diff lisp/cl-macs.el @ 4906:6ef8256a020a

implement equalp in C, fix case-folding, add equal() method for keymaps -------------------- ChangeLog entries follow: -------------------- lisp/ChangeLog addition: 2010-02-01 Ben Wing <ben@xemacs.org> * cl-extra.el: * cl-extra.el (cl-string-vector-equalp): Removed. * cl-extra.el (cl-bit-vector-vector-equalp): Removed. * cl-extra.el (cl-vector-array-equalp): Removed. * cl-extra.el (cl-hash-table-contents-equalp): Removed. * cl-extra.el (equalp): Removed. * cl-extra.el (cl-mapcar-many): Comment out the whole `equalp' implementation for the moment; remove once we're sure the C implementation works. * cl-macs.el: * cl-macs.el (equalp): Simplify the compiler-macro for `equalp' -- once it's in C, we don't need to try so hard to expand it. src/ChangeLog addition: 2010-02-01 Ben Wing <ben@xemacs.org> * abbrev.c (abbrev_match_mapper): * buffer.h (CANON_TABLE_OF): * buffer.h: * editfns.c (Fchar_equal): * minibuf.c (scmp_1): * text.c (qxestrcasecmp_i18n): * text.c (qxestrncasecmp_i18n): * text.c (qxetextcasecmp): * text.c (qxetextcasecmp_matching): Create new macro CANONCASE that converts to a canonical mapping and use it to do caseless comparisons instead of DOWNCASE. * alloc.c: * alloc.c (cons_equal): * alloc.c (vector_equal): * alloc.c (string_equal): * bytecode.c (compiled_function_equal): * chartab.c (char_table_entry_equal): * chartab.c (char_table_equal): * data.c (weak_list_equal): * data.c (weak_box_equal): * data.c (ephemeron_equal): * device-msw.c (equal_devmode): * elhash.c (hash_table_equal): * events.c (event_equal): * extents.c (properties_equal): * extents.c (extent_equal): * faces.c: * faces.c (face_equal): * faces.c (face_hash): * floatfns.c (float_equal): * fns.c: * fns.c (bit_vector_equal): * fns.c (plists_differ): * fns.c (Fplists_eq): * fns.c (Fplists_equal): * fns.c (Flax_plists_eq): * fns.c (Flax_plists_equal): * fns.c (internal_equal): * fns.c (internal_equalp): * fns.c (internal_equal_0): * fns.c (syms_of_fns): * glyphs.c (image_instance_equal): * glyphs.c (glyph_equal): * glyphs.c (glyph_hash): * gui.c (gui_item_equal): * lisp.h: * lrecord.h (struct lrecord_implementation): * marker.c (marker_equal): * number.c (bignum_equal): * number.c (ratio_equal): * number.c (bigfloat_equal): * objects.c (color_instance_equal): * objects.c (font_instance_equal): * opaque.c (equal_opaque): * opaque.c (equal_opaque_ptr): * rangetab.c (range_table_equal): * specifier.c (specifier_equal): Add a `foldcase' param to the equal() method and use it to implement `equalp' comparisons. Also add to plists_differ(), although we don't currently use it here. Rewrite internal_equalp(). Implement cross-type vector comparisons. Don't implement our own handling of numeric promotion -- just use the `=' primitive. Add internal_equal_0(), which takes a `foldcase' param and calls either internal_equal() or internal_equalp(). * buffer.h: When given a 0 for buffer (which is the norm when functions don't have a specific buffer available), use the current buffer's table, not `standard-case-table'; otherwise the current settings are ignored. * casetab.c: * casetab.c (set_case_table): When handling old-style vectors of 256 in `set-case-table' don't overwrite the existing table! Instead create a new table and populate. * device-msw.c (sync_printer_with_devmode): * lisp.h: * text.c (lisp_strcasecmp_ascii): Rename lisp_strcasecmp to lisp_strcasecmp_ascii and use lisp_strcasecmp_i18n for caseless comparisons in some places. * elhash.c: Delete unused lisp_string_hash and lisp_string_equal(). * events.h: * keymap-buttons.h: * keymap.h: * keymap.c (keymap_lookup_directly): * keymap.c (keymap_store): * keymap.c (FROB): * keymap.c (key_desc_list_to_event): * keymap.c (describe_map_mapper): * keymap.c (INCLUDE_BUTTON_ZERO): New file keymap-buttons.h; use to handle buttons 1-26 in place of duplicating code 26 times. * frame-gtk.c (allocate_gtk_frame_struct): * frame-msw.c (mswindows_init_frame_1): Fix some comments about internal_equal() in redisplay that don't apply any more. * keymap-slots.h: * keymap.c: New file keymap-slots.h. Use it to notate the slots in a keymap structure, similar to frameslots.h or coding-system-slots.h. * keymap.c (MARKED_SLOT): * keymap.c (keymap_equal): * keymap.c (keymap_hash): Implement. tests/ChangeLog addition: 2010-02-01 Ben Wing <ben@xemacs.org> * automated/case-tests.el: * automated/case-tests.el (uni-mappings): * automated/search-tests.el: Delete old pristine-case-table code. Rewrite the Unicode torture test to take into account whether overlapping mappings exist for more than one character, and not doing the upcase/downcase comparisons in such cases. * automated/lisp-tests.el (foo): * automated/lisp-tests.el (string-variable): * automated/lisp-tests.el (featurep): Replace Assert (equal ... with Assert-equal; same for other types of equality. Replace some awkward equivalents of Assert-equalp with Assert-equalp. Add lots of equalp tests. * automated/case-tests.el: * automated/regexp-tests.el: * automated/search-tests.el: Fix up the comments at the top of the files. Move rules about where to put tests into case-tests.el. * automated/test-harness.el: * automated/test-harness.el (test-harness-aborted-summary-template): New. * automated/test-harness.el (test-harness-from-buffer): * automated/test-harness.el (batch-test-emacs): Fix Assert-test-not. Create Assert-not-equal and variants. Delete the doc strings from all these convenience functions to avoid excessive repetition; instead use one copy in a comment.
author Ben Wing <ben@xemacs.org>
date Mon, 01 Feb 2010 01:02:40 -0600
parents 6772ce4d982b
children 6bc1f3f6cf0d
line wrap: on
line diff
--- a/lisp/cl-macs.el	Sat Jan 30 20:34:23 2010 -0600
+++ b/lisp/cl-macs.el	Mon Feb 01 01:02:40 2010 -0600
@@ -1,7 +1,7 @@
 ;;; cl-macs.el --- Common Lisp extensions for XEmacs Lisp (part four)
 
 ;; Copyright (C) 1993, 2003, 2004 Free Software Foundation, Inc.
-;; Copyright (C) 2002 Ben Wing.
+;; Copyright (C) 2002, 2010 Ben Wing.
 
 ;; Author: Dave Gillespie <daveg@synaptics.com>
 ;; Version: 2.02
@@ -3357,6 +3357,21 @@
       (regexp-quote string)
     form))
 
+;; NOTE: `equalp' is now a primitive, although as of yet it still doesn't
+;; have a byte-compiler opcode for it.  The compiler-macro for `equalp' used
+;; to try and remove as much as possible of the logic of the Lisp `equalp' as
+;; possible whenever one of the arguments is a constant, boiling things down
+;; to a few if-statements and some calls to various no-longer-defined
+;; helper functions.  Besides the fact that the helper functions aren't
+;; defined, there's little point in doing any of that expansion, since it will
+;; end up executing in Lisp what would otherwise be done in C by a direct
+;; call to `equalp'.  The only exception is when the reduction is quite
+;; simple and is to functions that do have op-codes; that may gain something.
+;; However, if `equalp' becomes an opcode itself, consider removing everything
+;; here except maybe when the call can directly be reduced to `equal' or `eq'.
+;;
+;; --ben
+
 (define-compiler-macro equalp (&whole form x y) 
   "Expand calls to `equalp' where X or Y is a constant expression.
 
@@ -3372,8 +3387,6 @@
   ;; constant here.
   (let* ((equalp-sym (eval-when-compile (gensym)))
 	(let-form '(progn))
-	(check-bit-vector t)
-	(check-string t)
 	(original-y y)
 	equalp-temp checked)
   (macrolet
@@ -3398,44 +3411,11 @@
 	    ((member x '("" #* []))
 	     ;; No need to protect against multiple evaluation here:
 	     `(and (member ,original-y '("" #* [])) t))
-	    ((stringp x)
-	     `(,@let-form
-	       (if (stringp ,y)
-		   (eq t (compare-strings ,x nil nil
-					  ,y nil nil t))
-		 (if (vectorp ,y) 
-		     (cl-string-vector-equalp ,x ,y)))))
-	    ((bit-vector-p x)
-	     `(,@let-form
-	       (if (bit-vector-p ,y)
-		   ;; No need to call equalp on each element here:
-		   (equal ,x ,y)
-		 (if (vectorp ,y) 
-		     (cl-bit-vector-vector-equalp ,x ,y)))))
-	    (t
-	     (loop
-	       for elt across x
-	       ;; We may not need to check the other argument if it's a
-	       ;; string or bit vector, depending on the contents of x:
-	       always (progn
-			(unless (characterp elt) (setq check-string nil))
-			(unless (and (numberp elt) (or (= elt 0) (= elt 1)))
-			  (setq check-bit-vector nil))
-			(or check-string check-bit-vector)))
-	     `(,@let-form
-	       (cond
-		,@(if check-string
-		      `(((stringp ,y) 
-			 (cl-string-vector-equalp ,y ,x))))
-		,@(if check-bit-vector 
-		      `(((bit-vector-p ,y)
-			 (cl-bit-vector-vector-equalp ,y ,x))))
-		((vectorp ,y)
-		 (cl-vector-array-equalp ,x ,y)))))))
+	    (t form)))
 	  ((unordered-check (and (characterp x) (not (cl-const-expr-p y))))
 	   `(,@let-form
 	     (or (eq ,x ,y)
-		  ;; eq has a bytecode, char-equal doesn't.
+		 ;; eq has a bytecode, char-equal doesn't.
 		 (and (characterp ,y)
 		      (eq (downcase ,x) (downcase ,y))))))
 	  ((unordered-check (and (numberp x) (not (cl-const-expr-p y))))
@@ -3443,31 +3423,146 @@
 	     (and (numberp ,y)
 		  (= ,x ,y))))
 	  ((unordered-check (and (hash-table-p x) (not (cl-const-expr-p y))))
-	   ;; Hash tables; follow the CL spec.
-	   `(,@let-form
-	     (and (hash-table-p ,y)
-		  (eq ',(hash-table-test x) (hash-table-test ,y))
-		  (= ,(hash-table-count x) (hash-table-count ,y))
-		  (cl-hash-table-contents-equalp ,x ,y))))
+	   form)
 	  ((unordered-check
 	    ;; Symbols; eq. 
 	    (and (not (cl-const-expr-p y))
 		 (or (memq x '(nil t))
 		     (and (eq (car-safe x) 'quote) (symbolp (second x))))))
 	   (cons 'eq (cdr form)))
-	  ((unordered-check
-	    ;; Compare conses at runtime, there's no real upside to
-	    ;; unrolling the function -> they fall through to the next
-	    ;; clause in this function.
-	    (and (cl-const-expr-p x) (not (consp x))
-		 (not (cl-const-expr-p y))))
-	   ;; All other types; use equal.
-	   (cons 'equal (cdr form)))
+
+	  ;; This clause is wrong -- e.g. when comparing a constant char-table
+	  ;; against a non-constant expression that evaluates to a char-table,
+	  ;; or some for range tables or certain other types, `equalp' is
+	  ;; not the same as `equal'.  We could insert the known list of
+	  ;; types with special `equalp' property, but it's fragile and may
+	  ;; not be much of an optimization, esp. since these types don't
+	  ;; occur that often are often big.
+	  ;;((unordered-check
+	  ;;  ;; Compare conses at runtime, there's no real upside to
+	  ;;  ;; unrolling the function -> they fall through to the next
+	  ;;  ;; clause in this function.
+	  ;;  (and (cl-const-expr-p x) (not (consp x))
+	  ;;       (not (cl-const-expr-p y))))
+	  ;; ;; All other types; use equal.
+	  ;; (cons 'equal (cdr form)))
+	  
 	  ;; Neither side is a constant expression, do all our evaluation at
 	  ;; runtime (or both are, and equalp will be called from
 	  ;; byte-optimize.el).
 	  (t form)))))
 
+;;(define-compiler-macro equalp (&whole form x y) 
+;;  "Expand calls to `equalp' where X or Y is a constant expression.
+;;
+;;Much of the processing that `equalp' does is dependent on the types of both
+;;of its arguments, and with type information for one of them, we can
+;;eliminate much of the body of the function at compile time.
+;;
+;;Where both X and Y are constant expressions, `equalp' is evaluated at
+;;compile time by byte-optimize.el--this compiler macro passes FORM through to
+;;the byte optimizer in those cases."
+;;  ;; Cases where both arguments are constant are handled in
+;;  ;; byte-optimize.el, we only need to handle those cases where one is
+;;  ;; constant here.
+;;  (let* ((equalp-sym (eval-when-compile (gensym)))
+;;	(let-form '(progn))
+;;	(check-bit-vector t)
+;;	(check-string t)
+;;	(original-y y)
+;;	equalp-temp checked)
+;;  (macrolet
+;;      ((unordered-check (check)
+;;	 `(prog1
+;;	     (setq checked
+;;		   (or ,check
+;;		       (prog1 ,(sublis '((x . y) (y . x)) check :test #'eq)
+;;			 (setq equalp-temp x x y y equalp-temp))))
+;;	   (when checked
+;;	     (unless (symbolp y)
+;;	       (setq let-form `(let ((,equalp-sym ,y))) y equalp-sym))))))
+;;    ;; In the bodies of the below clauses, x is always a constant expression
+;;    ;; of the type we're interested in, and y is always a symbol that refers
+;;    ;; to the result non-constant side of the comparison. 
+;;    (cond ((unordered-check (and (arrayp x) (not (cl-const-expr-p y))))
+;;	   ;; Strings and other arrays. A vector containing the same
+;;	   ;; character elements as a given string is equalp to that string;
+;;	   ;; a bit-vector can only be equalp to a string if both are
+;;	   ;; zero-length.
+;;	   (cond
+;;	    ((member x '("" #* []))
+;;	     ;; No need to protect against multiple evaluation here:
+;;	     `(and (member ,original-y '("" #* [])) t))
+;;	    ((stringp x)
+;;	     `(,@let-form
+;;	       (if (stringp ,y)
+;;		   (eq t (compare-strings ,x nil nil
+;;					  ,y nil nil t))
+;;		 (if (vectorp ,y) 
+;;		     (cl-string-vector-equalp ,x ,y)))))
+;;	    ((bit-vector-p x)
+;;	     `(,@let-form
+;;	       (if (bit-vector-p ,y)
+;;		   ;; No need to call equalp on each element here:
+;;		   (equal ,x ,y)
+;;		 (if (vectorp ,y) 
+;;		     (cl-bit-vector-vector-equalp ,x ,y)))))
+;;	    (t
+;;	     (loop
+;;	       for elt across x
+;;	       ;; We may not need to check the other argument if it's a
+;;	       ;; string or bit vector, depending on the contents of x:
+;;	       always (progn
+;;			(unless (characterp elt) (setq check-string nil))
+;;			(unless (and (numberp elt) (or (= elt 0) (= elt 1)))
+;;			  (setq check-bit-vector nil))
+;;			(or check-string check-bit-vector)))
+;;	     `(,@let-form
+;;	       (cond
+;;		,@(if check-string
+;;		      `(((stringp ,y) 
+;;			 (cl-string-vector-equalp ,y ,x))))
+;;		,@(if check-bit-vector 
+;;		      `(((bit-vector-p ,y)
+;;			 (cl-bit-vector-vector-equalp ,y ,x))))
+;;		((vectorp ,y)
+;;		 (cl-vector-array-equalp ,x ,y)))))))
+;;	  ((unordered-check (and (characterp x) (not (cl-const-expr-p y))))
+;;	   `(,@let-form
+;;	     (or (eq ,x ,y)
+;;		  ;; eq has a bytecode, char-equal doesn't.
+;;		 (and (characterp ,y)
+;;		      (eq (downcase ,x) (downcase ,y))))))
+;;	  ((unordered-check (and (numberp x) (not (cl-const-expr-p y))))
+;;	   `(,@let-form
+;;	     (and (numberp ,y)
+;;		  (= ,x ,y))))
+;;	  ((unordered-check (and (hash-table-p x) (not (cl-const-expr-p y))))
+;;	   ;; Hash tables; follow the CL spec.
+;;	   `(,@let-form
+;;	     (and (hash-table-p ,y)
+;;		  (eq ',(hash-table-test x) (hash-table-test ,y))
+;;		  (= ,(hash-table-count x) (hash-table-count ,y))
+;;		  (cl-hash-table-contents-equalp ,x ,y))))
+;;	  ((unordered-check
+;;	    ;; Symbols; eq. 
+;;	    (and (not (cl-const-expr-p y))
+;;		 (or (memq x '(nil t))
+;;		     (and (eq (car-safe x) 'quote) (symbolp (second x))))))
+;;	   (cons 'eq (cdr form)))
+;;	  ((unordered-check
+;;	    ;; Compare conses at runtime, there's no real upside to
+;;	    ;; unrolling the function -> they fall through to the next
+;;	    ;; clause in this function.
+;;	    (and (cl-const-expr-p x) (not (consp x))
+;;		 (not (cl-const-expr-p y))))
+;;	   ;; All other types; use equal.
+;;	   (cons 'equal (cdr form)))
+;;	  ;; Neither side is a constant expression, do all our evaluation at
+;;	  ;; runtime (or both are, and equalp will be called from
+;;	  ;; byte-optimize.el).
+;;	  (t form)))))
+
 (define-compiler-macro map (&whole form cl-type cl-func cl-seq
                             &rest cl-rest)
   "If CL-TYPE is a constant expression that we know how to handle, transform