* Jonathan Wakely: > +/** Stop the program with a breakpoint or debug trap. > + * > + * The details of how a breakpoint is implemented are platform-specific. > + * Some systems provide a special instruction, such as `int3` in x86. > + * When no more appropriate mechanism is available, this will stop the > + * program using `__builtin_trap()`. It might not be possible for the > + * program to continue after such a breakpoint. > + * > + * @since C++26 > + */ > +void > +breakpoint() noexcept;
Would it make sense to have a special function symbol for this, on which the debugger sets the breakpoint? We already have this for a slightly different purpose in glibc, see _dl_debug_state. Otherwise you'll get crashes if the debugger detection delivers a false positive (e.g. when running under an strace-equivalent if libstdc++ cannot check the process name). Or is_debugger_present should perhaps read a volatile bool that is updated by the debugger as needed, and not bother with any ptrace checks. Thanks, Florian
