Subject: Re: dlopen for xemacs Date: Mon, 20 Apr 1998 15:05:33 -0400 From: Joshua Rowe <rowe(a)eksystems.com&gt; To: SL Baur <steve(a)xemacs.org&gt; References: <9801152235.AA13534(a)meta.eksystems.com&gt; <m2n2djg4dy.fsf(a)altair.xemacs.org&gt; SL Baur writes: > I don't recall if I ever got back with you on this. We've integrated > dlopen into XEmacs and have been hacking on it. William Perry has > done some further work on it, particularly with regards to portability > and configure issues. The latest word is that Richard will be using > a version of it in FSF GNU Emacs as well. > > Thanks for kicking things off. Sure thing. Maybe I should have posted this to a mailing list/newsgroup, but I really don't pay attention to things like that. Forward this and forget it if you wish. I'm not sure if I ever told you the intent of the code. Maybe you can find someone to figure out the specifics: It is much more useful to use native compiled code if you don't have to recompile xemacs whenever the code changes. It is good to keep most of the code in xemacs in lisp form. It would be good to have a lisp->native-machine-code compiler. It is now possible to load native code without recompiling xemacs. The lisp->native-machine-code compiler can be platform independent by outputting xemacs-standard C - ie it uses all the C macros in lisp.h, etc - and compiling that code into a shared library which can then be dlopen'ed. The actual code that is output can be very simple for a first pass - there is already a byte compiler. Take a piece of byte compiled code and output the C that gets run for that bytecode. bytecode.c has the operations and the C actions. Output those C actions. I'm not sure exactly how much faster the code could become in this fashion, but I believe it could be _much_ faster if someone wrote a lisp->C instead of my current suggestion of lisp->bytecode->C. For example, from disassemble bbdb-add-hook: 0 constant boundp 1 varref hook 2 call 1 3 goto-if-not-nil 9 would become: Lisp_Object * bbdb_add_hook_constants; DEFUN ("bbdb-add-hook"...) { /* variable declarations, including constant's */ Lisp_Object v1, v2; int op; /* other stuff, including initialize the stack */ do switch (op = FETCH) { 0: /* constant boundp */ PUSH(bbdb_add_hook_constants [0]); 1: v1 = bbdb_add_hook_constants [5]; if (!SYMBOLP (v1)) v2 = Fsymbol_value (v1); else { v2 = XSYMBOL (v1)->value; if (SYMBOL_VALUE_MAGIC_P (v2)) v2 = Fsymbol_value (v1); } PUSH (v2); 2: DISCARD (1); TOP = Ffuncall (op + 1, &TOP); 3: op = FETCH2; if (!NILP (TOP)) { pc = op; break; } } while (1); ... } -- jmr - http://meta.eksystems.com/~rowe Just because you're paranoid doesn't mean they're not after you. -----BEGIN GEEK CODE BLOCK----- Version: 3.1 GCS dpu s: a--- C++++$ ULOU++++$ P+ L++++$ E+++$ W++ !N- !o>+ ?K w---(++) O+ M-- V+++$ PS+ PE Y PGP- t+ 5 X+++ R>+ !tv b+++ DI++ D- G e h++ r y ------END GEEK CODE BLOCK------