add more byte-code assertions and better failure output -------------------- ChangeLog entries follow: -------------------- src/ChangeLog addition: 2010-02-03 Ben Wing <ben@xemacs.org> * alloc.c (Fmake_byte_code): * bytecode.h: * lisp.h: * lread.c: * lread.c (readevalloop): * lread.c (Fread): * lread.c (Fread_from_string): * lread.c (read_list_conser): * lread.c (read_list): * lread.c (vars_of_lread): * symbols.c: * symbols.c (Fdefine_function): Turn on the "compiled-function annotation hack". Implement it properly by hooking into Fdefalias(). Note in the docstring to `defalias' that we do this. Remove some old broken code and change code that implemented the old kludgy way of hooking into the Lisp reader into bracketed by `#ifdef COMPILED_FUNCTION_ANNOTATION_HACK_OLD_WAY', which is not enabled. Also enable byte-code metering when DEBUG_XEMACS -- this is a form of profiling for computing histograms of which sequences of two bytecodes are used most often. * bytecode-ops.h: * bytecode-ops.h (OPCODE): New file. Extract out all the opcodes and declare them using OPCODE(), a bit like frame slots and such. This way the file can be included multiple times if necessary to iterate multiple times over the byte opcodes. * bytecode.c: * bytecode.c (NUM_REMEMBERED_BYTE_OPS): * bytecode.c (OPCODE): * bytecode.c (assert_failed_with_remembered_ops): * bytecode.c (READ_UINT_2): * bytecode.c (READ_INT_1): * bytecode.c (READ_INT_2): * bytecode.c (PEEK_INT_1): * bytecode.c (PEEK_INT_2): * bytecode.c (JUMP_RELATIVE): * bytecode.c (JUMP_NEXT): * bytecode.c (PUSH): * bytecode.c (POP_WITH_MULTIPLE_VALUES): * bytecode.c (DISCARD): * bytecode.c (UNUSED): * bytecode.c (optimize_byte_code): * bytecode.c (optimize_compiled_function): * bytecode.c (Fbyte_code): * bytecode.c (vars_of_bytecode): * bytecode.c (init_opcode_table_multi_op): * bytecode.c (reinit_vars_of_bytecode): * emacs.c (main_1): * eval.c (funcall_compiled_function): * symsinit.h: Any time we change either the instruction pointer or the stack pointer, assert that we're going to move it to a valid location. This should catch failures right when they occur rather than sometime later. This requires that we pass in another couple of parameters into some functions (only with error-checking enabled, see below). Also keep track, using a circular queue, of the last 100 byte opcodes seen, and when we hit an assert failure during byte-code execution, output the contents of the queue in a nice readable fashion. This requires that bytecode-ops.h be included a second time so that a table mapping opcodes to the name of their operation can be constructed. This table is constructed in new function reinit_vars_of_bytecode(). Everything in the last two paras happens only when ERROR_CHECK_BYTE_CODE. Add some longish comments describing how the arrays that hold the stack and instructions, and the pointers used to access them, work. * gc.c: Import some code from my `latest-fix' workspace to mark the staticpro's in order from lowest to highest, rather than highest to lowest, so it's easier to debug when something goes wrong. * lisp.h (abort_with_message): Renamed from abort_with_msg(). * symbols.c (defsymbol_massage_name_1): * symbols.c (defsymbol_nodump): * symbols.c (defsymbol): * symbols.c (defkeyword): * symeval.h (DEFVAR_SYMVAL_FWD_OBJECT): Make the various calls to staticpro() instead call staticpro_1(), passing in the name of the C var being staticpro'ed, so that it shows up in staticpro_names. Otherwise staticpro_names just has 1000+ copies of the word `location'.

--- a/src/bytecode.h	Tue Feb 02 15:19:15 2010 -0600
+++ b/src/bytecode.h	Wed Feb 03 08:01:55 2010 -0600
@@ -67,8 +67,19 @@
 #define COMPILED_INTERACTIVE	5
 #define COMPILED_DOMAIN		6
 
-/* It doesn't make sense to have this and also have load-history */
-/* #define COMPILED_FUNCTION_ANNOTATION_HACK */
+/* Someone claims: [[ It doesn't make sense to have this and also have
+   load-history ]] But in fact they are quite different things.  Perhaps
+   we should turn this on only when DEBUG_XEMACS but there's no speed
+   harm at all, so no reason not to do it always. */
+#define COMPILED_FUNCTION_ANNOTATION_HACK
+
+#ifdef DEBUG_XEMACS
+/* Define BYTE_CODE_METER to enable generation of a byte-op usage
+   histogram.  This isn't defined in FSF Emacs and isn't defined in XEmacs
+   v19.  But this is precisely the thing to turn on when DEBUG_XEMACS.  It
+   may lead to a slight speed penalty but nothing major. */
+#define BYTE_CODE_METER
+#endif
 
 struct Lisp_Compiled_Function
 {
@@ -131,6 +142,9 @@
 
 typedef unsigned char Opbyte;
 Lisp_Object execute_optimized_program (const Opbyte *program,
+#ifdef ERROR_CHECK_BYTE_CODE
+				       Elemcount program_length,
+#endif
 				       int stack_depth,
 				       Lisp_Object *constants_data);