Mercurial > emacs
annotate lisp/mail/mail-extr.el @ 1156:2a92ddfaf6ba
entered into RCS
author | Roland McGrath <roland@gnu.org> |
---|---|
date | Thu, 17 Sep 1992 01:12:50 +0000 |
parents | 9f3cc03dae67 |
children | 24f248525608 |
rev | line source |
---|---|
809 | 1 ;;; mail-extr.el --- extract full name and address from RFC 822 mail header. |
2 | |
846
20674ae6bf52
*** empty log message ***
Eric S. Raymond <esr@snark.thyrsus.com>
parents:
809
diff
changeset
|
3 ;; Copyright (C) 1992 Free Software Foundation, Inc. |
20674ae6bf52
*** empty log message ***
Eric S. Raymond <esr@snark.thyrsus.com>
parents:
809
diff
changeset
|
4 |
809 | 5 ;; Author: Joe Wells <jbw@cs.bu.edu> |
6 ;; Version: 1.0 | |
7 ;; Adapted-By: ESR | |
8 ;; Keywords: mail | |
9 | |
10 ;; This file is part of GNU Emacs. | |
11 | |
12 ;; GNU Emacs is free software; you can redistribute it and/or modify | |
13 ;; it under the terms of the GNU General Public License as published by | |
14 ;; the Free Software Foundation; either version 1, or (at your option) | |
15 ;; any later version. | |
16 | |
17 ;; GNU Emacs is distributed in the hope that it will be useful, | |
18 ;; but WITHOUT ANY WARRANTY; without even the implied warranty of | |
19 ;; MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the | |
20 ;; GNU General Public License for more details. | |
21 | |
22 ;; You should have received a copy of the GNU General Public License | |
23 ;; along with GNU Emacs; see the file COPYING. If not, write to | |
24 ;; the Free Software Foundation, 675 Mass Ave, Cambridge, MA 02139, USA. | |
25 | |
26 ;;; Commentary: | |
27 | |
28 ;; Here is `mail-extr', a package for extracting full names and canonical | |
29 ;; addresses from RFC 822 mail headers. It is intended to be hooked into | |
30 ;; other Emacs Lisp packages that deal with RFC 822 format messages, such as | |
31 ;; Gnews, GNUS, RMAIL, MH-E, BBDB, VM, Supercite, etc. Thus, this release is | |
32 ;; mainly for Emacs Lisp developers. | |
33 | |
34 ;; There are two main benefits: | |
35 | |
36 ;; 1. Higher probability of getting the correct full name for a human than | |
37 ;; any other package I know of. (On the other hand, it will cheerfully | |
38 ;; mangle non-human names/comments.) | |
39 ;; 2. Address part is put in a canonical form. | |
40 | |
41 ;; The interface is not yet carved in stone; please give me suggestions. | |
42 | |
43 ;; I have an extensive test-case collection of funny addresses if you want to | |
44 ;; work with the code. Developing this code requires frequent testing to | |
45 ;; make sure you're not breaking functionality. I'm not posting the | |
46 ;; test-cases because they take over 100K. | |
47 | |
48 ;; If you find an address that mail-extr fails on, please send it to me along | |
49 ;; with what you think the correct results should be. I do not consider it a | |
50 ;; bug if mail-extr mangles a comment that does not correspond to a real | |
51 ;; human full name, although I would prefer that mail-extr would return the | |
52 ;; comment as-is. | |
53 | |
54 ;; Features: | |
55 | |
56 ;; * Full name handling: | |
57 | |
58 ;; * knows where full names can be found in an address. | |
59 ;; * avoids using empty comments and quoted text. | |
60 ;; * extracts full names from mailbox names. | |
61 ;; * recognizes common formats for comments after a full name. | |
62 ;; * puts a period and a space after each initial. | |
63 ;; * understands & referring to the mailbox name capitalized. | |
64 ;; * strips name prefixes like "Prof.", etc.. | |
65 ;; * understands what characters can occur in names (not just letters). | |
66 ;; * figures out middle initial from mailbox name. | |
67 ;; * removes funny nicknames. | |
68 ;; * keeps suffixes such as Jr., Sr., III, etc. | |
69 ;; * reorders "Last, First" type names. | |
70 | |
71 ;; * Address handling: | |
72 | |
73 ;; * parses rfc822 quoted text, comments, and domain literals. | |
74 ;; * parses rfc822 multi-line headers. | |
75 ;; * does something reasonable with rfc822 GROUP addresses. | |
76 ;; * handles many rfc822 noncompliant and garbage addresses. | |
77 ;; * canonicalizes addresses (after stripping comments/phrases outside <>). | |
78 ;; * converts ! addresses into .UUCP and %-style addresses. | |
79 ;; * converts rfc822 ROUTE addresses to %-style addresses. | |
80 ;; * truncates %-style addresses at leftmost fully qualified domain name. | |
81 ;; * handles local relative precedence of ! vs. % and @ (untested). | |
82 | |
83 ;; It does almost no string creation. It primarily uses the built-in | |
84 ;; parsing routines with the appropriate syntax tables. This should | |
85 ;; result in greater speed. | |
86 | |
87 ;; TODO: | |
88 | |
89 ;; * handle all test cases. (This will take forever.) | |
90 ;; * software to pick the correct header to use (eg., "Senders-Name:"). | |
91 ;; * multiple addresses in the "From:" header (almost all of the necessary | |
92 ;; code is there). | |
93 ;; * flag to not treat `,' as an address separator. (This is useful when | |
94 ;; there is a "From:" header but no "Sender:" header, because then there | |
95 ;; is only allowed to be one address.) | |
96 ;; * mailbox name does not necessarily contain full name. | |
97 ;; * fixing capitalization when it's all upper or lowercase. (Hard!) | |
98 ;; * some of the domain literal handling is missing. (But I've never even | |
99 ;; seen one of these in a mail address, so maybe no big deal.) | |
100 ;; * arrange to have syntax tables byte-compiled. | |
101 ;; * speed hacks. | |
102 ;; * delete unused variables. | |
103 ;; * arrange for testing with different relative precedences of ! vs. @ | |
104 ;; and %. | |
105 ;; * put variant-method back into mail-extract-address-components. | |
106 ;; * insert documentation strings! | |
107 ;; * handle X.400-gatewayed addresses according to RFC 1148. | |
108 | |
109 ;;; Change Log: | |
110 ;; | |
111 ;; Mon Apr 6 23:59:09 1992 Joe Wells (jbw at bigbird.bu.edu) | |
112 ;; | |
113 ;; * Cleaned up some more. Release version 1.0 to world. | |
114 ;; | |
115 ;; Sun Apr 5 19:39:08 1992 Joe Wells (jbw at bigbird.bu.edu) | |
116 ;; | |
117 ;; * Cleaned up full name extraction extensively. | |
118 ;; | |
119 ;; Sun Feb 2 14:45:24 1992 Joe Wells (jbw at bigbird.bu.edu) | |
120 ;; | |
121 ;; * Total rewrite. Integrated mail-canonicalize-address into | |
122 ;; mail-extract-address-components. Now handles GROUP addresses more | |
123 ;; or less correctly. Better handling of lots of different cases. | |
124 ;; | |
125 ;; Fri Jun 14 19:39:50 1991 | |
126 ;; * Created. | |
127 | |
128 ;;; Code: | |
129 | |
130 ;; Variable definitions. | |
131 | |
132 (defvar mail-@-binds-tighter-than-! nil) | |
133 | |
134 ;;---------------------------------------------------------------------- | |
135 ;; what orderings are meaningful????? | |
136 ;;(defvar mail-operator-precedence-list '(?! ?% ?@)) | |
137 ;; Right operand of a % or a @ must be a domain name, period. No other | |
138 ;; operators allowed. Left operand of a @ is an address relative to that | |
139 ;; site. | |
140 | |
141 ;; Left operand of a ! must be a domain name. Right operand is an | |
142 ;; arbitrary address. | |
143 ;;---------------------------------------------------------------------- | |
144 | |
145 (defconst mail-space-char 32) | |
146 | |
147 (defconst mail-whitespace " \t\n") | |
148 | |
149 ;; Any character that can occur in a name in an RFC822 address. | |
150 ;; Yes, there are weird people with digits in their names. | |
151 (defconst mail-all-letters "A-Za-z---{|}'~0-9`.") | |
152 | |
153 ;; Any character that can occur in a name, not counting characters that | |
154 ;; separate parts of a multipart name. | |
155 (defconst mail-all-letters-but-separators "A-Za-z{|}'~0-9`") | |
156 | |
157 ;; Any character that can start a name | |
158 (defconst mail-first-letters "A-Za-z") | |
159 | |
160 ;; Any character that can end a name. | |
161 (defconst mail-last-letters "A-Za-z`'.") | |
162 | |
163 ;; Matches an initial not followed by both a period and a space. | |
164 (defconst mail-bad-initials-pattern | |
165 (format "\\(\\([^%s]\\|\\`\\)[%s]\\)\\(\\.\\([^ ]\\)\\| \\|\\([^%s .]\\)\\|\\'\\)" | |
166 mail-all-letters mail-first-letters mail-all-letters)) | |
167 | |
168 (defconst mail-non-name-chars (concat "^" mail-all-letters ".")) | |
169 | |
170 (defconst mail-non-begin-name-chars (concat "^" mail-first-letters)) | |
171 | |
172 (defconst mail-non-end-name-chars (concat "^" mail-last-letters)) | |
173 | |
174 ;; Matches periods used instead of spaces. Must not match the period | |
175 ;; following an initial. | |
176 (defconst mail-bad-\.-pattern | |
177 (format "\\([%s][%s]\\)\\.+\\([%s]\\)" | |
178 mail-all-letters mail-last-letters mail-first-letters)) | |
179 | |
180 ;; Matches an embedded or leading nickname that should be removed. | |
181 (defconst mail-nickname-pattern | |
182 (format "\\([ .]\\|\\`\\)[\"'`\[\(]\\([ .%s]+\\)[\]\"'\)] " | |
183 mail-all-letters)) | |
184 | |
185 ;; Matches a leading title that is not part of the name (does not | |
186 ;; contribute to uniquely identifying the person). | |
187 (defconst mail-full-name-prefixes | |
188 '"\\` *\\(Prof\\|Dr\\|Mrs?\\|Rev\\|Rabbi\\|SysOp\\|LCDR\\)\\.? ") | |
189 | |
190 ;; Matches the occurrence of a generational name suffix, and the last | |
191 ;; character of the preceding name. | |
192 (defconst mail-full-name-suffix-pattern | |
193 (format | |
194 "\\(,? ?\\([JjSs]r\\.?\\|V?I+V?\\)\\)\\([^%s]\\([^%s]\\|\\'\\)\\|\\'\\)" | |
195 mail-all-letters mail-all-letters)) | |
196 | |
197 (defconst mail-roman-numeral-pattern | |
198 "V?I+V?\\b") | |
199 | |
200 ;; Matches a trailing uppercase (with other characters possible) acronym. | |
201 ;; Must not match a trailing uppercase last name or trailing initial | |
202 (defconst mail-weird-acronym-pattern "\\([A-Z]+[-_/]\\|[A-Z][A-Z][A-Z]?\\b\\)") | |
203 | |
204 ;; Matches a mixed-case or lowercase name (not an initial). | |
205 (defconst mail-mixed-case-name-pattern | |
206 (format | |
207 "\\b\\([a-z][%s]*[%s]\\|[%s][%s]*[a-z][%s]*[%s]\\|[%s][%s]*[a-z]\\)" | |
208 mail-all-letters mail-last-letters | |
209 mail-first-letters mail-all-letters mail-all-letters mail-last-letters | |
210 mail-first-letters mail-all-letters)) | |
211 | |
212 ;; Matches a trailing alternative address. | |
213 (defconst mail-alternative-address-pattern "[a-zA-Z.]+[!@][a-zA-Z.]") | |
214 | |
215 ;; Matches a variety of trailing comments not including comma-delimited | |
216 ;; comments. | |
217 (defconst mail-trailing-comment-start-pattern " [-{]\\|--\\|[+@#></\;]") | |
218 | |
219 ;; Matches a name (not an initial). | |
220 ;; This doesn't force a word boundary at the end because sometimes a | |
221 ;; comment is separated by a `-' with no preceding space. | |
222 (defconst mail-name-pattern | |
223 (format | |
224 "\\b[%s][%s]*[%s]" | |
225 mail-first-letters mail-all-letters mail-last-letters)) | |
226 | |
227 (defconst mail-initial-pattern | |
228 (format "\\b[%s]\\([. ]\\|\\b\\)" mail-first-letters)) | |
229 | |
230 ;; Matches a single name before a comma. | |
231 (defconst mail-last-name-first-pattern | |
232 (concat "\\`" mail-name-pattern ",")) | |
233 | |
234 ;; Matches telephone extensions. | |
235 (defconst mail-telephone-extension-pattern | |
236 "\\(\\([Ee]xt\\|[Tt]el\\|[Xx]\\).?\\)? *\\+?[0-9][- 0-9]+") | |
237 | |
238 ;; Matches ham radio call signs. | |
239 (defconst mail-ham-call-sign-pattern | |
240 "\\b[A-Z]+[0-9][A-Z0-9]*") | |
241 | |
242 ;; Matches normal single-part name | |
243 (defconst mail-normal-name-pattern | |
244 (format | |
245 "\\b[%s][%s]+[%s]" | |
246 mail-first-letters mail-all-letters-but-separators mail-last-letters)) | |
247 | |
248 ;; Matches normal two names with missing middle initial | |
249 (defconst mail-two-name-pattern | |
250 (concat "\\`\\(" mail-normal-name-pattern | |
251 "\\|" mail-initial-pattern | |
252 "\\) +\\(" mail-normal-name-pattern "\\)\\(,\\|\\'\\)")) | |
253 | |
254 (defvar address-syntax-table (make-syntax-table)) | |
255 (defvar address-comment-syntax-table (make-syntax-table)) | |
256 (defvar address-domain-literal-syntax-table (make-syntax-table)) | |
257 (defvar address-text-comment-syntax-table (make-syntax-table)) | |
258 (defvar address-text-syntax-table (make-syntax-table)) | |
259 (mapcar | |
260 (function | |
261 (lambda (pair) | |
262 (let ((syntax-table (symbol-value (car pair)))) | |
263 (mapcar | |
264 (function | |
265 (lambda (item) | |
266 (if (eq 2 (length item)) | |
267 (modify-syntax-entry (car item) (car (cdr item)) syntax-table) | |
268 (let ((char (car item)) | |
269 (bound (car (cdr item))) | |
270 (syntax (car (cdr (cdr item))))) | |
271 (while (<= char bound) | |
272 (modify-syntax-entry char syntax syntax-table) | |
273 (setq char (1+ char))))))) | |
274 (cdr pair))))) | |
275 '((address-syntax-table | |
276 (0 31 "w") ;control characters | |
277 (32 " ") ;SPC | |
278 (?! ?~ "w") ;printable characters | |
279 (127 "w") ;DEL | |
280 (128 255 "w") ;high-bit-on characters | |
281 (?\t " ") | |
282 (?\r " ") | |
283 (?\n " ") | |
284 (?\( ".") | |
285 (?\) ".") | |
286 (?< ".") | |
287 (?> ".") | |
288 (?@ ".") | |
289 (?, ".") | |
290 (?\; ".") | |
291 (?: ".") | |
292 (?\\ "\\") | |
293 (?\" "\"") | |
294 (?. ".") | |
295 (?\[ ".") | |
296 (?\] ".") | |
297 ;; % and ! aren't RFC822 characters, but it is convenient to pretend | |
298 (?% ".") | |
299 (?! ".") | |
300 ) | |
301 (address-comment-syntax-table | |
302 (0 255 "w") | |
303 (?\( "\(\)") | |
304 (?\) "\)\(") | |
305 (?\\ "\\")) | |
306 (address-domain-literal-syntax-table | |
307 (0 255 "w") | |
308 (?\[ "\(\]") ;?????? | |
309 (?\] "\)\[") ;?????? | |
310 (?\\ "\\")) | |
311 (address-text-comment-syntax-table | |
312 (0 255 "w") | |
313 (?\( "\(\)") | |
314 (?\) "\)\(") | |
315 (?\[ "\(\]") | |
316 (?\] "\)\[") | |
317 (?\{ "\(\}") | |
318 (?\} "\)\{") | |
319 (?\\ "\\") | |
320 (?\" "\"") | |
321 ;; (?\' "\)\`") | |
322 ;; (?\` "\(\'") | |
323 ) | |
324 (address-text-syntax-table | |
325 (0 255 ".") | |
326 (?A ?Z "w") | |
327 (?a ?z "w") | |
328 (?- "w") | |
329 (?\} "w") | |
330 (?\{ "w") | |
331 (?| "w") | |
332 (?\' "w") | |
333 (?~ "w") | |
334 (?0 ?9 "w")) | |
335 )) | |
336 | |
337 | |
338 ;; Utility functions and macros. | |
339 | |
340 (defmacro undo-backslash-quoting (beg end) | |
341 (`(save-excursion | |
342 (save-restriction | |
343 (narrow-to-region (, beg) (, end)) | |
344 (goto-char (point-min)) | |
345 ;; undo \ quoting | |
346 (while (re-search-forward "\\\\\\(.\\)" nil t) | |
347 (replace-match "\\1") | |
348 ;; CHECK: does this leave point after the replacement? | |
349 ))))) | |
350 | |
351 (defmacro mail-nuke-char-at (pos) | |
352 (` (save-excursion | |
353 (goto-char (, pos)) | |
354 (delete-char 1) | |
355 (insert mail-space-char)))) | |
356 | |
357 (defmacro mail-nuke-elements-outside-range (list-symbol beg-symbol end-symbol | |
358 &optional no-replace) | |
359 (` (progn | |
360 (setq temp (, list-symbol)) | |
361 (while temp | |
362 (cond ((or (> (car temp) (, end-symbol)) | |
363 (< (car temp) (, beg-symbol))) | |
364 (, (or no-replace | |
365 (` (mail-nuke-char-at (car temp))))) | |
366 (setcar temp nil))) | |
367 (setq temp (cdr temp))) | |
368 (setq (, list-symbol) (delq nil (, list-symbol)))))) | |
369 | |
370 (defun mail-demarkerize (marker) | |
371 (and marker | |
372 (if (markerp marker) | |
373 (let ((temp (marker-position marker))) | |
374 (set-marker marker nil) | |
375 temp) | |
376 marker))) | |
377 | |
378 (defun mail-markerize (pos) | |
379 (and pos | |
380 (if (markerp pos) | |
381 pos | |
382 (copy-marker pos)))) | |
383 | |
384 (defmacro mail-last-element (list) | |
385 "Return last element of LIST." | |
386 (` (let ((list (, list))) | |
387 (while (not (null (cdr list))) | |
388 (setq list (cdr list))) | |
389 (car list)))) | |
390 | |
391 (defmacro safe-move-sexp (arg) | |
392 "Safely skip over one balanced sexp, if there is one. Return t if success." | |
393 (` (condition-case error | |
394 (progn | |
395 (goto-char (scan-sexps (point) (, arg))) | |
396 t) | |
397 (error | |
398 (if (string-equal (nth 1 error) "Unbalanced parentheses") | |
399 nil | |
400 (while t | |
401 (signal (car error) (cdr error)))))))) | |
402 | |
403 | |
404 ;; The main function to grind addresses | |
405 | |
406 (defun mail-extract-address-components (address) | |
407 "Given an rfc 822 ADDRESS, extract full name and canonical address. | |
408 Returns a list of the form (FULL-NAME CANONICAL-ADDRESS)." | |
409 (let ((canonicalization-buffer (get-buffer-create "*canonical address*")) | |
410 (extraction-buffer (get-buffer-create "*extract address components*")) | |
411 (foo 'bar) | |
412 char | |
413 multiple-addresses | |
414 <-pos >-pos @-pos :-pos ,-pos !-pos %-pos \;-pos | |
415 group-:-pos group-\;-pos route-addr-:-pos | |
416 record-pos-symbol | |
417 first-real-pos last-real-pos | |
418 phrase-beg phrase-end | |
419 comment-beg comment-end | |
420 quote-beg quote-end | |
421 atom-beg atom-end | |
422 mbox-beg mbox-end | |
423 \.-ends-name | |
424 temp | |
425 name-suffix | |
426 saved-point | |
427 fi mi li | |
428 saved-%-pos saved-!-pos saved-@-pos | |
429 domain-pos \.-pos insert-point) | |
430 | |
431 (save-excursion | |
432 (set-buffer extraction-buffer) | |
923 | 433 (buffer-disable-undo extraction-buffer) |
809 | 434 (set-syntax-table address-syntax-table) |
435 (widen) | |
436 (erase-buffer) | |
437 (setq case-fold-search nil) | |
438 | |
439 ;; Insert extra space at beginning to allow later replacement with < | |
440 ;; without having to move markers. | |
441 (insert mail-space-char address) | |
442 | |
443 ;; stolen from rfc822.el | |
444 ;; Unfold multiple lines. | |
445 (goto-char (point-min)) | |
446 (while (re-search-forward "\\([^\\]\\(\\\\\\\\\\)*\\)\n[ \t]" nil t) | |
447 (replace-match "\\1 " t)) | |
448 | |
449 ;; first pass grabs useful information about address | |
450 (goto-char (point-min)) | |
451 (while (progn | |
452 (skip-chars-forward mail-whitespace) | |
453 (not (eobp))) | |
454 (setq char (char-after (point))) | |
455 (or first-real-pos | |
456 (if (not (eq char ?\()) | |
457 (setq first-real-pos (point)))) | |
458 (cond | |
459 ;; comment | |
460 ((eq char ?\() | |
461 (set-syntax-table address-comment-syntax-table) | |
462 ;; only record the first non-empty comment's position | |
463 (if (and (not comment-beg) | |
464 (save-excursion | |
465 (forward-char 1) | |
466 (skip-chars-forward mail-whitespace) | |
467 (not (eq ?\) (char-after (point)))))) | |
468 (setq comment-beg (point))) | |
469 ;; TODO: don't record if unbalanced | |
470 (or (safe-move-sexp 1) | |
471 (forward-char 1)) | |
472 (set-syntax-table address-syntax-table) | |
473 (if (and comment-beg | |
474 (not comment-end)) | |
475 (setq comment-end (point)))) | |
476 ;; quoted text | |
477 ((eq char ?\") | |
478 ;; only record the first non-empty quote's position | |
479 (if (and (not quote-beg) | |
480 (save-excursion | |
481 (forward-char 1) | |
482 (skip-chars-forward mail-whitespace) | |
483 (not (eq ?\" (char-after (point)))))) | |
484 (setq quote-beg (point))) | |
485 ;; TODO: don't record if unbalanced | |
486 (or (safe-move-sexp 1) | |
487 (forward-char 1)) | |
488 (if (and quote-beg | |
489 (not quote-end)) | |
490 (setq quote-end (point)))) | |
491 ;; domain literals | |
492 ((eq char ?\[) | |
493 (set-syntax-table address-domain-literal-syntax-table) | |
494 (or (safe-move-sexp 1) | |
495 (forward-char 1)) | |
496 (set-syntax-table address-syntax-table)) | |
497 ;; commas delimit addresses when outside < > pairs. | |
498 ((and (eq char ?,) | |
499 (or (null <-pos) | |
500 (and >-pos | |
501 ;; handle weird munged addresses | |
502 (> (mail-last-element <-pos) (car >-pos))))) | |
503 (setq multiple-addresses t) | |
504 (delete-char 1) | |
505 (narrow-to-region (point-min) (point))) | |
506 ;; record the position of various interesting chars, determine | |
507 ;; legality later. | |
508 ((setq record-pos-symbol | |
509 (cdr (assq char | |
510 '((?< . <-pos) (?> . >-pos) (?@ . @-pos) | |
511 (?: . :-pos) (?, . ,-pos) (?! . !-pos) | |
512 (?% . %-pos) (?\; . \;-pos))))) | |
513 (set record-pos-symbol | |
514 (cons (point) (symbol-value record-pos-symbol))) | |
515 (forward-char 1)) | |
516 ((eq char ?.) | |
517 (forward-char 1)) | |
518 ((memq char '( | |
519 ;; comment terminator illegal | |
520 ?\) | |
521 ;; domain literal terminator illegal | |
522 ?\] | |
523 ;; \ allowed only within quoted strings, | |
524 ;; domain literals, and comments | |
525 ?\\ | |
526 )) | |
527 (mail-nuke-char-at (point)) | |
528 (forward-char 1)) | |
529 (t | |
530 (forward-word 1))) | |
531 (or (eq char ?\() | |
532 (setq last-real-pos (point)))) | |
533 | |
534 ;; Use only the leftmost <, if any. Replace all others with spaces. | |
535 (while (cdr <-pos) | |
536 (mail-nuke-char-at (car <-pos)) | |
537 (setq <-pos (cdr <-pos))) | |
538 | |
539 ;; Use only the rightmost >, if any. Replace all others with spaces. | |
540 (while (cdr >-pos) | |
541 (mail-nuke-char-at (nth 1 >-pos)) | |
542 (setcdr >-pos (nthcdr 2 >-pos))) | |
543 | |
544 ;; If multiple @s and a :, but no < and >, insert around buffer. | |
545 ;; This commonly happens on the UUCP "From " line. Ugh. | |
546 (cond ((and (> (length @-pos) 1) | |
547 :-pos ;TODO: check if between @s | |
548 (not <-pos)) | |
549 (goto-char (point-min)) | |
550 (delete-char 1) | |
551 (setq <-pos (list (point))) | |
552 (insert ?<))) | |
553 | |
554 ;; If < but no >, insert > in rightmost possible position | |
555 (cond ((and <-pos | |
556 (null >-pos)) | |
557 (goto-char (point-max)) | |
558 (setq >-pos (list (point))) | |
559 (insert ?>))) | |
560 | |
561 ;; If > but no <, replace > with space. | |
562 (cond ((and >-pos | |
563 (null <-pos)) | |
564 (mail-nuke-char-at (car >-pos)) | |
565 (setq >-pos nil))) | |
566 | |
567 ;; Turn >-pos and <-pos into non-lists | |
568 (setq >-pos (car >-pos) | |
569 <-pos (car <-pos)) | |
570 | |
571 ;; Trim other punctuation lists of items outside < > pair to handle | |
572 ;; stupid MTAs. | |
573 (cond (<-pos ; don't need to check >-pos also | |
574 ;; handle bozo software that violates RFC 822 by sticking | |
575 ;; punctuation marks outside of a < > pair | |
576 (mail-nuke-elements-outside-range @-pos <-pos >-pos t) | |
577 ;; RFC 822 says nothing about these two outside < >, but | |
578 ;; remove those positions from the lists to make things | |
579 ;; easier. | |
580 (mail-nuke-elements-outside-range !-pos <-pos >-pos t) | |
581 (mail-nuke-elements-outside-range %-pos <-pos >-pos t))) | |
582 | |
583 ;; Check for : that indicates GROUP list and for : part of | |
584 ;; ROUTE-ADDR spec. | |
585 ;; Can't possibly be more than two :. Nuke any extra. | |
586 (while :-pos | |
587 (setq temp (car :-pos) | |
588 :-pos (cdr :-pos)) | |
589 (cond ((and <-pos >-pos | |
590 (> temp <-pos) | |
591 (< temp >-pos)) | |
592 (if (or route-addr-:-pos | |
593 (< (length @-pos) 2) | |
594 (> temp (car @-pos)) | |
595 (< temp (nth 1 @-pos))) | |
596 (mail-nuke-char-at temp) | |
597 (setq route-addr-:-pos temp))) | |
598 ((or (not <-pos) | |
599 (and <-pos | |
600 (< temp <-pos))) | |
601 (setq group-:-pos temp)))) | |
602 | |
603 ;; Nuke any ; that is in or to the left of a < > pair or to the left | |
604 ;; of a GROUP starting :. Also, there may only be one ;. | |
605 (while \;-pos | |
606 (setq temp (car \;-pos) | |
607 \;-pos (cdr \;-pos)) | |
608 (cond ((and <-pos >-pos | |
609 (> temp <-pos) | |
610 (< temp >-pos)) | |
611 (mail-nuke-char-at temp)) | |
612 ((and (or (not group-:-pos) | |
613 (> temp group-:-pos)) | |
614 (not group-\;-pos)) | |
615 (setq group-\;-pos temp)))) | |
616 | |
617 ;; Handle junk like ";@host.company.dom" that sendmail adds. | |
618 ;; **** should I remember comment positions? | |
619 (and group-\;-pos | |
620 ;; this is fine for now | |
621 (mail-nuke-elements-outside-range !-pos group-:-pos group-\;-pos t) | |
622 (mail-nuke-elements-outside-range @-pos group-:-pos group-\;-pos t) | |
623 (mail-nuke-elements-outside-range %-pos group-:-pos group-\;-pos t) | |
624 (mail-nuke-elements-outside-range ,-pos group-:-pos group-\;-pos t) | |
625 (and last-real-pos | |
626 (> last-real-pos (1+ group-\;-pos)) | |
627 (setq last-real-pos (1+ group-\;-pos))) | |
628 (and comment-end | |
629 (> comment-end group-\;-pos) | |
630 (setq comment-end nil | |
631 comment-beg nil)) | |
632 (and quote-end | |
633 (> quote-end group-\;-pos) | |
634 (setq quote-end nil | |
635 quote-beg nil)) | |
636 (narrow-to-region (point-min) group-\;-pos)) | |
637 | |
638 ;; Any commas must be between < and : of ROUTE-ADDR. Nuke any | |
639 ;; others. | |
640 ;; Hell, go ahead an nuke all of the commas. | |
641 ;; **** This will cause problems when we start handling commas in | |
642 ;; the PHRASE part .... no it won't ... yes it will ... ????? | |
643 (mail-nuke-elements-outside-range ,-pos 1 1) | |
644 | |
645 ;; can only have multiple @s inside < >. The fact that some MTAs | |
646 ;; put de-bracketed ROUTE-ADDRs in the UUCP-style "From " line is | |
647 ;; handled above. | |
648 | |
649 ;; Locate PHRASE part of ROUTE-ADDR. | |
650 (cond (<-pos | |
651 (goto-char <-pos) | |
652 (skip-chars-backward mail-whitespace) | |
653 (setq phrase-end (point)) | |
654 (goto-char (or ;;group-:-pos | |
655 (point-min))) | |
656 (skip-chars-forward mail-whitespace) | |
657 (if (< (point) phrase-end) | |
658 (setq phrase-beg (point)) | |
659 (setq phrase-end nil)))) | |
660 | |
661 ;; handle ROUTE-ADDRS with real ROUTEs. | |
662 ;; If there are multiple @s, then we assume ROUTE-ADDR syntax, and | |
663 ;; any % or ! must be semantically meaningless. | |
664 ;; TODO: do this processing into canonicalization buffer | |
665 (cond (route-addr-:-pos | |
666 (setq !-pos nil | |
667 %-pos nil | |
668 >-pos (copy-marker >-pos) | |
669 route-addr-:-pos (copy-marker route-addr-:-pos)) | |
670 (goto-char >-pos) | |
671 (insert-before-markers ?X) | |
672 (goto-char (car @-pos)) | |
673 (while (setq @-pos (cdr @-pos)) | |
674 (delete-char 1) | |
675 (setq %-pos (cons (point-marker) %-pos)) | |
676 (insert "%") | |
677 (goto-char (1- >-pos)) | |
678 (save-excursion | |
679 (insert-buffer-substring extraction-buffer | |
680 (car @-pos) route-addr-:-pos) | |
681 (delete-region (car @-pos) route-addr-:-pos)) | |
682 (or (cdr @-pos) | |
683 (setq saved-@-pos (list (point))))) | |
684 (setq @-pos saved-@-pos) | |
685 (goto-char >-pos) | |
686 (delete-char -1) | |
687 (mail-nuke-char-at route-addr-:-pos) | |
688 (mail-demarkerize route-addr-:-pos) | |
689 (setq route-addr-:-pos nil | |
690 >-pos (mail-demarkerize >-pos) | |
691 %-pos (mapcar 'mail-demarkerize %-pos)))) | |
692 | |
693 ;; de-listify @-pos | |
694 (setq @-pos (car @-pos)) | |
695 | |
696 ;; TODO: remove comments in the middle of an address | |
697 | |
698 (set-buffer canonicalization-buffer) | |
699 | |
923 | 700 (buffer-disable-undo canonicalization-buffer) |
809 | 701 (set-syntax-table address-syntax-table) |
702 (setq case-fold-search nil) | |
703 | |
704 (widen) | |
705 (erase-buffer) | |
706 (insert-buffer-substring extraction-buffer) | |
707 | |
708 (if <-pos | |
709 (narrow-to-region (progn | |
710 (goto-char (1+ <-pos)) | |
711 (skip-chars-forward mail-whitespace) | |
712 (point)) | |
713 >-pos) | |
714 ;; ****** Oh no! What if the address is completely empty! | |
715 (narrow-to-region first-real-pos last-real-pos)) | |
716 | |
717 (and @-pos %-pos | |
718 (mail-nuke-elements-outside-range %-pos (point-min) @-pos)) | |
719 (and %-pos !-pos | |
720 (mail-nuke-elements-outside-range !-pos (point-min) (car %-pos))) | |
721 (and @-pos !-pos (not %-pos) | |
722 (mail-nuke-elements-outside-range !-pos (point-min) @-pos)) | |
723 | |
724 ;; Error condition:?? (and %-pos (not @-pos)) | |
725 | |
726 (cond (!-pos | |
727 ;; **** I don't understand this save-restriction and the | |
728 ;; narrow-to-region inside it. Why did I do that? | |
729 (save-restriction | |
730 (cond ((and @-pos | |
731 mail-@-binds-tighter-than-!) | |
732 (goto-char @-pos) | |
733 (setq %-pos (cons (point) %-pos) | |
734 @-pos nil) | |
735 (delete-char 1) | |
736 (insert "%") | |
737 (setq insert-point (point-max))) | |
738 (mail-@-binds-tighter-than-! | |
739 (setq insert-point (point-max))) | |
740 (%-pos | |
741 (setq insert-point (mail-last-element %-pos) | |
742 saved-%-pos (mapcar 'mail-markerize %-pos) | |
743 %-pos nil | |
744 @-pos (mail-markerize @-pos))) | |
745 (@-pos | |
746 (setq insert-point @-pos) | |
747 (setq @-pos (mail-markerize @-pos))) | |
748 (t | |
749 (setq insert-point (point-max)))) | |
750 (narrow-to-region (point-min) insert-point) | |
751 (setq saved-!-pos (car !-pos)) | |
752 (while !-pos | |
753 (goto-char (point-max)) | |
754 (cond ((and (not @-pos) | |
755 (not (cdr !-pos))) | |
756 (setq @-pos (point)) | |
757 (insert-before-markers "@ ")) | |
758 (t | |
759 (setq %-pos (cons (point) %-pos)) | |
760 (insert-before-markers "% "))) | |
761 (backward-char 1) | |
762 (insert-buffer-substring | |
763 (current-buffer) | |
764 (if (nth 1 !-pos) | |
765 (1+ (nth 1 !-pos)) | |
766 (point-min)) | |
767 (car !-pos)) | |
768 (delete-char 1) | |
769 (or (save-excursion | |
770 (safe-move-sexp -1) | |
771 (skip-chars-backward mail-whitespace) | |
772 (eq ?. (preceding-char))) | |
773 (insert-before-markers | |
774 (if (save-excursion | |
775 (skip-chars-backward mail-whitespace) | |
776 (eq ?. (preceding-char))) | |
777 "" | |
778 ".") | |
779 "uucp")) | |
780 (setq !-pos (cdr !-pos)))) | |
781 (and saved-%-pos | |
782 (setq %-pos (append (mapcar 'mail-demarkerize saved-%-pos) | |
783 %-pos))) | |
784 (setq @-pos (mail-demarkerize @-pos)) | |
785 (narrow-to-region (1+ saved-!-pos) (point-max)))) | |
786 (cond ((and %-pos | |
787 (not @-pos)) | |
788 (goto-char (car %-pos)) | |
789 (delete-char 1) | |
790 (setq @-pos (point)) | |
791 (insert "@") | |
792 (setq %-pos (cdr %-pos)))) | |
793 (setq %-pos (nreverse %-pos)) | |
794 ;; RFC 1034 doesn't approve of this, oh well: | |
795 (downcase-region (or (car %-pos) @-pos (point-max)) (point-max)) | |
796 (cond (%-pos ; implies @-pos valid | |
797 (setq temp %-pos) | |
798 (catch 'truncated | |
799 (while temp | |
800 (goto-char (or (nth 1 temp) | |
801 @-pos)) | |
802 (skip-chars-backward mail-whitespace) | |
803 (save-excursion | |
804 (safe-move-sexp -1) | |
805 (setq domain-pos (point)) | |
806 (skip-chars-backward mail-whitespace) | |
807 (setq \.-pos (eq ?. (preceding-char)))) | |
808 (cond ((and \.-pos | |
809 (get | |
810 (intern | |
811 (buffer-substring domain-pos (point))) | |
812 'domain-name)) | |
813 (narrow-to-region (point-min) (point)) | |
814 (goto-char (car temp)) | |
815 (delete-char 1) | |
816 (setq @-pos (point)) | |
817 (setcdr temp nil) | |
818 (setq %-pos (delq @-pos %-pos)) | |
819 (insert "@") | |
820 (throw 'truncated t))) | |
821 (setq temp (cdr temp)))))) | |
822 (setq mbox-beg (point-min) | |
823 mbox-end (if %-pos (car %-pos) | |
824 (or @-pos | |
825 (point-max)))) | |
826 | |
827 ;; Done canonicalizing address. | |
828 | |
829 (set-buffer extraction-buffer) | |
830 | |
831 ;; Find the full name | |
832 | |
833 (cond ((and phrase-beg | |
834 (eq quote-beg phrase-beg) | |
835 (<= quote-end phrase-end)) | |
836 (narrow-to-region (1+ quote-beg) (1- quote-end)) | |
837 (undo-backslash-quoting (point-min) (point-max))) | |
838 (phrase-beg | |
839 (narrow-to-region phrase-beg phrase-end)) | |
840 (comment-beg | |
841 (narrow-to-region (1+ comment-beg) (1- comment-end)) | |
842 (undo-backslash-quoting (point-min) (point-max))) | |
843 (t | |
844 ;; *** Work in canon buffer instead? No, can't. Hmm. | |
845 (delete-region (point-min) (point-max)) | |
846 (insert-buffer-substring canonicalization-buffer | |
847 mbox-beg mbox-end) | |
848 (goto-char (point-min)) | |
849 (setq \.-ends-name (search-forward "_" nil t)) | |
850 (goto-char (point-min)) | |
851 (while (progn | |
852 (skip-chars-forward mail-whitespace) | |
853 (not (eobp))) | |
854 (setq char (char-after (point))) | |
855 (cond | |
856 ((eq char ?\") | |
857 (setq quote-beg (point)) | |
858 (or (safe-move-sexp 1) | |
859 ;; TODO: handle this error condition!!!!! | |
860 (forward-char 1)) | |
861 ;; take into account deletions | |
862 (setq quote-end (- (point) 2)) | |
863 (save-excursion | |
864 (backward-char 1) | |
865 (delete-char 1) | |
866 (goto-char quote-beg) | |
867 (delete-char 1)) | |
868 (undo-backslash-quoting quote-beg quote-end) | |
869 (or (eq mail-space-char (char-after (point))) | |
870 (insert " ")) | |
871 (setq \.-ends-name t)) | |
872 ((eq char ?.) | |
873 (if (eq (char-after (1+ (point))) ?_) | |
874 (progn | |
875 (forward-char 1) | |
876 (delete-char 1) | |
877 (insert mail-space-char)) | |
878 (if \.-ends-name | |
879 (narrow-to-region (point-min) (point)) | |
880 (delete-char 1) | |
881 (insert " ")))) | |
882 ((memq (char-syntax char) '(?. ?\\)) | |
883 (delete-char 1) | |
884 (insert " ")) | |
885 (t | |
886 (setq atom-beg (point)) | |
887 (forward-word 1) | |
888 (setq atom-end (point)) | |
889 (save-restriction | |
890 (narrow-to-region atom-beg atom-end) | |
891 (goto-char (point-min)) | |
892 (while (re-search-forward "\\([^_]+\\)_" nil t) | |
893 (replace-match "\\1 ")) | |
894 (goto-char (point-max)))))))) | |
895 | |
896 (set-syntax-table address-text-syntax-table) | |
897 | |
898 (setq xxx (variant-method (buffer-string))) | |
899 (delete-region (point-min) (point-max)) | |
900 (insert xxx) | |
901 (goto-char (point-min)) | |
902 | |
903 ;; ;; Compress whitespace | |
904 ;; (goto-char (point-min)) | |
905 ;; (while (re-search-forward "[ \t\n]+" nil t) | |
906 ;; (replace-match " ")) | |
907 ;; | |
908 ;; ;; Fix . used as space | |
909 ;; (goto-char (point-min)) | |
910 ;; (while (re-search-forward mail-bad-\.-pattern nil t) | |
911 ;; (replace-match "\\1 \\2")) | |
912 ;; | |
913 ;; ;; Delete trailing parenthesized comment | |
914 ;; (goto-char (point-max)) | |
915 ;; (skip-chars-backward mail-whitespace) | |
916 ;; (cond ((memq (char-after (1- (point))) '(?\) ?\} ?\])) | |
917 ;; (setq comment-end (point)) | |
918 ;; (set-syntax-table address-text-comment-syntax-table) | |
919 ;; (or (safe-move-sexp -1) | |
920 ;; (backward-char 1)) | |
921 ;; (set-syntax-table address-text-syntax-table) | |
922 ;; (setq comment-beg (point)) | |
923 ;; (skip-chars-backward mail-whitespace) | |
924 ;; (if (bobp) | |
925 ;; (narrow-to-region (1+ comment-beg) (1- comment-end)) | |
926 ;; (narrow-to-region (point-min) (point))))) | |
927 ;; | |
928 ;; ;; Find, save, and delete any name suffix | |
929 ;; ;; *** Broken! | |
930 ;; (goto-char (point-min)) | |
931 ;; (cond ((re-search-forward mail-full-name-suffix-pattern nil t) | |
932 ;; (setq name-suffix (buffer-substring (match-beginning 3) | |
933 ;; (match-end 3))) | |
934 ;; (replace-match "\\1 \\4"))) | |
935 ;; | |
936 ;; ;; Delete ALL CAPS words and after, if preceded by mixed-case or | |
937 ;; ;; lowercase words. Eg. XT-DEM. | |
938 ;; (goto-char (point-min)) | |
939 ;; ;; ## This will lose on something like "SMITH MAX". | |
940 ;; ;; ## maybe it should be | |
941 ;; ;; ## " \\([A-Z]+[-_/][A-Z]+\\|[A-Z][A-Z][A-Z]\\)\\b.*[^A-Z \t]" | |
942 ;; ;; ## that is, three-letter-upper-case-word with non-upper-case | |
943 ;; ;; ## characters following it. | |
944 ;; (if (re-search-forward mail-mixed-case-name-pattern nil t) | |
945 ;; (if (re-search-forward mail-weird-acronym-pattern nil t) | |
946 ;; (narrow-to-region (point-min) (match-beginning 0)))) | |
947 ;; | |
948 ;; ;; Delete trailing alternative address | |
949 ;; (goto-char (point-min)) | |
950 ;; (if (re-search-forward mail-alternative-address-pattern nil t) | |
951 ;; (narrow-to-region (point-min) (match-beginning 0))) | |
952 ;; | |
953 ;; ;; Delete trailing comment | |
954 ;; (goto-char (point-min)) | |
955 ;; (if (re-search-forward mail-trailing-comment-start-pattern nil t) | |
956 ;; (or (progn | |
957 ;; (goto-char (match-beginning 0)) | |
958 ;; (skip-chars-backward mail-whitespace) | |
959 ;; (bobp)) | |
960 ;; (narrow-to-region (point-min) (match-beginning 0)))) | |
961 ;; | |
962 ;; ;; Delete trailing comma-separated comment | |
963 ;; (goto-char (point-min)) | |
964 ;; ;; ## doesn't this break "Smith, John"? Yes. | |
965 ;; (re-search-forward mail-last-name-first-pattern nil t) | |
966 ;; (while (search-forward "," nil t) | |
967 ;; (or (save-excursion | |
968 ;; (backward-char 2) | |
969 ;; (looking-at mail-full-name-suffix-pattern)) | |
970 ;; (narrow-to-region (point-min) (1- (point))))) | |
971 ;; | |
972 ;; ;; Delete telephone numbers and ham radio call signs | |
973 ;; (goto-char (point-min)) | |
974 ;; (if (re-search-forward mail-telephone-extension-pattern nil t) | |
975 ;; (narrow-to-region (point-min) (match-beginning 0))) | |
976 ;; (goto-char (point-min)) | |
977 ;; (if (re-search-forward mail-ham-call-sign-pattern nil t) | |
978 ;; (if (eq (match-beginning 0) (point-min)) | |
979 ;; (narrow-to-region (match-end 0) (point-max)) | |
980 ;; (narrow-to-region (point-min) (match-beginning 0)))) | |
981 ;; | |
982 ;; ;; Delete trailing word followed immediately by . | |
983 ;; (goto-char (point-min)) | |
984 ;; ;; ## what's this for? doesn't it mess up "Public, Harry Q."? No. | |
985 ;; (if (re-search-forward "\\b[A-Za-z][A-Za-z]+\\. *\\'" nil t) | |
986 ;; (narrow-to-region (point-min) (match-beginning 0))) | |
987 ;; | |
988 ;; ;; Handle & substitution | |
989 ;; ;; TODO: remember to disable middle initial guessing | |
990 ;; (goto-char (point-min)) | |
991 ;; (cond ((re-search-forward "\\( \\|\\`\\)&\\( \\|\\'\\)" nil t) | |
992 ;; (goto-char (match-end 1)) | |
993 ;; (delete-char 1) | |
994 ;; (capitalize-region | |
995 ;; (point) | |
996 ;; (progn | |
997 ;; (insert-buffer-substring canonicalization-buffer | |
998 ;; mbox-beg mbox-end) | |
999 ;; (point))))) | |
1000 ;; | |
1001 ;; ;; Delete nickname | |
1002 ;; (goto-char (point-min)) | |
1003 ;; (if (re-search-forward mail-nickname-pattern nil t) | |
1004 ;; (replace-match (if (eq (match-beginning 2) (1- (match-end 2))) | |
1005 ;; " \\2 " | |
1006 ;; " "))) | |
1007 ;; | |
1008 ;; ;; Fixup initials | |
1009 ;; (while (progn | |
1010 ;; (goto-char (point-min)) | |
1011 ;; (re-search-forward mail-bad-initials-pattern nil t)) | |
1012 ;; (replace-match | |
1013 ;; (if (match-beginning 4) | |
1014 ;; "\\1. \\4" | |
1015 ;; (if (match-beginning 5) | |
1016 ;; "\\1. \\5" | |
1017 ;; "\\1. ")))) | |
1018 ;; | |
1019 ;; ;; Delete title | |
1020 ;; (goto-char (point-min)) | |
1021 ;; (if (re-search-forward mail-full-name-prefixes nil t) | |
1022 ;; (narrow-to-region (point) (point-max))) | |
1023 ;; | |
1024 ;; ;; Delete trailing and preceding non-name characters | |
1025 ;; (goto-char (point-min)) | |
1026 ;; (skip-chars-forward mail-non-begin-name-chars) | |
1027 ;; (narrow-to-region (point) (point-max)) | |
1028 ;; (goto-char (point-max)) | |
1029 ;; (skip-chars-backward mail-non-end-name-chars) | |
1030 ;; (narrow-to-region (point-min) (point)) | |
1031 | |
1032 ;; If name is "First Last" and userid is "F?L", then assume | |
1033 ;; the middle initial is the second letter in the userid. | |
1034 ;; initially by Jamie Zawinski <jwz@lucid.com> | |
1035 (cond ((and (eq 3 (- mbox-end mbox-beg)) | |
1036 (progn | |
1037 (goto-char (point-min)) | |
1038 (looking-at mail-two-name-pattern))) | |
1039 (setq fi (char-after (match-beginning 0)) | |
1040 li (char-after (match-beginning 3))) | |
1041 (save-excursion | |
1042 (set-buffer canonicalization-buffer) | |
1043 ;; char-equal is ignoring case here, so no need to upcase | |
1044 ;; or downcase. | |
1045 (let ((case-fold-search t)) | |
1046 (and (char-equal fi (char-after mbox-beg)) | |
1047 (char-equal li (char-after (1- mbox-end))) | |
1048 (setq mi (char-after (1+ mbox-beg)))))) | |
1049 (cond ((and mi | |
1050 ;; TODO: use better table than syntax table | |
1051 (eq ?w (char-syntax mi))) | |
1052 (goto-char (match-beginning 3)) | |
1053 (insert (upcase mi) ". "))))) | |
1054 | |
1055 ;; ;; Restore suffix | |
1056 ;; (cond (name-suffix | |
1057 ;; (goto-char (point-max)) | |
1058 ;; (insert ", " name-suffix) | |
1059 ;; (backward-word 1) | |
1060 ;; (cond ((memq (following-char) '(?j ?J ?s ?S)) | |
1061 ;; (capitalize-word 1) | |
1062 ;; (or (eq (following-char) ?.) | |
1063 ;; (insert ?.))) | |
1064 ;; (t | |
1065 ;; (upcase-word 1))))) | |
1066 | |
1067 ;; Result | |
1068 (list (buffer-string) | |
1069 (progn | |
1070 (set-buffer canonicalization-buffer) | |
1071 (buffer-string))) | |
1072 ))) | |
1073 | |
1074 ;; TODO: put this back in the above function now that it's proven: | |
1075 (defun variant-method (string) | |
1076 (let ((variant-buffer (get-buffer-create "*variant method buffer*")) | |
1077 (word-count 0) | |
1078 mixed-case-flag lower-case-flag upper-case-flag | |
1079 suffix-flag last-name-comma-flag | |
1080 comment-beg comment-end initial beg end | |
1081 ) | |
1082 (save-excursion | |
1083 (set-buffer variant-buffer) | |
923 | 1084 (buffer-disable-undo variant-buffer) |
809 | 1085 (set-syntax-table address-text-syntax-table) |
1086 (widen) | |
1087 (erase-buffer) | |
1088 (setq case-fold-search nil) | |
1089 | |
1090 (insert string) | |
1091 | |
1092 ;; Fix . used as space | |
1093 (goto-char (point-min)) | |
1094 (while (re-search-forward mail-bad-\.-pattern nil t) | |
1095 (replace-match "\\1 \\2")) | |
1096 | |
1097 ;; Skip any initial garbage. | |
1098 (goto-char (point-min)) | |
1099 (skip-chars-forward mail-non-begin-name-chars) | |
1100 (skip-chars-backward "& \"") | |
1101 (narrow-to-region (point) (point-max)) | |
1102 | |
1103 (catch 'stop | |
1104 (while t | |
1105 (skip-chars-forward mail-whitespace) | |
1106 | |
1107 (cond | |
1108 | |
1109 ;; Delete title | |
1110 ((and (eq word-count 0) | |
1111 (looking-at mail-full-name-prefixes)) | |
1112 (goto-char (match-end 0)) | |
1113 (narrow-to-region (point) (point-max))) | |
1114 | |
1115 ;; Stop after name suffix | |
1116 ((and (>= word-count 2) | |
1117 (looking-at mail-full-name-suffix-pattern)) | |
1118 (skip-chars-backward mail-whitespace) | |
1119 (setq suffix-flag (point)) | |
1120 (if (eq ?, (following-char)) | |
1121 (forward-char 1) | |
1122 (insert ?,)) | |
1123 ;; Enforce at least one space after comma | |
1124 (or (eq mail-space-char (following-char)) | |
1125 (insert mail-space-char)) | |
1126 (skip-chars-forward mail-whitespace) | |
1127 (cond ((memq (following-char) '(?j ?J ?s ?S)) | |
1128 (capitalize-word 1) | |
1129 (if (eq (following-char) ?.) | |
1130 (forward-char 1) | |
1131 (insert ?.))) | |
1132 (t | |
1133 (upcase-word 1))) | |
1134 (setq word-count (1+ word-count)) | |
1135 (throw 'stop t)) | |
1136 | |
1137 ;; Handle SCA names | |
1138 ((looking-at "MKA \\(.+\\)") ; "Mundanely Known As" | |
1139 (setq word-count 0) | |
1140 (goto-char (match-beginning 1)) | |
1141 (narrow-to-region (point) (point-max))) | |
1142 | |
1143 ;; Various stopping points | |
1144 ((or | |
1145 ;; Stop before ALL CAPS acronyms, if preceded by mixed-case or | |
1146 ;; lowercase words. Eg. XT-DEM. | |
1147 (and (>= word-count 2) | |
1148 (or mixed-case-flag lower-case-flag) | |
1149 (looking-at mail-weird-acronym-pattern) | |
1150 (not (looking-at mail-roman-numeral-pattern))) | |
1151 ;; Stop before 4-or-more letter lowercase words preceded by | |
1152 ;; mixed case or uppercase words. | |
1153 (and (>= word-count 2) | |
1154 (or upper-case-flag mixed-case-flag) | |
1155 (looking-at "[a-z][a-z][a-z][a-z]+\\b")) | |
1156 ;; Stop before trailing alternative address | |
1157 (looking-at mail-alternative-address-pattern) | |
1158 ;; Stop before trailing comment not introduced by comma | |
1159 (looking-at mail-trailing-comment-start-pattern) | |
1160 ;; Stop before telephone numbers | |
1161 (looking-at mail-telephone-extension-pattern)) | |
1162 (throw 'stop t)) | |
1163 | |
1164 ;; Check for initial last name followed by comma | |
1165 ((and (eq ?, (following-char)) | |
1166 (eq word-count 1)) | |
1167 (forward-char 1) | |
1168 (setq last-name-comma-flag t) | |
1169 (or (eq mail-space-char (following-char)) | |
1170 (insert mail-space-char))) | |
1171 | |
1172 ;; Stop before trailing comma-separated comment | |
1173 ((eq ?, (following-char)) | |
1174 (throw 'stop t)) | |
1175 | |
1176 ;; Delete parenthesized/quoted comment/nickname | |
1177 ((memq (following-char) '(?\( ?\{ ?\[ ?\" ?\' ?\`)) | |
1178 (setq comment-beg (point)) | |
1179 (set-syntax-table address-text-comment-syntax-table) | |
1180 (cond ((memq (following-char) '(?\' ?\`)) | |
1181 (if (eq ?\' (following-char)) | |
1182 (forward-char 1)) | |
1183 (or (search-forward "'" nil t) | |
1184 (delete-char 1))) | |
1185 (t | |
1186 (or (safe-move-sexp 1) | |
1187 (goto-char (point-max))))) | |
1188 (set-syntax-table address-text-syntax-table) | |
1189 (setq comment-end (point)) | |
1190 (cond | |
1191 ;; Handle case of entire name being quoted | |
1192 ((and (eq word-count 0) | |
1193 (looking-at " *\\'") | |
1194 (>= (- comment-end comment-beg) 2)) | |
1195 (narrow-to-region (1+ comment-beg) (1- comment-end)) | |
1196 (goto-char (point-min))) | |
1197 (t | |
1198 ;; Handle case of quoted initial | |
1199 (if (and (or (= 3 (- comment-end comment-beg)) | |
1200 (and (= 4 (- comment-end comment-beg)) | |
1201 (eq ?. (char-after (+ 2 comment-beg))))) | |
1202 (not (looking-at " *\\'"))) | |
1203 (setq initial (char-after (1+ comment-beg))) | |
1204 (setq initial nil)) | |
1205 (delete-region comment-beg comment-end) | |
1206 (if initial | |
1207 (insert initial ". "))))) | |
1208 | |
1209 ;; Delete ham radio call signs | |
1210 ((looking-at mail-ham-call-sign-pattern) | |
1211 (delete-region (match-beginning 0) (match-end 0))) | |
1212 | |
1213 ;; Handle & substitution | |
1214 ;; TODO: remember to disable middle initial guessing | |
1215 ((and (or (bobp) | |
1216 (eq mail-space-char (preceding-char))) | |
1217 (looking-at "&\\( \\|\\'\\)")) | |
1218 (delete-char 1) | |
1219 (capitalize-region | |
1220 (point) | |
1221 (progn | |
1222 (insert-buffer-substring canonicalization-buffer | |
1223 mbox-beg mbox-end) | |
1224 (point)))) | |
1225 | |
1226 ;; Fixup initials | |
1227 ((looking-at mail-initial-pattern) | |
1228 (or (eq (following-char) (upcase (following-char))) | |
1229 (setq lower-case-flag t)) | |
1230 (forward-char 1) | |
1231 (if (eq ?. (following-char)) | |
1232 (forward-char 1) | |
1233 (insert ?.)) | |
1234 (or (eq mail-space-char (following-char)) | |
1235 (insert mail-space-char)) | |
1236 (setq word-count (1+ word-count))) | |
1237 | |
1238 ;; Regular name words | |
1239 ((looking-at mail-name-pattern) | |
1240 (setq beg (point)) | |
1241 (setq end (match-end 0)) | |
1242 (set (if (re-search-forward "[a-z]" end t) | |
1243 (if (progn | |
1244 (goto-char beg) | |
1245 (re-search-forward "[A-Z]" end t)) | |
1246 'mixed-case-flag | |
1247 'lower-case-flag) | |
1248 'upper-case-flag) t) | |
1249 (goto-char end) | |
1250 (setq word-count (1+ word-count))) | |
1251 | |
1252 (t | |
1253 (throw 'stop t))))) | |
1254 | |
1255 (narrow-to-region (point-min) (point)) | |
1256 | |
1257 ;; Delete trailing word followed immediately by . | |
1258 (cond ((not suffix-flag) | |
1259 (goto-char (point-min)) | |
1260 (if (re-search-forward "\\b[A-Za-z][A-Za-z]+\\. *\\'" nil t) | |
1261 (narrow-to-region (point-min) (match-beginning 0))))) | |
1262 | |
1263 ;; If last name first put it at end (but before suffix) | |
1264 (cond (last-name-comma-flag | |
1265 (goto-char (point-min)) | |
1266 (search-forward ",") | |
1267 (setq end (1- (point))) | |
1268 (goto-char (or suffix-flag (point-max))) | |
1269 (or (eq mail-space-char (preceding-char)) | |
1270 (insert mail-space-char)) | |
1271 (insert-buffer-substring (current-buffer) (point-min) end) | |
1272 (narrow-to-region (1+ end) (point-max)))) | |
1273 | |
1274 (goto-char (point-max)) | |
1275 (skip-chars-backward mail-non-end-name-chars) | |
1276 (if (eq ?. (following-char)) | |
1277 (forward-char 1)) | |
1278 (narrow-to-region (point) | |
1279 (progn | |
1280 (goto-char (point-min)) | |
1281 (skip-chars-forward mail-non-begin-name-chars) | |
1282 (point))) | |
1283 | |
1284 ;; Compress whitespace | |
1285 (goto-char (point-min)) | |
1286 (while (re-search-forward "[ \t\n]+" nil t) | |
1287 (replace-match " ")) | |
1288 | |
1289 (buffer-substring (point-min) (point-max)) | |
1290 | |
1291 ))) | |
1292 | |
1293 ;; The country names are just in there for show right now, and because | |
1294 ;; Jamie thought it would be neat. They aren't used yet. | |
1295 | |
1296 ;; Keep in mind that the country abbreviations follow ISO-3166. There is | |
1297 ;; a U.S. FIPS that specifies a different set of two-letter country | |
1298 ;; abbreviations. | |
1299 | |
1300 ;; TODO: put this in its own obarray, instead of cluttering up the main | |
1301 ;; symbol table with junk. | |
1302 | |
1303 (mapcar | |
1304 (function | |
1305 (lambda (x) | |
1306 (if (symbolp x) | |
1307 (put x 'domain-name t) | |
1308 (put (car x) 'domain-name (nth 1 x))))) | |
1309 '((ag "Antigua") | |
1310 (ar "Argentina") ; Argentine Republic | |
1311 arpa ; Advanced Projects Research Agency | |
1312 (at "Austria") ; The Republic of _ | |
1313 (au "Australia") | |
1314 (bb "Barbados") | |
1315 (be "Belgium") ; The Kingdom of _ | |
1316 (bg "Bulgaria") | |
1317 bitnet ; Because It's Time NET | |
1318 (bo "Bolivia") ; Republic of _ | |
1319 (br "Brazil") ; The Federative Republic of _ | |
1320 (bs "Bahamas") | |
1321 (bz "Belize") | |
1322 (ca "Canada") | |
1323 (ch "Switzerland") ; The Swiss Confederation | |
1324 (cl "Chile") ; The Republic of _ | |
1325 (cn "China") ; The People's Republic of _ | |
1326 (co "Columbia") | |
1327 com ; Commercial | |
1328 (cr "Costa Rica") ; The Republic of _ | |
1329 (cs "Czechoslovakia") | |
1330 (de "Germany") | |
1331 (dk "Denmark") | |
1332 (dm "Dominica") | |
1333 (do "Dominican Republic") ; The _ | |
1334 (ec "Ecuador") ; The Republic of _ | |
1335 edu ; Educational | |
1336 (eg "Egypt") ; The Arab Republic of _ | |
1337 (es "Spain") ; The Kingdom of _ | |
1338 (fi "Finland") ; The Republic of _ | |
1339 (fj "Fiji") | |
1340 (fr "France") | |
1341 gov ; Government (U.S.A.) | |
1342 (gr "Greece") ; The Hellenic Republic | |
1343 (hk "Hong Kong") | |
1344 (hu "Hungary") ; The Hungarian People's Republic (???) | |
1345 (ie "Ireland") | |
1346 (il "Israel") ; The State of _ | |
1347 (in "India") ; The Republic of _ | |
1348 int ; something British, don't know what | |
1349 (is "Iceland") ; The Republic of _ | |
1350 (it "Italy") ; The Italian Republic | |
1351 (jm "Jamaica") | |
1352 (jp "Japan") | |
1353 (kn "St. Kitts and Nevis") | |
1354 (kr "South Korea") | |
1355 (lc "St. Lucia") | |
1356 (lk "Sri Lanka") ; The Democratic Socialist Republic of _ | |
1357 mil ; Military (U.S.A.) | |
1358 (mx "Mexico") ; The United Mexican States | |
1359 (my "Malaysia") ; changed to Myanmar???? | |
1360 (na "Namibia") | |
1361 nato ; North Atlantic Treaty Organization | |
1362 net ; Network | |
1363 (ni "Nicaragua") ; The Republic of _ | |
1364 (nl "Netherlands") ; The Kingdom of the _ | |
1365 (no "Norway") ; The Kingdom of _ | |
1366 (nz "New Zealand") | |
1367 org ; Organization | |
1368 (pe "Peru") | |
1369 (pg "Papua New Guinea") | |
1370 (ph "Philippines") ; The Republic of the _ | |
1371 (pl "Poland") | |
1372 (pr "Puerto Rico") | |
1373 (pt "Portugal") ; The Portugese Republic | |
1374 (py "Paraguay") | |
1375 (se "Sweden") ; The Kingdom of _ | |
1376 (sg "Singapore") ; The Republic of _ | |
1377 (sr "Suriname") | |
1378 (su "Soviet Union") | |
1379 (th "Thailand") ; The Kingdom of _ | |
1380 (tn "Tunisia") | |
1381 (tr "Turkey") ; The Republic of _ | |
1382 (tt "Trinidad and Tobago") | |
1383 (tw "Taiwan") | |
1384 (uk "United Kingdom") ; The _ of Great Britain | |
1385 unter-dom ; something German | |
1386 (us "U.S.A.") ; The United States of America | |
1387 uucp ; Unix to Unix CoPy | |
1388 (uy "Uruguay") ; The Eastern Republic of _ | |
1389 (vc "St. Vincent and the Grenadines") | |
1390 (ve "Venezuela") ; The Republic of _ | |
1391 (yu "Yugoslavia") ; The Socialist Federal Republic of _ | |
1392 ;; Also said to be Zambia ... | |
1393 (za "South Africa") ; The Republic of _ (why not Zaire???) | |
1394 (zw "Zimbabwe") ; Republic of _ | |
1395 )) | |
1396 ;; fipnet | |
1397 | |
1398 | |
1399 ;; Code for testing. | |
1400 | |
1401 (defun time-extract () | |
1402 (let (times list) | |
1403 (setq times (cons (current-time-string) times) | |
1404 list problem-address-alist) | |
1405 (while list | |
1406 (mail-extract-address-components (car (car list))) | |
1407 (setq list (cdr list))) | |
1408 (setq times (cons (current-time-string) times)) | |
1409 (nreverse times))) | |
1410 | |
1411 (defun test-extract (&optional starting-point) | |
1412 (interactive) | |
1413 (set-buffer (get-buffer-create "*Testing*")) | |
1414 (erase-buffer) | |
1415 (sit-for 0) | |
1416 (mapcar 'test-extract-internal | |
1417 (if starting-point | |
1418 (memq starting-point problem-address-alist) | |
1419 problem-address-alist))) | |
1420 | |
1421 (defvar failed-item) | |
1422 (defun test-extract-internal (item) | |
1423 (setq failed-item item) | |
1424 (let* ((address (car item)) | |
1425 (correct-name (nth 1 item)) | |
1426 (correct-canon (nth 2 item)) | |
1427 (result (mail-extract-address-components address)) | |
1428 (name (car result)) | |
1429 (canon (nth 1 result)) | |
1430 (name-correct (or (null correct-name) | |
1431 (string-equal (downcase correct-name) | |
1432 (downcase name)))) | |
1433 (canon-correct (or (null correct-canon) | |
1434 (string-equal correct-canon canon)))) | |
1435 (cond ((not (and name-correct canon-correct)) | |
1436 (pop-to-buffer "*Testing*") | |
1437 (select-window (get-buffer-window (current-buffer))) | |
1438 (goto-char (point-max)) | |
1439 (insert "Address: " address "\n") | |
1440 (if (not name-correct) | |
1441 (insert " Correct Name: [" correct-name | |
1442 "]\; Result: [" name "]\n")) | |
1443 (if (not canon-correct) | |
1444 (insert " Correct Canon: [" correct-canon | |
1445 "]\; Result: [" canon "]\n")) | |
1446 (insert "\n") | |
1447 (sit-for 0)))) | |
1448 (setq failed-item nil)) | |
1449 | |
1450 (defun test-continue-extract () | |
1451 (interactive) | |
1452 (test-extract failed-item)) | |
1453 | |
1454 | |
1455 ;; Assorted junk. | |
1456 | |
1457 ;; warsaw@nlm.nih.gov (A Bad Dude -- Barry Warsaw) | |
1458 | |
1459 ;;'(from | |
1460 ;; reply-to | |
1461 ;; return-path | |
1462 ;; x-uucp-from | |
1463 ;; sender | |
1464 ;; resent-from | |
1465 ;; resent-sender | |
1466 ;; resent-reply-to) | |
1467 | |
1468 ;;; mail-extr.el ends here |