view lisp/undo-stack.el @ 826:6728e641994e

[xemacs-hg @ 2002-05-05 11:30:15 by ben] syntax cache, 8-bit-format, lots of code cleanup README.packages: Update info about --package-path. i.c: Create an inheritable event and pass it on to XEmacs, so that ^C can be handled properly. Intercept ^C and signal the event. "Stop Build" in VC++ now works. bytecomp-runtime.el: Doc string changes. compat.el: Some attempts to redo this to make it truly useful and fix the "multiple versions interacting with each other" problem. Not yet done. Currently doesn't work. files.el: Use with-obsolete-variable to avoid warnings in new revert-buffer code. xemacs.mak: Split up CFLAGS into a version without flags specifying the C library. The problem seems to be that minitar depends on zlib, which depends specifically on libc.lib, not on any of the other C libraries. Unless you compile with libc.lib, you get errors -- specifically, no _errno in the other libraries, which must make it something other than an int. (#### But this doesn't seem to obtain in XEmacs, which also uses zlib, and can be linked with any of the C libraries. Maybe zlib is used differently and doesn't need errno, or maybe XEmacs provides an int errno; ... I don't understand. Makefile.in.in: Fix so that packages are around when testing. abbrev.c, alloc.c, buffer.c, buffer.h, bytecode.c, callint.c, casefiddle.c, casetab.c, casetab.h, charset.h, chartab.c, chartab.h, cmds.c, console-msw.h, console-stream.c, console-x.c, console.c, console.h, data.c, device-msw.c, device.c, device.h, dialog-msw.c, dialog-x.c, dired-msw.c, dired.c, doc.c, doprnt.c, dumper.c, editfns.c, elhash.c, emacs.c, eval.c, event-Xt.c, event-gtk.c, event-msw.c, event-stream.c, events.c, events.h, extents.c, extents.h, faces.c, file-coding.c, file-coding.h, fileio.c, fns.c, font-lock.c, frame-gtk.c, frame-msw.c, frame-x.c, frame.c, frame.h, glade.c, glyphs-gtk.c, glyphs-msw.c, glyphs-msw.h, glyphs-x.c, glyphs.c, glyphs.h, gui-msw.c, gui-x.c, gui.h, gutter.h, hash.h, indent.c, insdel.c, intl-win32.c, intl.c, keymap.c, lisp-disunion.h, lisp-union.h, lisp.h, lread.c, lrecord.h, lstream.c, lstream.h, marker.c, menubar-gtk.c, menubar-msw.c, menubar-x.c, menubar.c, minibuf.c, mule-ccl.c, mule-charset.c, mule-coding.c, mule-wnnfns.c, nas.c, objects-msw.c, objects-x.c, opaque.c, postgresql.c, print.c, process-nt.c, process-unix.c, process.c, process.h, profile.c, rangetab.c, redisplay-gtk.c, redisplay-msw.c, redisplay-output.c, redisplay-x.c, redisplay.c, redisplay.h, regex.c, regex.h, scrollbar-msw.c, search.c, select-x.c, specifier.c, specifier.h, symbols.c, symsinit.h, syntax.c, syntax.h, syswindows.h, tests.c, text.c, text.h, tooltalk.c, ui-byhand.c, ui-gtk.c, unicode.c, win32.c, window.c: Another big Ben patch. -- FUNCTIONALITY CHANGES: add partial support for 8-bit-fixed, 16-bit-fixed, and 32-bit-fixed formats. not quite done yet. (in particular, needs functions to actually convert the buffer.) NOTE: lots of changes to regex.c here. also, many new *_fmt() inline funs that take an Internal_Format argument. redo syntax cache code. make the cache per-buffer; keep the cache valid across calls to functions that use it. also keep it valid across insertions/deletions and extent changes, as much as is possible. eliminate the junky regex-reentrancy code by passing in the relevant lisp info to the regex routines as local vars. add general mechanism in extents code for signalling extent changes. fix numerous problems with the case-table implementation; yoshiki never properly transferred many algorithms from old-style to new-style case tables. redo char tables to support a default argument, so that mapping only occurs over changed args. change many chartab functions to accept Lisp_Object instead of Lisp_Char_Table *. comment out the code in font-lock.c by default, because font-lock.el no longer uses it. we should consider eliminating it entirely. Don't output bell as ^G in console-stream when not a TTY. add -mswindows-termination-handle to interface with i.c, so we can properly kill a build. add more error-checking to buffer/string macros. add some additional buffer_or_string_() funs. -- INTERFACE CHANGES AFFECTING MORE CODE: switch the arguments of write_c_string and friends to be consistent with write_fmt_string, which must have printcharfun first. change BI_* macros to BYTE_* for increased clarity; similarly for bi_* local vars. change VOID_TO_LISP to be a one-argument function. eliminate no-longer-needed CVOID_TO_LISP. -- char/string macro changes: rename MAKE_CHAR() to make_emchar() for slightly less confusion with make_char(). (The former generates an Emchar, the latter a Lisp object. Conceivably we should rename make_char() -> wrap_char() and similarly for make_int(), make_float().) Similar changes for other *CHAR* macros -- we now consistently use names with `emchar' whenever we are working with Emchars. Any remaining name with just `char' always refers to a Lisp object. rename macros with XSTRING_* to string_* except for those that reference actual fields in the Lisp_String object, following conventions used elsewhere. rename set_string_{data,length} macros (the only ones to work with a Lisp_String_* instead of a Lisp_Object) to set_lispstringp_* to make the difference clear. try to be consistent about caps vs. lowercase in macro/inline-fun names for chars and such, which wasn't the case before. we now reserve caps either for XFOO_ macros that reference object fields (e.g. XSTRING_DATA) or for things that have non-function semantics, e.g. directly modifying an arg (BREAKUP_EMCHAR) or evaluating an arg (any arg) more than once. otherwise, use lowercase. here is a summary of most of the macros/inline funs changed by all of the above changes: BYTE_*_P -> byte_*_p XSTRING_BYTE -> string_byte set_string_data/length -> set_lispstringp_data/length XSTRING_CHAR_LENGTH -> string_char_length XSTRING_CHAR -> string_emchar INTBYTE_FIRST_BYTE_P -> intbyte_first_byte_p INTBYTE_LEADING_BYTE_P -> intbyte_leading_byte_p charptr_copy_char -> charptr_copy_emchar LEADING_BYTE_* -> leading_byte_* CHAR_* -> EMCHAR_* *_CHAR_* -> *_EMCHAR_* *_CHAR -> *_EMCHAR CHARSET_BY_ -> charset_by_* BYTE_SHIFT_JIS* -> byte_shift_jis* BYTE_BIG5* -> byte_big5* REP_BYTES_BY_FIRST_BYTE -> rep_bytes_by_first_byte char_to_unicode -> emchar_to_unicode valid_char_p -> valid_emchar_p Change intbyte_strcmp -> qxestrcmp_c (duplicated functionality). -- INTERFACE CHANGES AFFECTING LESS CODE: use DECLARE_INLINE_HEADER in various places. remove '#ifdef emacs' from XEmacs-only files. eliminate CHAR_TABLE_VALUE(), which duplicated the functionality of get_char_table(). add BUFFER_TEXT_LOOP to simplify iterations over buffer text. define typedefs for signed and unsigned types of fixed sizes (INT_32_BIT, UINT_32_BIT, etc.). create ALIGN_FOR_TYPE as a higher-level interface onto ALIGN_SIZE; fix code to use it. add charptr_emchar_len to return the text length of the character pointed to by a ptr; use it in place of charcount_to_bytecount(..., 1). add emchar_len to return the text length of a given character. add types Bytexpos and Charxpos to generalize Bytebpos/Bytecount and Charbpos/Charcount, in code (particularly, the extents code and redisplay code) that works with either kind of index. rename redisplay struct params with names such as `charbpos' to e.g. `charpos' when they are e.g. a Charxpos, not a Charbpos. eliminate xxDEFUN in place of DEFUN; no longer necessary with changes awhile back to doc.c. split up big ugly combined list of EXFUNs in lisp.h on a file-by-file basis, since other prototypes are similarly split. rewrite some "*_UNSAFE" macros as inline funs and eliminate the _UNSAFE suffix. move most string code from lisp.h to text.h; the string code and text.h code is now intertwined in such a fashion that they need to be in the same place and partially interleaved. (you can't create forward references for inline funs) automated/lisp-tests.el, automated/symbol-tests.el, automated/test-harness.el: Fix test harness to output FAIL messages to stderr when in batch mode. Fix up some problems in lisp-tests/symbol-tests that were causing spurious failures.
author ben
date Sun, 05 May 2002 11:33:57 +0000
parents 3ecd8885ac67
children 308d34e9f07d
line wrap: on
line source

;;; undo-stack.el --- An "undoable stack" object.

;; Copyright (C) 1997 Free Software Foundation, Inc.
;; Copyright (C) 1996 Ben Wing.

;; Maintainer: XEmacs Development Team
;; Keywords: extensions, dumped

;; This file is part of XEmacs.

;; XEmacs is free software; you can redistribute it and/or modify it
;; under the terms of the GNU General Public License as published by
;; the Free Software Foundation; either version 2, or (at your option)
;; any later version.

;; XEmacs is distributed in the hope that it will be useful, but
;; WITHOUT ANY WARRANTY; without even the implied warranty of
;; MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
;; General Public License for more details.

;; You should have received a copy of the GNU General Public License
;; along with XEmacs; see the file COPYING.  If not, write to the 
;; Free Software Foundation, 59 Temple Place - Suite 330,
;; Boston, MA 02111-1307, USA.

;;; Synched up with: Not in FSF.

;;; Commentary:

;; This file is dumped with XEmacs.

;; An "undoable stack" is an object that can be used to implement
;; a history of positions, with undo and redo.  Conceptually, it
;; is the kind of data structure used to keep track of (e.g.)
;; visited Web pages, so that the "Back" and "Forward" operations
;; in the browser work.  Basically, I can successively visit a
;; number of Web pages through links, and then hit "Back" a
;; few times to go to previous positions, and then "Forward" a
;; few times to reverse this process.  This is similar to an
;; "undo" and "redo" mechanism.

;; Note that Emacs does not standardly contain structures like
;; this.  Instead, it implements history using either a ring
;; (the kill ring, the mark ring), or something like the undo
;; stack, where successive "undo" operations get recorded as
;; normal modifications, so that if you do a bunch of successive
;; undo's, then something else, then start undoing, you will
;; be redoing all your undo's back to the point before you did
;; the undo's, and then further undo's will act like the previous
;; round of undo's.  I think that both of these paradigms are
;; inferior to the "undoable-stack" paradigm because they're
;; confusing and difficult to keep track of.

;; Conceptually, imagine a position history like this:

;;   1 -> 2 -> 3 -> 4 -> 5 -> 6
;;                            ^^

;; where the arrow indicates where you currently are.  "Going back"
;; and "going forward" just amount to moving the arrow.  However,
;; what happens if the history state is this:

;;   1 -> 2 -> 3 -> 4 -> 5 -> 6
;;                  ^^

;; and then I visit new positions (7) and (8)?  In the most general
;; implementation, you've just caused a new branch like this:

;;   1 -> 2 -> 3 -> 4 -> 5 -> 6
;;                  |
;;                  |
;;                  7 -> 8
;;                       ^^

;; But then you can end up with a whole big tree, and you need
;; more sophisticated ways of navigating ("Forward" might involve
;; a choice of paths to follow) and managing its size (if you don't
;; want to keep unlimited history, you have to truncate at some point,
;; and how do you truncate a tree?)

;; My solution to this is just to insert the new positions like
;; this:

;;   1 -> 2 -> 3 -> 4 -> 7 -> 8 -> 5 -> 6
;;                            ^^

;; (Netscape, I think, would just truncate 5 and 6 completely,
;; but that seems a bit drastic.  In the Emacs-standard "ring"
;; structure, this problem is avoided by simply moving 5 and 6
;; to the beginning of the ring.  However, it doesn't seem
;; logical to me to have "going back past 1" get you to 6.)

;; Now what if we have a "maximum" size of (say) 7 elements?
;; When we add 8, we could truncate either 1 or 6.  Since 5 and
;; 6 are "undone" positions, we should presumably truncate
;; them before 1.  So, adding 8 truncates 6, adding 9 truncates
;; 5, and adding 10 truncates 1 because there is nothing more
;; that is forward of the insertion point.

;; Interestingly, this method of truncation is almost like
;; how a ring would truncate.  A ring would move 5 and 6
;; around to the back, like this:

;;   5 -> 6 -> 1 -> 2 -> 3 -> 4 -> 7 -> 8
;;                                      ^^

;; However, when 8 is added, the ring truncates 5 instead of
;; 6, which is less than optimal.

;; Conceptually, we can implement the "undoable stack" using
;; two stacks of a sort called "truncatable stack", which are
;; just simple stacks, but where you can truncate elements
;; off of the bottom of the stack.  Then, the undoable stack

;;   1 -> 2 -> 3 -> 4 -> 5 -> 6
;;                  ^^

;; is equivalent to two truncatable stacks:

;;   4 <- 3 <- 2 <- 1
;;   5 <- 6

;; where I reversed the direction to accord with the probable
;; implementation of a standard list.  To do another undo,
;; I pop 4 off of the first stack and move it to the top of
;; the second stack.  A redo operation does the opposite.
;; To truncate to the proper size, first chop off 6, then 5,
;; then 1 -- in all cases, truncating off the bottom.

;;; Code:

(define-error 'trunc-stack-bottom "Bottom of stack reached")

(defsubst trunc-stack-stack (stack)
  ;; return the list representing the trunc-stack's elements.
  ;; the head of the list is the most recent element.
  (aref stack 1))

(defsubst trunc-stack-length (stack)
  ;; return the number of elements in the trunc-stack.
  (aref stack 2))

(defsubst set-trunc-stack-stack (stack new)
  ;; set the list representing the trunc-stack's elements.
  (aset stack 1 new))

(defsubst set-trunc-stack-length (stack new)
  ;; set the length of the trunc-stack.
  (aset stack 2 new))

;; public functions:

(defun make-trunc-stack ()
  ;; make an empty trunc-stack.
  (vector 'trunc-stack nil 0))

(defun trunc-stack-push (stack el)
  ;; push a new element onto the head of the trunc-stack.
  (set-trunc-stack-stack stack (cons el (trunc-stack-stack stack)))
  (set-trunc-stack-length stack (1+ (trunc-stack-length stack))))

(defun trunc-stack-top (stack &optional n)
  ;; return the nth topmost element from the trunc-stack.
  ;; signal an error if the stack doesn't have that many elements.
  (or n (setq n 0))
  (if (>= n (trunc-stack-length stack))
      (signal-error 'trunc-stack-bottom (list stack))
    (nth n (trunc-stack-stack stack))))

(defun trunc-stack-pop (stack)
  ;; pop and return the topmost element from the stack.
  (prog1 (trunc-stack-top stack)
    (set-trunc-stack-stack stack (cdr (trunc-stack-stack stack)))
    (set-trunc-stack-length stack (1- (trunc-stack-length stack)))))

(defun trunc-stack-truncate (stack &optional n)
  ;; truncate N items off the bottom of the stack.  If the stack is
  ;; not that big, it just becomes empty.
  (or n (setq n 1))
  (if (> n 0)
      (let ((len (trunc-stack-length stack)))
	(if (>= n len)
	    (progn
	      (set-trunc-stack-length stack 0)
	      (set-trunc-stack-stack stack nil))
	  (setcdr (nthcdr (1- (- len n)) (trunc-stack-stack stack)) nil)
	  (set-trunc-stack-length stack (- len n))))))

;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;

;;; FMH! FMH! FMH!  This object-oriented stuff doesn't really work
;;; properly without built-in structures (vectors suck) and without
;;; public and private functions and fields.

(defsubst undoable-stack-max (stack)
  (aref stack 1))

(defsubst undoable-stack-a (stack)
  (aref stack 2))

(defsubst undoable-stack-b (stack)
  (aref stack 3))

;; public functions:

(defun make-undoable-stack (max)
  ;; make an empty undoable stack of max size MAX.
  (vector 'undoable-stack max (make-trunc-stack) (make-trunc-stack)))

(defsubst set-undoable-stack-max (stack new)
  ;; change the max size of an undoable stack.
  (aset stack 1 new))

(defun undoable-stack-a-top (stack)
  ;; return the topmost element off the "A" stack of an undoable stack.
  ;; this is the most recent position pushed on the undoable stack.
  (trunc-stack-top (undoable-stack-a stack)))

(defun undoable-stack-a-length (stack)
  (trunc-stack-length (undoable-stack-a stack)))

(defun undoable-stack-b-top (stack)
  ;; return the topmost element off the "B" stack of an undoable stack.
  ;; this is the position that will become the most recent position,
  ;; after a redo operation.
  (trunc-stack-top (undoable-stack-b stack)))

(defun undoable-stack-b-length (stack)
  (trunc-stack-length (undoable-stack-b stack)))

(defun undoable-stack-push (stack el)
  ;; push an element onto the stack.
  (let*
      ((lena (trunc-stack-length (undoable-stack-a stack)))
       (lenb (trunc-stack-length (undoable-stack-b stack)))
       (max (undoable-stack-max stack))
       (len (+ lena lenb)))
    ;; maybe truncate some elements.  We have to deal with the
    ;; possibility that we have more elements than our max
    ;; (someone might have reduced the max).
    (if (>= len max)
	(let ((must-nuke (1+ (- len max))))
	  ;; chop off must-nuke elements from the B stack.
	  (trunc-stack-truncate (undoable-stack-b stack) must-nuke)
	  ;; but if there weren't that many elements to chop,
	  ;; take the rest off the A stack.
	  (if (< lenb must-nuke)
	      (trunc-stack-truncate (undoable-stack-a stack)
				    (- must-nuke lenb)))))
    (trunc-stack-push (undoable-stack-a stack) el)))

(defun undoable-stack-pop (stack)
  ;; pop an element off the stack.
  (trunc-stack-pop (undoable-stack-a stack)))

(defun undoable-stack-undo (stack)
  ;; transfer an element from the top of A to the top of B.
  ;; return value is undefined.
  (trunc-stack-push (undoable-stack-b stack)
		    (trunc-stack-pop (undoable-stack-a stack))))

(defun undoable-stack-redo (stack)
  ;; transfer an element from the top of B to the top of A.
  ;; return value is undefined.
  (trunc-stack-push (undoable-stack-a stack)
		    (trunc-stack-pop (undoable-stack-b stack))))


;;; undo-stack.el ends here