in for std::stacktrace
        static stacktrace current(std::size_t skip = 0);
        static stacktrace current(std::size_t skip, std::size_t max_depth);
        void print() const;
        void print(std::ostream& stream) const;
        void print(std::ostream& stream, bool color) const;
        void print_with_snippets() const;
        void print_with_snippets(std::ostream& stream) const;
        void print_with_snippets(std::ostream& stream, bool color) const;
        std::string to_string(bool color = false) const;
        void clear();
        bool empty() const noexcept;
        /* operator<<(ostream, ..), std::format support, and iterators exist for this object */
    };

    stacktrace generate_trace(std::size_t skip = 0);
    stacktrace generate_trace(std::size_t skip, std::size_t max_depth);
}

Object Traces

Object traces contain the most basic information needed to construct a stack trace outside the currently running executable. It contains the raw address, the address in the binary (ASLR and the object file's memory space and whatnot is resolved), and the path to the object the instruction pointer is located in.

namespace cpptrace {
    struct object_frame {
        std::string object_path;
        frame_ptr raw_address;
        frame_ptr object_address;
    };

    struct object_trace {
        std::vector<object_frame> frames;
        static object_trace current(std::size_t skip = 0);
        static object_trace current(std::size_t skip, std::size_t max_depth);
        stacktrace resolve() const;
        void clear();
        bool empty() const noexcept;
        /* iterators exist for this object */
    };

    object_trace generate_object_trace(std::size_t skip = 0);
    object_trace generate_object_trace(std::size_t skip, std::size_t max_depth);
}

Raw Traces

Raw trace access: A vector of program counters. These are ideal for fast and cheap traces you want to resolve later.

Note it is important executables and shared libraries in memory aren't somehow unmapped otherwise libdl calls (and GetModuleFileName in windows) will fail to figure out where the program counter corresponds to.

namespace cpptrace {
    struct raw_trace {
        std::vector<frame_ptr> frames;
        static raw_trace current(std::size_t skip = 0);
        static raw_trace current(std::size_t skip, std::size_t max_depth);
        object_trace resolve_object_trace() const;
        stacktrace resolve() const;
        void clear();
        bool empty() const noexcept;
        /* iterators exist for this object */
    };

    raw_trace generate_raw_trace(std::size_t skip = 0);
    raw_trace generate_raw_trace(std::size_t skip, std::size_t max_depth);
}

Utilities

cpptrace::demangle provides a helper function for name demangling, since it has to implement that helper internally anyways.

cpptrace::get_snippet gets a text snippet, if possible, from for the given source file for +/- context_size lines around line.

cpptrace::isatty and the fileno definitions are useful for deciding whether to use color when printing stack traces.

cpptrace::register_terminate_handler() is a helper function to set a custom std::terminate handler that prints a stack trace from a cpptrace exception (more info below) and otherwise behaves like the normal terminate handler.

namespace cpptrace {
    std::string demangle(const std::string& name);
    std::string get_snippet(
        const std::string& path,
        std::size_t line,
        std::size_t context_size,
        bool color = false
    );
    bool isatty(int fd);

    extern const int stdin_fileno;
    extern const int stderr_fileno;
    extern const int stdout_fileno;

    void register_terminate_handler();
}

Configuration

cpptrace::absorb_trace_exceptions: Configure whether the library silently absorbs internal exceptions and continues. Default is true.

cpptrace::enable_inlined_call_resolution: Configure whether the library will attempt to resolve inlined call information for release builds. Default is true.

cpptrace::experimental::set_cache_mode: Control time-memory tradeoffs within the library. By default speed is prioritized. If using this function, set the cache mode at the very start of your program before any traces are performed.

namespace cpptrace {
    void absorb_trace_exceptions(bool absorb);
    void enable_inlined_call_resolution(bool enable);

    enum class cache_mode {
        // Only minimal lookup tables
        prioritize_memory,
        // Build lookup tables but don't keep them around between trace calls
        hybrid,
        // Build lookup tables as needed
        prioritize_speed
    };

    namespace experimental {
        void set_cache_mode(cache_mode mode);
    }
}

Traces From All Exceptions

Cpptrace provides CPPTRACE_TRY and CPPTRACE_CATCH macros that allow a stack trace to be collected from the current thrown exception object, with minimal or no overhead in the non-throwing path:

#include <cpptrace/from_current.hpp>
void foo() {
    throw std::runtime_error("foo failed");
}
int main() {
    CPPTRACE_TRY {
        foo();
    } CPPTRACE_CATCH(const std::exception& e) {
        std::cerr<<"Exception: "<<e.what()<<std::endl;
        cpptrace::from_current_exception().print();
    }
}

This functionality is entirely opt-in, to access this use #include <cpptrace/from_current.hpp>.

Any declarator catch accepts works with CPPTRACE_CATCH, including .... This works with any thrown object, not just std::exceptions, it even works with throw 0;

from_current

There are a few extraneous frames at the top of the stack corresponding to standard library exception handling internals. These are a small price to pay for stack traces on all exceptions.

API functions:

There is a performance tradeoff with this functionality: Either the try-block can be zero overhead in the non-throwing path with potential expense in the throwing path, or the try-block can have very minimal overhead in the non-throwing path due to bookkeeping with guarantees about the expense of the throwing path. More details on this tradeoff below. Cpptrace provides macros for both sides of this tradeoff:

Note: It's important to not mix the Z variants with the non-Z variants.

Unfortunately the try/catch macros are needed to insert some magic to perform a trace during the unwinding search phase. In order to have multiple catch alternatives, either CPPTRACE_CATCH_ALT or a normal catch must be used:

CPPTRACE_TRY {
    foo();
} CPPTRACE_CATCH(const std::exception&) { // <- First catch must be CPPTRACE_CATCH
    // ...
} CPPTRACE_CATCH_ALT(const std::exception&) { // <- Ok
    // ...
} catch(const std::exception&) { // <- Also Ok
    // ...
} CPPTRACE_CATCH(const std::exception&) { // <- Not Ok
    // ...
}

Note: The current exception is the exception most recently seen by a cpptrace try-catch macro block.

CPPTRACE_TRY {
    throw std::runtime_error("foo");
} CPPTRACE_CATCH(const std::exception& e) {
    cpptrace::from_current_exception().print(); // the trace for std::runtime_error("foo")
    CPPTRACE_TRY {
        throw std::runtime_error("bar");
    } CPPTRACE_CATCH(const std::exception& e) {
        cpptrace::from_current_exception().print(); // the trace for std::runtime_error("bar")
    }
    cpptrace::from_current_exception().print(); // the trace for std::runtime_error("bar"), again
}

Removing the CPPTRACE_ prefix

CPPTRACE_TRY is a little cumbersome to type. To remove the CPPTRACE_ prefix you can use the CPPTRACE_UNPREFIXED_TRY_CATCH cmake option or the CPPTRACE_UNPREFIXED_TRY_CATCH preprocessor definition:

TRY {
    foo();
} CATCH(const std::exception& e) {
    std::cerr<<"Exception: "<<e.what()<<std::endl;
    cpptrace::from_current_exception().print();
}

This is not done by default for macro safety/hygiene reasons. If you do not want TRY/CATCH macros defined, as they are common macro names, you can easily modify the following snippet to provide your own aliases:

#define TRY CPPTRACE_TRY
#define CATCH(param) CPPTRACE_CATCH(param)
#define TRYZ CPPTRACE_TRYZ
#define CATCHZ(param) CPPTRACE_CATCHZ(param)
#define CATCH_ALT(param) CPPTRACE_CATCH_ALT(param)

How it works

C++ does not provide any language support for collecting stack traces when exceptions are thrown, however, exception handling under both the Itanium ABI and by SEH (used to implement C++ exceptions on windows) involves unwinding the stack twice. The first unwind searches for an appropriate catch handler, the second actually unwinds the stack and calls destructors. Since the stack remains intact during the search phase it's possible to collect a stack trace with little to no overhead when the catch is considered for matching the exception. The try/catch macros for cpptrace set up a special try/catch system that can collect a stack trace when considered during a search phase.

N.b.: This mechanism is also discussed in P2490R3.

Performance

The fundamental mechanism for this functionality is generating a trace when a catch block is considered during an exception handler search phase. Internally a lightweight raw trace is generated upon consideration, which is quite fast. This raw trace is only resolved when cpptrace::from_current_exception is called, or when the user manually resolves a trace from cpptrace::raw_trace_from_current_exception.

It's tricky, however, from the library's standpoint to check if the catch will end up matching. The library could simply generate a trace every time a CPPTRACE_CATCH is considered, however, in a deep nesting of catch's, e.g. as a result of recusion, where a matching handler is not found quickly this could introduce a non-trivial cost in the throwing pat due to tracing the stack multiple times. Thus, there is a performance tradeoff between a little book keeping to prevent duplicate tracing or biting the bullet, so to speak, in the throwing path and unwinding multiple times.

[!TIP] The choice between the Z and non-Z (zero-overhead and non-zero-overhead) variants of the exception handlers should not matter 99% of the time, however, both are provided in the rare case that it does.

CPPTRACE_TRY/CPPTRACE_CATCH could only hurt performance if used in a hot loop where the compiler can't optimize away the internal bookkeeping, otherwise the bookkeeping should be completely negligible.

CPPTRACE_TRYZ/CPPTRACE_CATCHZ could only hurt performance when there is an exceptionally deep nesting of exception handlers in a call stack before a matching handler.

More information on performance considerations with the zero-overhead variant:

Tracing the stack multiple times in throwing paths should not matter for the vast majority applications given that:

  1. Performance very rarely is critical in throwing paths and exceptions should be exceptionally rare
  2. Exception handling is not usually used in such a way that you could have a deep nesting of handlers before finding a matching handler
  3. Most call stacks are fairly shallow

To put the scale of this performance consideration into perspective: In my benchmarking I have found generation of raw traces to take on the order of 100ns per frame. Thus, even if there were 100 non-matching handlers before a matching handler in a 100-deep call stack the total time would stil be on the order of one millisecond.

Nonetheless, I chose a default bookkeeping behavior for CPPTRACE_TRY/CPPTRACE_CATCH since it is safer with better performance guarantees for the most general possible set of users.

Traced Exception Objects

Cpptrace provides a handful of traced exception classes which automatically collect stack traces when thrown. These are useful when throwing exceptions that may not be caught by CPPTRACE_CATCH.

The base traced exception class is cpptrace::exception and cpptrace provides a handful of helper classes for working with traced exceptions. These exceptions generate relatively lightweight raw traces and resolve symbols and line numbers lazily if and when requested.

These are provided both as a useful utility and as a reference implementation for traced exceptions.

The basic interface is:

namespace cpptrace {
    class exception : public std::exception {
    public:
        virtual const char* what() const noexcept = 0; // The what string both the message and trace
        virtual const char* message() const noexcept = 0;
        virtual const stacktrace& trace() const noexcept = 0;
    };
}

There are two ways to go about traced exception objects: Traces can be resolved eagerly or lazily. Cpptrace provides the basic implementation of exceptions as lazy exceptions. I hate to have anything about the implementation exposed in the interface or type system but this seems to be the best way to do this.

namespace cpptrace {
    class lazy_exception : public exception {
        // lazy_trace_holder is basically a std::variant<raw_trace, stacktrace>, more docs later
        mutable detail::lazy_trace_holder trace_holder;
        mutable std::string what_string;
    public:
        explicit lazy_exception(
            raw_trace&& trace = detail::get_raw_trace_and_absorb()
        ) noexcept : trace_holder(std::move(trace)) {}
        const char* what() const noexcept override;
        const char* message() const noexcept override;
        const stacktrace& trace() const noexcept override;
    };
}

cpptrace::lazy_exception can be freely thrown or overridden. Generally message() is the only field to override.

Lastly cpptrace provides an exception class that takes a user-provided message, cpptrace::exception_with_message, as well as a number of traced exception classes resembling <stdexcept>:

namespace cpptrace {
    class exception_with_message : public lazy_exception {
        mutable std::string user_message;
    public:
        explicit exception_with_message(
            std::string&& message_arg,
            raw_trace&& trace = detail::get_raw_trace_and_absorb()
        ) noexcept : lazy_exception(std::move(trace)), user_message(std::move(message_arg)) {}
        const char* message() const noexcept override;
    };

    // All stdexcept errors have analogs here. All but system_error have the constructor:
    // explicit the_error(
    //     std::string&& message_arg,
    //     raw_trace&& trace = detail::get_raw_trace_and_absorb()
    // ) noexcept
    //     : exception_with_message(std::move(message_arg), std::move(trace)) {}
    class logic_error      : public exception_with_message { ... };
    class domain_error     : public exception_with_message { ... };
    class invalid_argument : public exception_with_message { ... };
    class length_error     : public exception_with_message { ... };
    class out_of_range     : public exception_with_message { ... };
    class runtime_error    : public exception_with_message { ... };
    class range_error      : public exception_with_message { ... };
    class overflow_error   : public exception_with_message { ... };
    class underflow_error  : public exception_with_message { ... };
    class system_error : public runtime_error {
    public:
        explicit system_error(
            int error_code,
            std::string&& message_arg,
            raw_trace&& trace = detail::get_raw_trace_and_absorb()
        ) noexcept;
        const std::error_code& code() const noexcept;
    };
}

Wrapping std::exceptions

[!NOTE] This section is largely obsolete now that cpptrace provides a better mechanism for collecting traces from exceptions

Cpptrace exceptions can provide great information for user-controlled exceptions. For non-cpptrace::exceptions that may originate outside of code you control, e.g. the standard library, cpptrace provides some wrapper utilities that can rethrow these exceptions nested in traced cpptrace exceptions. The trace won't be perfect, the trace will start where the wrapper caught it, but these utilities can provide good diagnostic information. Unfortunately this is the best solution for this problem, as far as I know.

std::vector<int> foo = {1, 2, 3};
CPPTRACE_WRAP_BLOCK(
    foo.at(4) = 2;
    foo.at(5)++;
);
std::cout<<CPPTRACE_WRAP(foo.at(12))<<std::endl;

Exception handling with cpptrace exception objects

[!NOTE] This section pertains to cpptrace traced exception objects and not the mechanism for collecting traces from arbitrary exceptions

Working with cpptrace exceptions in your code:

try {
    foo();
} catch(cpptrace::exception& e) {
    // Prints the exception info and stack trace, conditionally enabling color codes depending on
    // whether stderr is a terminal
    std::cerr << "Error: " << e.message() << '\n';
    e.trace().print(std::cerr, cpptrace::isatty(cpptrace::stderr_fileno));
} catch(std::exception& e) {
    std::cerr << "Error: " << e.what() << '\n';
}

Terminate Handling

Cpptrace provides a custom std::terminate handler that prints stacktraces while otherwise behaving like the normal std::terminate handler. If a cpptrace exception object reaches std::terminate the trace from that exception is printed, otherwise a stack trace is generated at the point of the terminate handler. Often std::terminate is called directly without unwinding so the trace is preserved.

To register this custom handler:

cpptrace::register_terminate_handler();

Signal-Safe Tracing

Stack traces from signal handlers can provide very helpful information for debugging application crashes, e.g. from SIGSEGV or SIGTRAP handlers. Signal handlers are really restrictive environments as your application could be interrupted by a signal at any point, including in the middle of malloc or buffered IO or while holding a lock. Doing a stack trace in a signal handler is possible but it requires a lot of care. This is difficult to do correctly and most examples online do this incorrectly.

Cpptrace offers an API to walk the stack in a signal handler and produce a raw trace safely. The library also provides an interface for producing a object frame safely:

namespace cpptrace {
    std::size_t safe_generate_raw_trace(frame_ptr* buffer, std::size_t size, std::size_t skip = 0);
    std::size_t safe_generate_raw_trace(frame_ptr* buffer, std::size_t size, std::size_t skip, std::size_t max_depth);
    struct safe_object_frame {
        frame_ptr raw_address;
        frame_ptr address_relative_to_object_start;
        char object_path[CPPTRACE_PATH_MAX + 1];
        object_frame resolve() const; // To be called outside a signal handler. Not signal safe.
    };
    void get_safe_object_frame(frame_ptr address, safe_object_frame* out);
    bool can_signal_safe_unwind();
}

It is not possible to resolve debug symbols safely in the process from a signal handler without heroic effort. In order to produce a full trace there are three options:

  1. Carefully save the object trace information to be resolved at a later time outside the signal handler
  2. Write the object trace information to a file to be resolved later
  3. Spawn a new process, communicate object trace information to that process, and have that process do the trace resolution

For traces on segfaults, e.g., only options 2 and 3 are viable. For more information an implementation of approach 3, see the comprehensive overview and demo at signal-safe-tracing.md.

[!IMPORTANT] Currently signal-safe stack unwinding is only possible with libunwind, which must be manually enabled. If signal-safe unwinding isn't supported, safe_generate_raw_trace will just produce an empty trace. can_signal_safe_unwind can be used to check for signal-safe unwinding support. If object information can't be resolved in a signal-safe way then get_safe_object_frame will not populate fields beyond the raw_address.

[!IMPORTANT] _dl_find_object is required for signal-safe stack tracing. This is a relatively recent addition to glibc, added in glibc 2.35.

[!CAUTION] Calls to shared objects can be lazy-loaded where the first call to the shared object invokes non-signal-safe functions such as malloc(). To avoid this, call these routines in main() ahead of a signal handler to "warm up" the library.

Utility Types

A couple utility types are used to provide the library with a good interface.

nullable<T> is used for a nullable integer type. Internally the maximum value for T is used as a sentinel. std::optional would be used if this library weren't c++11. But, nullable<T> provides an std::optional-like interface and it's less heavy-duty for this use than an std::optional.

detail::lazy_trace_holder is a utility type for lazy_exception used in place of an std::variant<raw_trace, stacktrace>.

namespace cpptrace {
    template<typename T, typename std::enable_if<std::is_integral<T>::value, int>::type = 0>
    struct nullable {
        T raw_value;
        nullable& operator=(T value)
        bool has_value() const noexcept;
        T& value() noexcept;
        const T& value() const noexcept;
        T value_or(T alternative) const noexcept;
        void swap(nullable& other) noexcept;
        void reset() noexcept;
        bool operator==(const nullable& other) const noexcept;
        bool operator!=(const nullable& other) const noexcept;
        constexpr static nullable null() noexcept; // returns a null instance
    };

    namespace detail {
        class lazy_trace_holder {
            bool resolved;
            union {
                raw_trace trace;
                stacktrace resolved_trace;
            };
        public:
            // constructors
            lazy_trace_holder() : trace() {}
            explicit lazy_trace_holder(raw_trace&& _trace);
            explicit lazy_trace_holder(stacktrace&& _resolved_trace);
            // logistics
            lazy_trace_holder(const lazy_trace_holder& other);
            lazy_trace_holder(lazy_trace_holder&& other) noexcept;
            lazy_trace_holder& operator=(const lazy_trace_holder& other);
            lazy_trace_holder& operator=(lazy_trace_holder&& other) noexcept;
            ~lazy_trace_holder();
            // access
            const raw_trace& get_raw_trace() const;
            stacktrace& get_resolved_trace();
            const stacktrace& get_resolved_trace() const; // throws if not already resolved
        private:
            void clear();
        };
    }
}

Headers

Cpptrace provides a handful of headers to make inclusion more minimal.

HeaderContents
cpptrace/forward.hppcpptrace::frame_ptr and a few trace class forward declarations
cpptrace/basic.hppDefinitions for trace classes and the basic tracing APIs (Stack Traces, Object Traces, Raw Traces, and Signal-Safe Tracing)
cpptrace/exceptions.hppTraced Exception Objects and related utilities (Wrapping std::exceptions)
cpptrace/from_current.hppTraces From All Exceptions
cpptrace/io.hppoperator<< overloads for std::ostream and std::formatters
cpptrace/utils.hppUtility functions, configuration functions, and terminate utilities (Utilities, Configuration, and Terminate Handling)
cpptrace/version.hppLibrary version macros

The main cpptrace header is cpptrace/cpptrace.hpp which includes everything other than from_current.hpp and version.hpp.

Supported Debug Formats

FormatSupported
DWARF in binary✔️
GNU debug link️️✔️
Split dwarf (debug fission)✔️
DWARF in dSYM✔️
DWARF via Mach-O debug map✔️
Windows debug symbols in PDB✔️

DWARF5 added DWARF package files. As far as I can tell no compiler implements these yet.

How to Include The Library

CMake FetchContent

With CMake FetchContent:

include(FetchContent)
FetchContent_Declare(
  cpptrace
  GIT_REPOSITORY https://github.com/jeremy-rifkin/cpptrace.git
  GIT_TAG        v0.7.4 # <HASH or TAG>
)
FetchContent_MakeAvailable(cpptrace)
target_link_libraries(your_target cpptrace::cpptrace)

It's as easy as that. Cpptrace will automatically configure itself for your system. Note: On windows and macos some extra work is required, see Platform Logistics below.

Be sure to configure with -DCMAKE_BUILD_TYPE=Debug or -DCMAKE_BUILD_TYPE=RelWithDebInfo for symbols and line information.

System-Wide Installation

git clone https://github.com/jeremy-rifkin/cpptrace.git
git checkout v0.7.4
mkdir cpptrace/build
cd cpptrace/build
cmake .. -DCMAKE_BUILD_TYPE=Release
make -j
sudo make install

Using through cmake:

find_package(cpptrace REQUIRED)
target_link_libraries(<your target> cpptrace::cpptrace)

Be sure to configure with -DCMAKE_BUILD_TYPE=Debug or -DCMAKE_BUILD_TYPE=RelWithDebInfo for symbols and line information.

Or compile with -lcpptrace:

g++ main.cpp -o main -g -Wall -lcpptrace
./main

[!IMPORTANT] If you aren't using cmake and are linking statically you must manually specify -DCPPTRACE_STATIC_DEFINE.

If you get an error along the lines of

error while loading shared libraries: libcpptrace.so: cannot open shared object file: No such file or directory

You may have to run sudo /sbin/ldconfig to create any necessary links and update caches so the system can find libcpptrace.so (I had to do this on Ubuntu). Only when installing system-wide. Usually your package manager does this for you when installing new libraries.

[!NOTE] Libdwarf requires a relatively new version of libdwarf. Sometimes a previously-installed system-wide libdwarf may cause issues due to being too old. Libdwarf 8 and newer is known to work.

<details> <summary>System-wide install on windows</summary>
git clone https://github.com/jeremy-rifkin/cpptrace.git
git checkout v0.7.4
mkdir cpptrace/build
cd cpptrace/build
cmake .. -DCMAKE_BUILD_TYPE=Release
msbuild cpptrace.sln
msbuild INSTALL.vcxproj

Note: You'll need to run as an administrator in a developer powershell, or use vcvarsall.bat distributed with visual studio to get the correct environment variables set.

</details>

Local User Installation

To install just for the local user (or any custom prefix):

git clone https://github.com/jeremy-rifkin/cpptrace.git
git checkout v0.7.4
mkdir cpptrace/build
cd cpptrace/build
cmake .. -DCMAKE_BUILD_TYPE=Release -DCMAKE_INSTALL_PREFIX=$HOME/wherever
make -j
make install

Using through cmake:

find_package(cpptrace REQUIRED PATHS $ENV{HOME}/wherever)
target_link_libraries(<your target> cpptrace::cpptrace)

Using manually:

g++ main.cpp -o main -g -Wall -I$HOME/wherever/include -L$HOME/wherever/lib -lcpptrace

[!IMPORTANT] If you aren't using cmake and are linking statically you must manually specify -DCPPTRACE_STATIC_DEFINE.

Use Without CMake

To use the library without cmake first follow the installation instructions at System-Wide Installation, Local User Installation, or Package Managers.

In addition to any include or library paths you'll need to specify to tell the compiler where cpptrace was installed. The typical dependencies for cpptrace are:

CompilerPlatformDependencies
gcc, clang, intel, etc.Linux/macos/unix-lcpptrace -ldwarf -lz -lzstd -ldl
gccWindows-lcpptrace -ldbghelp -ldwarf -lz -lzstd
msvcWindowscpptrace.lib dbghelp.lib
clangWindows-lcpptrace -ldbghelp

Note: Newer libdwarf requires -lzstd, older libdwarf does not.

[!IMPORTANT] If you are linking statically, you will additionally need to specify -DCPPTRACE_STATIC_DEFINE.

Dependencies may differ if different back-ends are manually selected.

Installation Without Package Managers or FetchContent

Some users may prefer, or need to, to install cpptrace without package managers or fetchcontent (e.g. if their system does not have internet access). Below are instructions for how to install libdwarf and cpptrace.

<details> <summary>Installation Without Package Managers or FetchContent</summary>

Here is an example for how to build cpptrace and libdwarf. ~/scratch/cpptrace-test is used as a working directory and the libraries are installed to ~/scratch/cpptrace-test/resources.

mkdir -p ~/scratch/cpptrace-test/resources

cd ~/scratch/cpptrace-test
git clone https://github.com/facebook/zstd.git
cd zstd
git checkout 63779c798237346c2b245c546c40b72a5a5913fe
cd build/cmake
mkdir build
cd build
cmake .. -DCMAKE_INSTALL_PREFIX=~/scratch/cpptrace-test/resources -DZSTD_BUILD_PROGRAMS=On -DZSTD_BUILD_CONTRIB=On -DZSTD_BUILD_TESTS=On -DZSTD_BUILD_STATIC=On -DZSTD_BUILD_SHARED=On -DZSTD_LEGACY_SUPPORT=On
make -j
make install

cd ~/scratch/cpptrace-test
git clone https://github.com/jeremy-rifkin/libdwarf-lite.git
cd libdwarf-lite
git checkout fe09ca800b988e2ff21225ac5e7468ceade2a30e
mkdir build
cd build
cmake .. -DPIC_ALWAYS=On -DBUILD_DWARFDUMP=Off -DCMAKE_PREFIX_PATH=~/scratch/cpptrace-test/resources -DCMAKE_INSTALL_PREFIX=~/scratch/cpptrace-test/resources
make -j
make install

cd ~/scratch/cpptrace-test
git clone https://github.com/jeremy-rifkin/cpptrace.git
cd cpptrace
git checkout v0.7.4
mkdir build
cd build
cmake .. -DCMAKE_BUILD_TYPE=Release -DBUILD_SHARED_LIBS=On -DCPPTRACE_USE_EXTERNAL_LIBDWARF=On -DCMAKE_PREFIX_PATH=~/scratch/cpptrace-test/resources -DCMAKE_INSTALL_PREFIX=~/scratch/cpptrace-test/resources
make -j
make install

The ~/scratch/cpptrace-test/resources directory also serves as a bundle you can ship with all the installed files for cpptrace and its dependencies.

</details>

Package Managers

Conan

Cpptrace is available through conan at https://conan.io/center/recipes/cpptrace.

[requires]
cpptrace/0.7.4
[generators]
CMakeDeps
CMakeToolchain
[layout]
cmake_layout
# ...
find_package(cpptrace REQUIRED)
# ...
target_link_libraries(YOUR_TARGET cpptrace::cpptrace)

Vcpkg

vcpkg install cpptrace
find_package(cpptrace CONFIG REQUIRED)
target_link_libraries(main PRIVATE cpptrace::cpptrace)

Platform Logistics

Windows and macOS require a little extra work to get everything in the right place.

Windows

Copying the library .dll on Windows:

# Copy the cpptrace.dll on windows to the same directory as the executable for your_target.
# Not required if static linking.
if(WIN32)
  add_custom_command(
    TARGET your_target POST_BUILD
    COMMAND ${CMAKE_COMMAND} -E copy_if_different
    $<TARGET_FILE:cpptrace::cpptrace>
    $<TARGET_FILE_DIR:your_target>
  )
endif()

macOS

On macOS, it is recommended to generate a dSYM file containing debug information for your program. This is not required as cpptrace makes a good effort at finding and reading the debug information without this, but having a dSYM file is the most robust method.

When using Xcode with CMake, this can be done with:

set_target_properties(your_target PROPERTIES XCODE_ATTRIBUTE_DEBUG_INFORMATION_FORMAT "dwarf-with-dsym")

Outside of Xcode, this can be done with dsymutil yourbinary:

# Create a .dSYM file on macOS
if(APPLE)
  add_custom_command(
    TARGET your_target
    POST_BUILD
    COMMAND dsymutil $<TARGET_FILE:your_target>
  )
endif()

Library Back-Ends

Cpptrace supports a number of back-ends to produce stack traces. Stack traces are produced in roughly three steps: Unwinding, symbol resolution, and demangling.

The library's CMake automatically configures itself for what your system supports. The ideal configuration is as follows:

PlatformUnwindingSymbolsDemangling
Linux_Unwindlibdwarfcxxabi.h
MacOS_Unwind for gcc, execinfo.h for clang and apple clanglibdwarfcxxabi.h
WindowsStackWalk64dbghelpNo demangling needed
MinGWStackWalk64libdwarf + dbghelpcxxabi.h

Support for these back-ends is the main development focus and they should work well. If you want to use a different back-end such as addr2line, for example, you can configure the library to do so.

Unwinding

LibraryCMake configPlatformsInfo
libgcc unwindCPPTRACE_UNWIND_WITH_UNWINDlinux, macos, mingwFrames are captured with libgcc's _Unwind_Backtrace, which currently produces the most accurate stack traces on gcc/clang/mingw. Libgcc is often linked by default, and llvm has something equivalent.
execinfo.hCPPTRACE_UNWIND_WITH_EXECINFOlinux, macosFrames are captured with execinfo.h's backtrace, part of libc on linux/unix systems.
winapiCPPTRACE_UNWIND_WITH_WINAPIwindows, mingwFrames are captured with CaptureStackBackTrace.
dbghelpCPPTRACE_UNWIND_WITH_DBGHELPwindows, mingwFrames are captured with StackWalk64.
libunwindCPPTRACE_UNWIND_WITH_LIBUNWINDlinux, macos, windows, mingwFrames are captured with libunwind. Note: This is the only back-end that requires a library to be installed by the user, and a CMAKE_PREFIX_PATH may also be needed.
N/ACPPTRACE_UNWIND_WITH_NOTHINGallUnwinding is not done, stack traces will be empty.

Some back-ends (execinfo and CaptureStackBackTrace) require a fixed buffer has to be created to read addresses into while unwinding. By default the buffer can hold addresses for 400 frames (beyond the skip frames). This is configurable with CPPTRACE_HARD_MAX_FRAMES.

Symbol resolution

LibraryCMake configPlatformsInfo
libdwarfCPPTRACE_GET_SYMBOLS_WITH_LIBDWARFlinux, macos, mingwLibdwarf is the preferred method for symbol resolution for cpptrace. Cpptrace will get it via FetchContent or find_package depending on CPPTRACE_USE_EXTERNAL_LIBDWARF.
dbghelpCPPTRACE_GET_SYMBOLS_WITH_DBGHELPwindowsDbghelp.h is the preferred method for symbol resolution on windows under msvc/clang and is supported on all windows machines.
libbacktraceCPPTRACE_GET_SYMBOLS_WITH_LIBBACKTRACElinux, macos*, mingw*Libbacktrace is already installed on most systems or available through the compiler directly. For clang you must specify the absolute path to backtrace.h using CPPTRACE_BACKTRACE_PATH.
addr2lineCPPTRACE_GET_SYMBOLS_WITH_ADDR2LINElinux, macos, mingwSymbols are resolved by invoking addr2line (or atos on mac) via fork() (on linux/unix, and popen under mingw).
libdlCPPTRACE_GET_SYMBOLS_WITH_LIBDLlinux, macosLibdl uses dynamic export information. Compiling with -rdynamic is needed for symbol information to be retrievable. Line numbers won't be retrievable.
N/ACPPTRACE_GET_SYMBOLS_WITH_NOTHINGallNo attempt is made to resolve symbols.

*: Requires installation

One back-end should be used. For MinGW CPPTRACE_GET_SYMBOLS_WITH_LIBDWARF and CPPTRACE_GET_SYMBOLS_WITH_DBGHELP can be used in conjunction.

Note for addr2line: By default cmake will resolve an absolute path to addr2line to bake into the library. This path can be configured with CPPTRACE_ADDR2LINE_PATH, or CPPTRACE_ADDR2LINE_SEARCH_SYSTEM_PATH can be used to have the library search the system path for addr2line at runtime. This is not the default to prevent against path injection attacks.

Demangling

Lastly, depending on other back-ends used a demangler back-end may be needed.

LibraryCMake configPlatformsInfo
cxxabi.hCPPTRACE_DEMANGLE_WITH_CXXABILinux, macos, mingwShould be available everywhere other than msvc.
dbghelp.hCPPTRACE_DEMANGLE_WITH_WINAPIWindowsDemangle with UnDecorateSymbolName.
N/ACPPTRACE_DEMANGLE_WITH_NOTHINGallDon't attempt to do anything beyond what the symbol resolution back-end does.

More?

There are plenty more libraries that can be used for unwinding, parsing debug information, and demangling. In the future more back-ends can be added. Ideally this library can "just work" on systems, without additional installation work.

Summary of Library Configurations

Summary of all library configuration options:

Back-ends:

Back-end configuration:

Other useful configurations:

Testing:

Testing Methodology

Cpptrace currently uses integration and functional testing, building and running under every combination of back-end options. The implementation is based on github actions matrices and driven by python scripts located in the ci/ folder. Testing used to be done by github actions matrices directly, however, launching hundreds of two second jobs was extremely inefficient. Test outputs are compared against expected outputs located in test/expected/. Stack trace addresses may point to the address after an instruction depending on the unwinding back-end, and the python script will check for an exact or near-match accordingly.

Notes About the Library

For the most part I'm happy with the state of the library. But I'm sure that there is room for improvement and issues will exist. If you encounter any issue, please let me know! If you find any pain-points in the library, please let me know that too.

A note about performance: For handling of DWARF symbols there is a lot of room to explore for performance optimizations and time-memory tradeoffs. If you find the current implementation is either slow or using too much memory, I'd be happy to explore some of these options.

A couple things I'd like to improve in the future:

Contributing

I'm grateful for the help I've received with this library and I welcome contributions! For information on contributing please refer to CONTRIBUTING.md.

License

This library is under the MIT license.

Cpptrace uses libdwarf on linux, macos, and mingw/cygwin unless configured to use something else. If this library is statically linked with libdwarf then the library's binary will itself be LGPL.