Comment fix.

[bpt/emacs.git] / lisp / format.el

;;; format.el --- read and save files in multiple formats

;; Copyright (c) 1994, 1995, 1997, 1999 Free Software Foundation

;; Author: Boris Goldowsky <boris@gnu.org>

;; This file is part of GNU Emacs.

;; GNU Emacs 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.

;; GNU Emacs 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 GNU Emacs; see the file COPYING.  If not, write to the
;; Free Software Foundation, Inc., 59 Temple Place - Suite 330,
;; Boston, MA 02111-1307, USA.

;;; Commentary:

;; This file defines a unified mechanism for saving & loading files stored
;; in different formats.  `format-alist' contains information that directs
;; Emacs to call an encoding or decoding function when reading or writing
;; files that match certain conditions.
;;
;; When a file is visited, its format is determined by matching the
;; beginning of the file against regular expressions stored in
;; `format-alist'.  If this fails, you can manually translate the buffer
;; using `format-decode-buffer'.  In either case, the formats used are
;; listed in the variable `buffer-file-format', and become the default
;; format for saving the buffer.  To save a buffer in a different format,
;; change this variable, or use `format-write-file'.
;;
;; Auto-save files are normally created in the same format as the visited
;; file, but the variable `auto-save-file-format' can be set to a
;; particularly fast or otherwise preferred format to be used for
;; auto-saving (or nil to do no encoding on auto-save files, but then you
;; risk losing any text-properties in the buffer).
;;
;; You can manually translate a buffer into or out of a particular format
;; with the functions `format-encode-buffer' and `format-decode-buffer'.
;; To translate just the region use the functions `format-encode-region'
;; and `format-decode-region'.
;;
;; You can define a new format by writing the encoding and decoding
;; functions, and adding an entry to `format-alist'.  See enriched.el for
;; an example of how to implement a file format.  There are various
;; functions defined in this file that may be useful for writing the
;; encoding and decoding functions:
;;  * `format-annotate-region' and `format-deannotate-region' allow a
;;     single alist of information to be used for encoding and decoding.
;;     The alist defines a correspondence between strings in the file
;;     ("annotations") and text-properties in the buffer.
;;  * `format-replace-strings' is similarly useful for doing simple
;;     string->string translations in a reversible manner.

;;; Code:

(put 'buffer-file-format 'permanent-local t)

(defvar format-alist
  '((text/enriched "Extended MIME text/enriched format."
		   "Content-[Tt]ype:[ \t]*text/enriched"
		   enriched-decode enriched-encode t enriched-mode)
    (plain "ISO 8859-1 standard format, no text properties."
	   ;; Plain only exists so that there is an obvious neutral choice in
	   ;; the completion list.
	   nil nil nil nil nil)
    (ibm   "IBM Code Page 850 (DOS)"
	   nil				; The original "1\\(^\\)" is obscure.
	   "recode -f ibm-pc:latin1" "recode -f latin1:ibm-pc" t nil)
    (mac   "Apple Macintosh"
	   nil
	   "recode -f mac:latin1" "recode -f latin1:mac" t nil)
    (hp    "HP Roman8"
	   nil
	   "recode -f roman8:latin1" "recode -f latin1:roman8" t nil)
    (TeX   "TeX (encoding)"
	   nil
	   iso-tex2iso iso-iso2tex t nil)
    (gtex  "German TeX (encoding)"
	   nil
	   iso-gtex2iso iso-iso2gtex t nil)
    (html  "HTML/SGML \"ISO 8879:1986//ENTITIES Added Latin 1//EN\" (encoding)"
	   nil
	   iso-sgml2iso iso-iso2sgml t nil)
    (rot13 "rot13"
	   nil
	   "tr a-mn-z n-za-m" "tr a-mn-z n-za-m" t nil)
    (duden "Duden Ersatzdarstellung"
	   nil
	   "diac" iso-iso2duden t nil)
    (de646 "German ASCII (ISO 646)"
	   nil
	   "recode -f iso646-ge:latin1" "recode -f latin1:iso646-ge" t nil)
    (denet "net German"
	   nil
	   iso-german iso-cvt-read-only t nil)
    (esnet "net Spanish"
	   nil
	   iso-spanish iso-cvt-read-only t nil))
  "List of information about understood file formats.
Elements are of the form \(NAME DOC-STR REGEXP FROM-FN TO-FN MODIFY MODE-FN).

NAME    is a symbol, which is stored in `buffer-file-format'.

DOC-STR should be a single line providing more information about the
        format.  It is currently unused, but in the future will be shown to
        the user if they ask for more information.

REGEXP  is a regular expression to match against the beginning of the file;
        it should match only files in that format.  Use nil to avoid
        matching at all for formats for which this isn't appropriate to
        require explicit encoding/decoding.

FROM-FN is called to decode files in that format; it gets two args, BEGIN
        and END, and can make any modifications it likes, returning the new
        end.  It must make sure that the beginning of the file no longer
        matches REGEXP, or else it will get called again.
	Alternatively, FROM-FN can be a string, which specifies a shell command
	(including options) to be used as a filter to perform the conversion.

TO-FN   is called to encode a region into that format; it is passed three
        arguments: BEGIN, END, and BUFFER.  BUFFER is the original buffer that
        the data being written came from, which the function could use, for
        example, to find the values of local variables.  TO-FN should either
        return a list of annotations like `write-region-annotate-functions',
        or modify the region and return the new end.
	Alternatively, TO-FN can be a string, which specifies a shell command
	(including options) to be used as a filter to perform the conversion.

MODIFY, if non-nil, means the TO-FN wants to modify the region.  If nil,
        TO-FN will not make any changes but will instead return a list of
        annotations.

MODE-FN, if specified, is called when visiting a file with that format.
         It is called with a single positive argument, on the assumption
         that it turns on some Emacs mode.")

;;; Basic Functions (called from Lisp)

(defun format-encode-run-method (method from to &optional buffer)
  "Translate using function or shell script METHOD the text from FROM to TO.
If METHOD is a string, it is a shell command;
otherwise, it should be a Lisp function.
BUFFER should be the buffer that the output originally came from."
  (if (stringp method)
      (let ((error-buff (get-buffer-create "*Format Errors*"))
	    (coding-system-for-read 'no-conversion)
	    format-alist)
	(with-current-buffer error-buff
	  (widen)
	  (erase-buffer))
    	(if (and (zerop (save-window-excursion
			  (shell-command-on-region from to method t t
						   error-buff)))
		 ;; gzip gives zero exit status with bad args, for instance.
		 (zerop (with-current-buffer error-buff
			  (buffer-size))))
	    (bury-buffer error-buff)
	  (switch-to-buffer-other-window error-buff)
	  (error "Format encoding failed")))
    (funcall method from to buffer)))

(defun format-decode-run-method (method from to &optional buffer)
  "Decode using function or shell script METHOD the text from FROM to TO.
If METHOD is a string, it is a shell command; otherwise, it should be
a Lisp function.  Decoding is done for the given BUFFER."
  (if (stringp method)
      (let ((error-buff (get-buffer-create "*Format Errors*"))
	    (coding-system-for-write 'no-conversion)
	    format-alist)
	(with-current-buffer error-buff
	  (widen)
	  (erase-buffer))
	;; We should perhaps go via a temporary buffer and copy it
	;; back, in case of errors.
	(if (and (zerop (save-window-excursion
			  (shell-command-on-region (point-min) (point-max)
						   method t t
						   error-buff)))
		 ;; gzip gives zero exit status with bad args, for instance.
		 (zerop (with-current-buffer error-buff
			  (buffer-size))))
	    (bury-buffer error-buff)
	  (switch-to-buffer-other-window error-buff)
	  (error "Format decoding failed"))
	(point))
    (funcall method from to)))

(defun format-annotate-function (format from to orig-buf format-count)
  "Return annotations for writing region as FORMAT.
FORMAT is a symbol naming one of the formats defined in `format-alist',
it must be a single symbol, not a list like `buffer-file-format'.
FROM and TO delimit the region to be operated on in the current buffer.
ORIG-BUF is the original buffer that the data came from.

FORMAT-COUNT is an integer specifying how many times this function has
been called in the process of decoding ORIG-BUF.

This function works like a function on `write-region-annotate-functions':
it either returns a list of annotations, or returns with a different buffer
current, which contains the modified text to write.  In the latter case,
this function's value is nil.

For most purposes, consider using `format-encode-region' instead."
  ;; This function is called by write-region (actually
  ;; build_annotations) for each element of buffer-file-format.
  (let* ((info (assq format format-alist))
	 (to-fn  (nth 4 info))
	 (modify (nth 5 info)))
    (if to-fn
	(if modify
	    ;; To-function wants to modify region.  Copy to safe place.
	    (let ((copy-buf (get-buffer-create (format " *Format Temp %d*"
						       format-count))))
	      (copy-to-buffer copy-buf from to)
	      (set-buffer copy-buf)
	      (format-insert-annotations write-region-annotations-so-far from)
	      (format-encode-run-method to-fn (point-min) (point-max) orig-buf)
	      nil)
	  ;; Otherwise just call function, it will return annotations.
	  (funcall to-fn from to orig-buf)))))

(defun format-decode (format length &optional visit-flag)
  ;; This function is called by insert-file-contents whenever a file is read.
  "Decode text from any known FORMAT.
FORMAT is a symbol appearing in `format-alist' or a list of such symbols,
or nil, in which case this function tries to guess the format of the data by
matching against the regular expressions in `format-alist'.  After a match is
found and the region decoded, the alist is searched again from the beginning
for another match.

Second arg LENGTH is the number of characters following point to operate on.
If optional third arg VISIT-FLAG is true, set `buffer-file-format'
to the list of formats used, and call any mode functions defined for those
formats.

Returns the new length of the decoded region.

For most purposes, consider using `format-decode-region' instead."
  (let ((mod (buffer-modified-p))
	(begin (point))
	(end (+ (point) length)))
    (unwind-protect
	(progn
	  ;; Don't record undo information for the decoding.
	  
	  (if (null format)
	      ;; Figure out which format it is in, remember list in `format'.
	      (let ((try format-alist))
		(while try
		  (let* ((f (car try))
			 (regexp (nth 2 f))
			 (p (point)))
		    (if (and regexp (looking-at regexp)
			     (< (match-end 0) (+ begin length)))
			(progn
			  (setq format (cons (car f) format))
			  ;; Decode it
			  (if (nth 3 f)
			      (setq end (format-decode-run-method (nth 3 f) begin end)))
			  ;; Call visit function if required
			  (if (and visit-flag (nth 6 f)) (funcall (nth 6 f) 1))
			  ;; Safeguard against either of the functions changing pt.
			  (goto-char p)
			  ;; Rewind list to look for another format
			  (setq try format-alist))
		      (setq try (cdr try))))))
	    ;; Deal with given format(s)
	    (or (listp format) (setq format (list format)))
	    (let ((do format) f)
	      (while do
		(or (setq f (assq (car do) format-alist))
		    (error "Unknown format" (car do)))
		;; Decode:
		(if (nth 3 f)
		    (setq end (format-decode-run-method (nth 3 f) begin end)))
		;; Call visit function if required
		(if (and visit-flag (nth 6 f)) (funcall (nth 6 f) 1))
		(setq do (cdr do)))))
	  (if visit-flag
	      (setq buffer-file-format format)))
      
      (set-buffer-modified-p mod))

      ;; Return new length of region
    (- end begin)))

;;;
;;; Interactive functions & entry points
;;;

(defun format-decode-buffer (&optional format)
  "Translate the buffer from some FORMAT.
If the format is not specified, this function attempts to guess.
`buffer-file-format' is set to the format used, and any mode-functions
for the format are called."
  (interactive
   (list (format-read "Translate buffer from format (default: guess): ")))
  (save-excursion
    (goto-char (point-min))
    (format-decode format (buffer-size) t)))

(defun format-decode-region (from to &optional format)
  "Decode the region from some format.
Arg FORMAT is optional; if omitted the format will be determined by looking
for identifying regular expressions at the beginning of the region."
  (interactive
   (list (region-beginning) (region-end)
	 (format-read "Translate region from format (default: guess): ")))
  (save-excursion
    (goto-char from)
    (format-decode format (- to from) nil)))

(defun format-encode-buffer (&optional format)
  "Translate the buffer into FORMAT.
FORMAT defaults to `buffer-file-format'.  It is a symbol naming one of the
formats defined in `format-alist', or a list of such symbols."
  (interactive
   (list (format-read (format "Translate buffer to format (default %s): "
			      buffer-file-format))))
  (format-encode-region (point-min) (point-max) format))

(defun format-encode-region (beg end &optional format)
  "Translate the region into some FORMAT.
FORMAT defaults to `buffer-file-format', it is a symbol naming
one of the formats defined in `format-alist', or a list of such symbols."
  (interactive
   (list (region-beginning) (region-end)
	 (format-read (format "Translate region to format (default %s): "
			      buffer-file-format))))
  (if (null format)    (setq format buffer-file-format))
  (if (symbolp format) (setq format (list format)))
  (save-excursion
    (goto-char end)
    (let ((cur-buf (current-buffer))
	  (end (point-marker)))
      (while format
	(let* ((info (assq (car format) format-alist))
	       (to-fn  (nth 4 info))
	       (modify (nth 5 info))
	       result)
	  (if to-fn
	      (if modify
		  (setq end (format-encode-run-method to-fn beg end
						      (current-buffer)))
		(format-insert-annotations
		 (funcall to-fn beg end (current-buffer)))))
	  (setq format (cdr format)))))))

(defun format-write-file (filename format)
  "Write current buffer into file FILENAME using some FORMAT.
Makes buffer visit that file and sets the format as the default for future
saves.  If the buffer is already visiting a file, you can specify a directory
name as FILENAME, to write a file of the same old name in that directory."
  (interactive
   ;; Same interactive spec as write-file, plus format question.
   (let* ((file (if buffer-file-name
		    (read-file-name "Write file: "
				    nil nil nil nil)
		  (read-file-name "Write file: "
				  (cdr (assq 'default-directory
					     (buffer-local-variables)))
				  nil nil (buffer-name))))
	  (fmt (format-read (format "Write file `%s' in format: "
				    (file-name-nondirectory file)))))
     (list file fmt)))
  (setq buffer-file-format format)
  (write-file filename))

(defun format-find-file (filename format)
  "Find the file FILENAME using data format FORMAT.
If FORMAT is nil then do not do any format conversion."
  (interactive
   ;; Same interactive spec as write-file, plus format question.
   (let* ((file (read-file-name "Find file: "))
	  (fmt (format-read (format "Read file `%s' in format: "
				    (file-name-nondirectory file)))))
     (list file fmt)))
  (let ((format-alist nil))
     (find-file filename))
  (if format
      (format-decode-buffer format)))

(defun format-insert-file (filename format &optional beg end)
  "Insert the contents of file FILENAME using data format FORMAT.
If FORMAT is nil then do not do any format conversion.
The optional third and fourth arguments BEG and END specify
the part of the file to read.

The return value is like the value of `insert-file-contents':
a list (ABSOLUTE-FILE-NAME . SIZE)."
  (interactive
   ;; Same interactive spec as write-file, plus format question.
   (let* ((file (read-file-name "Find file: "))
	  (fmt (format-read (format "Read file `%s' in format: "
				    (file-name-nondirectory file)))))
     (list file fmt)))
  (let (value size)
    (let ((format-alist nil))
      (setq value (insert-file-contents filename nil beg end))
      (setq size (nth 1 value)))
    (if format
	(setq size (format-decode format size)
	      value (cons (car value) size)))
    value))

(defun format-read (&optional prompt)
  "Read and return the name of a format.
Return value is a list, like `buffer-file-format'; it may be nil.
Formats are defined in `format-alist'.  Optional arg is the PROMPT to use."
  (let* ((table (mapcar (lambda (x) (list (symbol-name (car x))))
			format-alist))
	 (ans (completing-read (or prompt "Format: ") table nil t)))
    (if (not (equal "" ans)) (list (intern ans)))))


;;;
;;; Below are some functions that may be useful in writing encoding and
;;; decoding functions for use in format-alist.
;;;

(defun format-replace-strings (alist &optional reverse beg end)
  "Do multiple replacements on the buffer.
ALIST is a list of (from . to) pairs, which should be proper arguments to
`search-forward' and `replace-match' respectively.
Optional 2nd arg REVERSE, if non-nil, means the pairs are (to . from), so that
you can use the same list in both directions if it contains only literal
strings.
Optional args BEG and END specify a region of the buffer on which to operate."
  (save-excursion
    (save-restriction
      (or beg (setq beg (point-min)))
      (if end (narrow-to-region (point-min) end))
      (while alist
	(let ((from (if reverse (cdr (car alist)) (car (car alist))))
	      (to   (if reverse (car (car alist)) (cdr (car alist)))))
	  (goto-char beg)
	  (while (search-forward from nil t)
	    (goto-char (match-beginning 0))
	    (insert to)
	    (set-text-properties (- (point) (length to)) (point)
				 (text-properties-at (point)))
	    (delete-region (point) (+ (point) (- (match-end 0)
						 (match-beginning 0)))))
	  (setq alist (cdr alist)))))))

;;; Some list-manipulation functions that we need.

(defun format-delq-cons (cons list)
  "Remove the given CONS from LIST by side effect and return the new LIST.
Since CONS could be the first element of LIST, write
`\(setq foo \(format-delq-cons element foo))' to be sure of changing
the value of `foo'."
  (if (eq cons list)
      (cdr list)
    (let ((p list))
      (while (not (eq (cdr p) cons))
	(if (null p) (error "format-delq-cons: not an element."))
	(setq p (cdr p)))
      ;; Now (cdr p) is the cons to delete
      (setcdr p (cdr cons))
      list)))
    
(defun format-make-relatively-unique (a b)
  "Delete common elements of lists A and B, return as pair.
Compares using `equal'."
  (let* ((acopy (copy-sequence a))
	 (bcopy (copy-sequence b))
	 (tail acopy))
    (while tail
      (let ((dup (member (car tail) bcopy))
	    (next (cdr tail)))
	(if dup (setq acopy (format-delq-cons tail acopy)
		      bcopy (format-delq-cons dup  bcopy)))
	(setq tail next)))
    (cons acopy bcopy)))

(defun format-common-tail (a b)
  "Given two lists that have a common tail, return it.
Compares with `equal', and returns the part of A that is equal to the
equivalent part of B.  If even the last items of the two are not equal,
returns nil."
  (let ((la (length a))
	(lb (length b)))
    ;; Make sure they are the same length
    (if (> la lb)
	(setq a (nthcdr (- la lb) a))
      (setq b (nthcdr (- lb la) b))))
  (while (not (equal a b))
    (setq a (cdr a)
	  b (cdr b)))
  a)

(defun format-proper-list-p (list)
  "Return t if LIST is a proper list.
A proper list is a list ending with a nil cdr, not with an atom "
  (when (listp list)
    (while (consp list)
      (setq list (cdr list)))
    (null list)))

(defun format-reorder (items order)
  "Arrange ITEMS to following partial ORDER.
Elements of ITEMS equal to elements of ORDER will be rearranged to follow the
ORDER.  Unmatched items will go last."
  (if order
      (let ((item (member (car order) items)))
	(if item
	    (cons (car item)
		  (format-reorder (format-delq-cons item items)
			   (cdr order)))
	  (format-reorder items (cdr order))))
    items))

(put 'face 'format-list-valued t)	; These text-properties take values
(put 'unknown 'format-list-valued t)	; that are lists, the elements of which
					; should be considered separately.
					; See format-deannotate-region and
					; format-annotate-region.

;; This text property has list values, but they are treated atomically.

(put 'display 'format-list-atomic-p t)

;;;
;;; Decoding
;;;

(defun format-deannotate-region (from to translations next-fn)
  "Translate annotations in the region into text properties.
This sets text properties between FROM to TO as directed by the
TRANSLATIONS and NEXT-FN arguments.

NEXT-FN is a function that searches forward from point for an annotation.
It should return a list of 4 elements: \(BEGIN END NAME POSITIVE).  BEGIN and
END are buffer positions bounding the annotation, NAME is the name searched
for in TRANSLATIONS, and POSITIVE should be non-nil if this annotation marks
the beginning of a region with some property, or nil if it ends the region.
NEXT-FN should return nil if there are no annotations after point.

The basic format of the TRANSLATIONS argument is described in the
documentation for the `format-annotate-region' function.  There are some
additional things to keep in mind for decoding, though:

When an annotation is found, the TRANSLATIONS list is searched for a
text-property name and value that corresponds to that annotation.  If the
text-property has several annotations associated with it, it will be used only
if the other annotations are also in effect at that point.  The first match
found whose annotations are all present is used.

The text property thus determined is set to the value over the region between
the opening and closing annotations.  However, if the text-property name has a
non-nil `format-list-valued' property, then the value will be consed onto the
surrounding value of the property, rather than replacing that value.

There are some special symbols that can be used in the \"property\" slot of
the TRANSLATIONS list: PARAMETER and FUNCTION \(spelled in uppercase).
Annotations listed under the pseudo-property PARAMETER are considered to be
arguments of the immediately surrounding annotation; the text between the
opening and closing parameter annotations is deleted from the buffer but saved
as a string.  The surrounding annotation should be listed under the
pseudo-property FUNCTION.  Instead of inserting a text-property for this
annotation, the function listed in the VALUE slot is called to make whatever
changes are appropriate.  The function's first two arguments are the START and
END locations, and the rest of the arguments are any PARAMETERs found in that
region.

Any annotations that are found by NEXT-FN but not defined by TRANSLATIONS
are saved as values of the `unknown' text-property \(which is list-valued).
The TRANSLATIONS list should usually contain an entry of the form
    \(unknown \(nil format-annotate-value))
to write these unknown annotations back into the file."
  (save-excursion
    (save-restriction
      (narrow-to-region (point-min) to)
      (goto-char from)
      (let (next open-ans todo loc unknown-ans)
	(while (setq next (funcall next-fn))
	  (let* ((loc      (nth 0 next))
		 (end      (nth 1 next))
		 (name     (nth 2 next))
		 (positive (nth 3 next))
		 (found    nil))

	    ;; Delete the annotation
	    (delete-region loc end)
	    (cond
	     ;; Positive annotations are stacked, remembering location
	     (positive (setq open-ans (cons `(,name ((,loc . nil))) open-ans)))
	     ;; It is a negative annotation:
	     ;; Close the top annotation & add its text property.
	     ;; If the file's nesting is messed up, the close might not match
	     ;; the top thing on the open-annotations stack.
	     ;; If no matching annotation is open, just ignore the close.
	     ((not (assoc name open-ans))
	      (message "Extra closing annotation (%s) in file" name))
	     ;; If one is open, but not on the top of the stack, close
	     ;; the things in between as well.  Set `found' when the real
	     ;; one is closed.
	     (t
	      (while (not found)
		(let* ((top (car open-ans))	; first on stack: should match.
		       (top-name (car top))	; text property name
		       (top-extents (nth 1 top)) ; property regions
		       (params (cdr (cdr top)))	; parameters
		       (aalist translations)
		       (matched nil))
		  (if (equal name top-name)
		      (setq found t)
		    (message "Improper nesting in file."))
		  ;; Look through property names in TRANSLATIONS
		  (while aalist
		    (let ((prop (car (car aalist)))
			  (alist (cdr (car aalist))))
		      ;; And look through values for each property
		      (while alist
			(let ((value (car (car alist)))
			      (ans (cdr (car alist))))
			  (if (member top-name ans)
			      ;; This annotation is listed, but still have to
			      ;; check if multiple annotations are satisfied
			      (if (member nil (mapcar (lambda (r)
							(assoc r open-ans))
						      ans))
				  nil	; multiple ans not satisfied
				;; If there are multiple annotations going
				;; into one text property, split up the other
				;; annotations so they apply individually to
				;; the other regions.
				(setcdr (car top-extents) loc)
				(let ((to-split ans) this-one extents)
				  (while to-split
				    (setq this-one
					  (assoc (car to-split) open-ans)
					  extents (nth 1 this-one))
				    (if (not (eq this-one top))
					(setcar (cdr this-one)
						(format-subtract-regions
						 extents top-extents)))
				    (setq to-split (cdr to-split))))
				;; Set loop variables to nil so loop
				;; will exit.
				(setq alist nil aalist nil matched t
				      ;; pop annotation off stack.
				      open-ans (cdr open-ans))
				(let ((extents top-extents)
				      (start (car (car top-extents)))
				      (loc (cdr (car top-extents))))
				  (while extents
				    (cond
				     ;; Check for pseudo-properties
				     ((eq prop 'PARAMETER)
				      ;; A parameter of the top open ann:
				      ;; delete text and use as arg.
				      (if open-ans
					  ;; (If nothing open, discard).
					  (setq open-ans
						(cons
						 (append (car open-ans)
							 (list
							  (buffer-substring
							   start loc)))
						 (cdr open-ans))))
				      (delete-region start loc))
				     ((eq prop 'FUNCTION)
				      ;; Not a property, but a function.
				      (let ((rtn
					     (apply value start loc params)))
					(if rtn (setq todo (cons rtn todo)))))
				     (t
				      ;; Normal property/value pair
				      (setq todo
					    (cons (list start loc prop value)
						  todo))))
				    (setq extents (cdr extents)
					  start (car (car extents))
					  loc (cdr (car extents))))))))
			(setq alist (cdr alist))))
		    (setq aalist (cdr aalist)))
		  (if (not matched)
		      ;; Didn't find any match for the annotation:
		      ;; Store as value of text-property `unknown'.
		      (let ((extents top-extents)
			    (start (car (car top-extents)))
			    (loc (or (cdr (car top-extents)) loc)))
			(while extents
			  (setq open-ans (cdr open-ans)
				todo (cons (list start loc 'unknown top-name)
					   todo)
				unknown-ans (cons name unknown-ans)
				extents (cdr extents)
				start (car (car extents))
				loc (cdr (car extents))))))))))))

	;; Once entire file has been scanned, add the properties.
	(while todo
	  (let* ((item (car todo))
		 (from (nth 0 item))
		 (to   (nth 1 item))
		 (prop (nth 2 item))
		 (val  (nth 3 item)))

	    (if (numberp val)	; add to ambient value if numeric
		(format-property-increment-region from to prop val 0)
	      (put-text-property
	       from to prop
	       (cond ((get prop 'format-list-valued) ; value gets consed onto
						     ; list-valued properties
		      (let ((prev (get-text-property from prop)))
			(cons val (if (listp prev) prev (list prev)))))
		     (t val))))) ; normally, just set to val.
	  (setq todo (cdr todo)))

	(if unknown-ans
	    (message "Unknown annotations: %s" unknown-ans))))))

(defun format-subtract-regions (minu subtra)
  "Remove from the regions in MINUend the regions in SUBTRAhend.
A region is a dotted pair (from . to).  Both parameters are lists of
regions.  Each list must contain nonoverlapping, noncontiguous
regions, in descending order.  The result is also nonoverlapping,
noncontiguous, and in descending order.  The first element of MINUEND
can have a cdr of nil, indicating that the end of that region is not
yet known."
  (let* ((minuend (copy-alist minu))
	 (subtrahend (copy-alist subtra))
	 (m (car minuend))
	 (s (car subtrahend))
	 results)
    (while (and minuend subtrahend)
      (cond
       ;; The minuend starts after the subtrahend ends; keep it.
       ((> (car m) (cdr s))
	(setq results (cons m results)
	      minuend (cdr minuend)
	      m (car minuend)))
       ;; The minuend extends beyond the end of the subtrahend.  Chop it off.
       ((or (null (cdr m)) (> (cdr m) (cdr s)))
	(setq results (cons (cons (1+ (cdr s)) (cdr m)) results))
	(setcdr m (cdr s)))
       ;; The subtrahend starts after the minuend ends; throw it away.
       ((< (cdr m) (car s))
	(setq subtrahend (cdr subtrahend) s (car subtrahend)))
       ;; The subtrahend extends beyond the end of the minuend.  Chop it off.
       (t	;(<= (cdr m) (cdr s)))
	(if (>= (car m) (car s))
	    (setq minuend (cdr minuend) m (car minuend))
	  (setcdr m (1- (car s)))
	  (setq subtrahend (cdr subtrahend) s (car subtrahend))))))
    (nconc (nreverse results) minuend)))

;; This should probably go somewhere other than format.el.  Then again,
;; indent.el has alter-text-property.  NOTE: We can also use
;; next-single-property-change instead of text-property-not-all, but then
;; we have to see if we passed TO.
(defun format-property-increment-region (from to prop delta default)
  "Over the region between FROM and TO increment property PROP by amount DELTA.
DELTA may be negative.  If property PROP is nil anywhere
in the region, it is treated as though it were DEFAULT."
  (let ((cur from) val newval next)
    (while cur
      (setq val    (get-text-property cur prop)
	    newval (+ (or val default) delta)
	    next   (text-property-not-all cur to prop val))
      (put-text-property cur (or next to) prop newval)
      (setq cur next))))

;;;
;;; Encoding
;;;

(defun format-insert-annotations (list &optional offset)
  "Apply list of annotations to buffer as `write-region' would.
Inserts each element of the given LIST of buffer annotations at its
appropriate place.  Use second arg OFFSET if the annotations' locations are
not relative to the beginning of the buffer: annotations will be inserted
at their location-OFFSET+1 \(ie, the offset is treated as the character number
of the first character in the buffer)."
  (if (not offset)
      (setq offset 0)
    (setq offset (1- offset)))
  (let ((l (reverse list)))
    (while l
      (goto-char (- (car (car l)) offset))
      (insert (cdr (car l)))
      (setq l (cdr l)))))

(defun format-annotate-value (old new)
  "Return OLD and NEW as a \(close . open) annotation pair.
Useful as a default function for TRANSLATIONS alist when the value of the text
property is the name of the annotation that you want to use, as it is for the
`unknown' text property."
  (cons (if old (list old))
	(if new (list new))))

(defun format-annotate-region (from to translations format-fn ignore)
  "Generate annotations for text properties in the region.
Searches for changes between FROM and TO, and describes them with a list of
annotations as defined by alist TRANSLATIONS and FORMAT-FN.  IGNORE lists text
properties not to consider; any text properties that are neither ignored nor
listed in TRANSLATIONS are warned about.
If you actually want to modify the region, give the return value of this
function to `format-insert-annotations'.

Format of the TRANSLATIONS argument:

Each element is a list whose car is a PROPERTY, and the following
elements are VALUES of that property followed by the names of zero or more
ANNOTATIONS.  Whenever the property takes on that value, the annotations
\(as formatted by FORMAT-FN) are inserted into the file.
When the property stops having that value, the matching negated annotation
will be inserted \(it may actually be closed earlier and reopened, if
necessary, to keep proper nesting).

If the property's value is a list, then each element of the list is dealt with
separately.

If a VALUE is numeric, then it is assumed that there is a single annotation
and each occurrence of it increments the value of the property by that number.
Thus, given the entry \(left-margin \(4 \"indent\")), if the left margin
changes from 4 to 12, two <indent> annotations will be generated.

If the VALUE is nil, then instead of annotations, a function should be
specified.  This function is used as a default: it is called for all
transitions not explicitly listed in the table.  The function is called with
two arguments, the OLD and NEW values of the property.  It should return
lists of annotations like `format-annotate-location' does.

    The same structure can be used in reverse for reading files."
  (let ((all-ans nil)    ; All annotations - becomes return value
	(open-ans nil)   ; Annotations not yet closed
	(loc nil)	 ; Current location
	(not-found nil)) ; Properties that couldn't be saved
    (while (or (null loc)
	       (and (setq loc (next-property-change loc nil to))
		    (< loc to)))
      (or loc (setq loc from))
      (let* ((ans (format-annotate-location loc (= loc from) ignore translations))
	     (neg-ans (format-reorder (aref ans 0) open-ans))
	     (pos-ans (aref ans 1))
	     (ignored (aref ans 2)))
	(setq not-found (append ignored not-found)
	      ignore    (append ignored ignore))
	;; First do the negative (closing) annotations
	(while neg-ans
	  ;; Check if it's missing.  This can happen (eg, a numeric property
	  ;; going negative can generate closing annotations before there are
	  ;; any open).  Warn user & ignore.
	  (if (not (member (car neg-ans) open-ans))
	      (message "Can't close %s: not open." (car neg-ans))
	    (while (not (equal (car neg-ans) (car open-ans)))
	      ;; To close anno. N, need to first close ans 1 to N-1,
	      ;; remembering to re-open them later.
	      (setq pos-ans (cons (car open-ans) pos-ans))
	      (setq all-ans
		    (cons (cons loc (funcall format-fn (car open-ans) nil))
			  all-ans))
	      (setq open-ans (cdr open-ans)))
	    ;; Now remove the one we're really interested in from open list.
	    (setq open-ans (cdr open-ans))
	    ;; And put the closing annotation here.
	    (setq all-ans
		  (cons (cons loc (funcall format-fn (car neg-ans) nil))
			all-ans)))
	  (setq neg-ans (cdr neg-ans)))
	;; Now deal with positive (opening) annotations
	(let ((p pos-ans))
	  (while pos-ans
	    (setq open-ans (cons (car pos-ans) open-ans))
	    (setq all-ans
		  (cons (cons loc (funcall format-fn (car pos-ans) t))
			all-ans))
	    (setq pos-ans (cdr pos-ans))))))

    ;; Close any annotations still open
    (while open-ans
      (setq all-ans
	    (cons (cons to (funcall format-fn (car open-ans) nil))
		  all-ans))
      (setq open-ans (cdr open-ans)))
    (if not-found
	(message "These text properties could not be saved:\n    %s"
		 not-found))
    (nreverse all-ans)))

;;; Internal functions for format-annotate-region.

(defun format-annotate-location (loc all ignore translations)
  "Return annotation(s) needed at location LOC.
This includes any properties that change between LOC-1 and LOC.
If ALL is true, don't look at previous location, but generate annotations for
all non-nil properties.
Third argument IGNORE is a list of text-properties not to consider.
Use the TRANSLATIONS alist.

Return value is a vector of 3 elements:
1. List of names of the annotations to close
2. List of the names of annotations to open.
3. List of properties that were ignored or couldn't be annotated."
  (let* ((prev-loc (1- loc))
	 (before-plist (if all nil (text-properties-at prev-loc)))
	 (after-plist (text-properties-at loc))
	 p negatives positives prop props not-found)
    ;; make list of all property names involved
    (setq p before-plist)
    (while p
      (if (not (memq (car p) props))
	  (setq props (cons (car p) props)))
      (setq p (cdr (cdr p))))
    (setq p after-plist)
    (while p
      (if (not (memq (car p) props))
	  (setq props (cons (car p) props)))
      (setq p (cdr (cdr p))))

    (while props
      (setq prop (car props)
	    props (cdr props))
      (if (memq prop ignore)
	  nil  ; If it's been ignored before, ignore it now.
	(let ((before (if all nil (car (cdr (memq prop before-plist)))))
	      (after (car (cdr (memq prop after-plist)))))
	  (if (equal before after)
	      nil ; no change; ignore
	    (let ((result (format-annotate-single-property-change
			   prop before after translations)))
	      (if (not result)
		  (setq not-found (cons prop not-found))
		(setq negatives (nconc negatives (car result))
		      positives (nconc positives (cdr result)))))))))
    (vector negatives positives not-found)))

(defun format-annotate-single-property-change (prop old new trans)
  "Return annotations for property PROP changing from OLD to NEW.
These are searched for in the translations alist TRANS.
If NEW does not appear in the list, but there is a default function, then that
function is called.
Annotations to open and to close are returned as a dotted pair."
  (let ((prop-alist (cdr (assoc prop trans)))
	default)
    (if (not prop-alist)
	nil
      ;; If either old or new is a list, have to treat both that way.
      (if (and (or (listp old) (listp new))
	       (not (get prop 'format-list-atomic-p)))
	  (if (or (not (format-proper-list-p old))
		  (not (format-proper-list-p new)))
	      (format-annotate-atomic-property-change prop-alist old new)
	    (let* ((old (if (listp old) old (list old)))
		   (new (if (listp new) new (list new)))
		   (tail (format-common-tail old new))
		   close open)
	      (while old
		(setq close
		      (append (car (format-annotate-atomic-property-change
				    prop-alist (car old) nil))
			      close)
		      old (cdr old)))
	      (while new
		(setq open
		      (append (cdr (format-annotate-atomic-property-change
				    prop-alist nil (car new)))
			      open)
		      new (cdr new)))
	      (format-make-relatively-unique close open)))
	(format-annotate-atomic-property-change prop-alist old new)))))

(defun format-annotate-atomic-property-change (prop-alist old new)
  "Internal function annotate a single property change.
PROP-ALIST is the relevant segment of a TRANSLATIONS list.
OLD and NEW are the values."
  (let (num-ann)
    ;; If old and new values are numbers,
    ;; look for a number in PROP-ALIST.
    (if (and (or (null old) (numberp old))
	     (or (null new) (numberp new)))
	(progn
	  (setq num-ann prop-alist)
	  (while (and num-ann (not (numberp (car (car num-ann)))))
	    (setq num-ann (cdr num-ann)))))
    (if num-ann
	;; Numerical annotation - use difference
	(progn
	  ;; If property is numeric, nil means 0
	  (cond ((and (numberp old) (null new))
		 (setq new 0))
		((and (numberp new) (null old))
		 (setq old 0)))

	  (let* ((entry (car num-ann))
		 (increment (car entry))
		 (n (ceiling (/ (float (- new old)) (float increment))))
		 (anno (car (cdr entry))))
	    (if (> n 0)
		(cons nil (make-list n anno))
	      (cons (make-list (- n) anno) nil))))

      ;; Standard annotation
      (let ((close (and old (cdr (assoc old prop-alist))))
	    (open  (and new (cdr (assoc new prop-alist)))))
	(if (or close open)
	    (format-make-relatively-unique close open)
	  ;; Call "Default" function, if any
	  (let ((default (assq nil prop-alist)))
	    (if default
		(funcall (car (cdr default)) old new))))))))

(provide 'format)

;;; format.el ends here