Latest crashes with profiling on

Hello All! I got profiling to work fairly reliable for short profiling windows of seconds or minutes. When I leave XEmacs unattended for many minutes to hours it eventually crashes. The typical signature is: current_profile_function() line 168 + 4 bytes sigprof_handler(int 0x0000001d) line 306 + 5 bytes setitimer_helper_proc(unsigned int 0x0000f631, unsigned int 0x00000000, unsigned long 0x0000001d, unsigned long 0x00000000, unsigned long 0x00000000) line 788 + 19 bytes WINMM! 76b454f3() WINMM! 76b5ae36() WINMM! 76b5af3a() KERNEL32! 7c80b50b() This is at "=>" in the following source fragment. At this point the internal state of variables is already beyond belief. I'll send in my patch to eval.c which gets me even to that point of "stability" for review. I'll attach my full log of recent backtraces in case anyone has time and interest to take a look. Best regards! Adrian ---------------------------------------------------------------------------- static Lisp_Object current_profile_function (void) { Lisp_Object fun; struct backtrace *bt = backtrace_list; /* 2 because we set in_profiling when we entered the current routine. */ if (in_profiling >= 2) return QSprofile_overhead; /* Find a function actually being called. Potentially (?) there could be a number of non-calling funs -- calling foo autoloads, which tries to call bar, but requires evalling its args first, which calls baz, ... If profiling was not enabled when the function was called, just treat the function as actually called, because the info about whether we've finished the preamble will not have been recorded. */ =>for (; bt && !bt->function_being_called; bt = bt->next) ---------------------------------------------------------------------------- -- Adrian Aichner mailto:adrian＠xemacs.org http://www.xemacs.org/ _______________________________________________ XEmacs-Beta mailing list XEmacs-Beta(a)xemacs.org http://calypso.tux.org/cgi-bin/mailman/listinfo/xemacs-beta