Memory leak due to calling backtrace_create_state multiple times

[Cogitri]

Hello,

currently boost-stacktrace with the backtrace backend will call backtrace_create_state once for each thread. This creates a memory leak since each time a new thread is created, a new backtrace_state is allocated but it's never freed (and can't be since there's no backtrace_destroy_state). See e.g. https://patchwork.ozlabs.org/project/gcc/patch/3defaafeedc9fa6b42e61401d4c8c84e@starynkevitch.net/#1625428 for more discussion on this.

[apolukhin]

Unfortunately, it's not a bug but a feature:

stacktrace/include/boost/stacktrace/detail/libbacktrace_impls.hpp

Lines 74 to 85 in 08d720a

	// TODO: The most obvious solution:
	//
	//static ::backtrace_state* state = ::backtrace_create_state(
	// prog_location.name(),
	// 1, // allow safe concurrent usage of the same state
	// boost::stacktrace::detail::libbacktrace_error_callback,
	// 0 // pointer to data that will be passed to callback
	//);
	//
	//
	// Unfortunately, that solution segfaults when `construct_state()` function is in .so file
	// and multiple threads concurrently work with state.

There's some issue either with the packaged versions of libbacktrace, or some issue in the libbacktrace with some compiler flags that I failed to localize https://gcc.gnu.org/bugzilla//show_bug.cgi?id=87653

I have to add more docs and references into the sources and provide a macro to force single instance for those' who wish to take the risk

[pergraa-celonis]

Should that be static and not ststic here?
308b7f6#diff-1f3dccb3928ad7b52e4c986270cb619b663197283b764c2fb68256d285f21a3fR124

[apolukhin]

Oops, thanks!