backward::SignalHandling crash_status not resetting to crash_status::running #351
Description
On Windows at least, there appears to be an edge case that only shows up when not linking against backward.cpp's singleton instance of backward::SignalHandling. Instead, in my code I was creating and destroying SignalHandling instances as needed. As a result, the reporter_thread_ exits shortly after starting, since crashed() returns crash_status::ending by the time that thread starts back up again in the new SignalHandling instance. Do you intend to support re-creating SignalHandling instances? If so, I have a potential fix below. It works in my testing, but only if at most one SignalHandling instance exists at a time (this doesn't make sense to support anyway).
diff --git a/backward.hpp b/backward.hpp
index 670aa45..632b372 100644
--- a/backward.hpp
+++ b/backward.hpp
@@ -4311,28 +4311,38 @@ private:
class SignalHandling {
public:
SignalHandling(const std::vector<int> & = std::vector<int>())
- : reporter_thread_([]() {
- /* We handle crashes in a utility thread:
- backward structures and some Windows functions called here
- need stack space, which we do not have when we encounter a
- stack overflow.
- To support reporting stack traces during a stack overflow,
- we create a utility thread at startup, which waits until a
- crash happens or the program exits normally. */
-
- {
- std::unique_lock<std::mutex> lk(mtx());
- cv().wait(lk, [] { return crashed() != crash_status::running; });
- }
- if (crashed() == crash_status::crashed) {
- handle_stacktrace(skip_recs());
- }
- {
- std::unique_lock<std::mutex> lk(mtx());
- crashed() = crash_status::ending;
- }
- cv().notify_one();
- }) {
+ {
+ // Reset crash status
+ {
+ std::unique_lock<std::mutex> lk(mtx());
+ crashed() = crash_status::running;
+ }
+
+ reporter_thread_ = std::thread([]() {
+ /* We handle crashes in a utility thread:
+ backward structures and some Windows functions called here
+ need stack space, which we do not have when we encounter a
+ stack overflow.
+ To support reporting stack traces during a stack overflow,
+ we create a utility thread at startup, which waits until a
+ crash happens or the program exits normally. */
+
+ crash_status status;
+ {
+ std::unique_lock<std::mutex> lk(mtx());
+ cv().wait(lk, [] { return crashed() != crash_status::running; });
+ status = crashed();
+ }
+ if (status == crash_status::crashed) {
+ handle_stacktrace(skip_recs());
+ }
+ {
+ std::unique_lock<std::mutex> lk(mtx());
+ crashed() = crash_status::ending;
+ }
+ cv().notify_one();
+ });
+
SetUnhandledExceptionFilter(crash_handler);
signal(SIGABRT, signal_handler);