Re: fontlock-speed

"Stephen J. Turnbull" <stephen(a)xemacs.org&gt;, on 08 Feb 2002 at 14:56:48 +0900 you wrote: > Sorry, I know nothing about profiling C yet, I guess it's time to > learn. Well, if you take the tack I just tried, you'll soon find out that XEmacs is NOT set up to work with gprof. This is what I've done so far. 1) Configure with --cflags="-g -pg" and --ldflags="-pg". 2) Edit src/Makefile.in so that it does not explicitly specify the crt files to link with. I wasn't sure what would cause problems and what wouldn't, so I: a) Removed "-lgcc -lc -lgcc /usr/lib/crtn.o" from the end of ld_libs_all b) Removed the -nostdlib flag from LD c) Emptied start_files and lib_cc (Is there some platform on which all this is necessary? Letting the compiler sort all of this out certainly works fine on Linux and Solaris, as far as I can tell.) 3) Watch temacs dump perfectly good profiling data ... and then watch xemacs run, exit normally, and fail to write any profiling data. I thought at first that maybe XEmacs was calling _exit() instead of exit(). There are a number of places in the source where _exit() is used instead of exit(): a) callproc.c: child_setup, relocate_fd b) ecrt0.c: various startup routines c) process-unix.c: unix_create_process d) sysdep.c: sys_subshell, mkdir, rmdir But, no, XEmacs is exiting via Fkill_emacs, which definitely calls exit(). It took me awhile to figure this one out, but I finally got it. I've been bitten by the dumping process. The glibc code for __gmon_start__ looks like this: ------------------------------------------------------------------------ void __gmon_start__ (void) { #ifdef HAVE_INITFINI /* Protect from being called more than once. Since crti.o is linked into every shared library, each of their init functions will call us. */ static int called; if (called) return; called = 1; #endif /* Start keeping profiling records. */ __monstartup ((u_long) &_start, (u_long) &etext); /* Call _mcleanup before exiting; it will write out gmon.out from the collected data. */ atexit (&_mcleanup); } ------------------------------------------------------------------------ So temacs, before dumping, has called = 0, and sets it to 1. The value of 1 is then dumped, so xemacs enters __gmon_start__, says "I've been here before!", and leaves. Argh! This is going to make it kind of hard to profile XEmacs 21.1, where --pdump is not available. Anyway, I used hexl mode to edit the binary and changed the dumped value back to 0. Now I am getting profile data out of xemacs. But this took so much time that I don't have time to do the same to XEmacs 21.4 now. I'll get back to you later on that. -- Jerry James