| blob | 2fd7b8322f74ff9ed931064d0b64b6c74743bbbf |
1 /*-
2 * Copyright (c) 1993, 1994
3 * The Regents of the University of California. All rights reserved.
4 * Copyright (c) 1992, 1993, 1994, 1995, 1996
5 * Keith Bostic. All rights reserved.
6 *
7 * See the LICENSE file for redistribution information.
8 */
12 #ifndef lint
13 static const char sccsid[] = "$Id: v_txt.c,v 10.108 2003/07/18 21:27:42 skimo Exp $ (Berkeley) $Date: 2003/07/18 21:27:42 $";
16 #include <sys/types.h>
17 #include <sys/queue.h>
18 #include <sys/stat.h>
19 #include <sys/time.h>
21 #include <bitstring.h>
22 #include <ctype.h>
23 #include <errno.h>
24 #include <limits.h>
25 #include <stdio.h>
26 #include <stdlib.h>
27 #include <string.h>
28 #include <unistd.h>
53 /* Cursor character (space is hard to track on the screen). */
54 #if defined(DEBUG) && 0
55 #undef CH_CURSOR
57 #endif
59 /*
60 * v_tcmd --
61 * Fill a buffer from the terminal for vi.
62 *
63 * PUBLIC: int v_tcmd __P((SCR *, VICMD *, ARG_CHAR_T, u_int));
64 */
65 int
67 {
68 /* Normally, we end up where we started. */
72 /* Initialize the map. */
76 /* Move to the last line. */
80 /* Don't update the modeline for now. */
83 /* Set the input flags. */
91 /* Do the input thing. */
95 /* Reenable the modeline updates. */
98 /* Clean up the map. */
105 /* Set the cursor to the resulting position. */
110 }
112 /*
113 * txt_map_init
114 * Initialize the screen map for colon command-line input.
115 */
116 static int
118 {
124 /*
125 * Fake like the user is doing input on the last line of the
126 * screen. This makes all of the scrolling work correctly,
127 * and allows us the use of the vi text editing routines, not
128 * to mention practically infinite length ex commands.
129 *
130 * Save the current location.
131 */
139 /*
140 * If it's a small screen, TMAP may be small for the screen.
141 * Fix it, filling in fake lines as we go.
142 */
149 }
151 /* Build the fake entry. */
158 /* Reset the screen information. */
160 }
162 }
164 /*
165 * txt_map_end
166 * Reset the screen map for colon command-line input.
167 */
168 static int
170 {
176 /* Restore the screen information. */
181 /*
182 * If it's a small screen, TMAP may be wrong. Clear any
183 * lines that might have been overwritten.
184 */
189 }
194 /*
195 * The map may be wrong if the user entered more than one
196 * (logical) line. Fix it. If the user entered a whole
197 * screen, this will be slow, but we probably don't care.
198 */
204 }
206 /*
207 * Invalidate the cursor and the line size cache, the line never
208 * really existed. This fixes bugs where the user searches for
209 * the last line on the screen + 1 and the refresh routine thinks
210 * that's where we just were.
211 */
216 }
218 /*
219 * If doing input mapping on the colon command line, may need to unmap
220 * based on the command.
221 */
222 #define UNMAP_TST \
223 FL_ISSET(ec_flags, EC_MAPINPUT) && LF_ISSET(TXT_INFOLINE)
225 /*
226 * Internally, we maintain tp->lno and tp->cno, externally, everyone uses
227 * sp->lno and sp->cno. Make them consistent as necessary.
228 */
229 #define UPDATE_POSITION(sp, tp) { \
230 (sp)->lno = (tp)->lno; \
231 (sp)->cno = (tp)->cno; \
232 }
234 /*
235 * v_txt --
236 * Vi text input.
237 *
238 * PUBLIC: int v_txt __P((SCR *, VICMD *, MARK *,
239 * PUBLIC: const CHAR_T *, size_t, ARG_CHAR_T, db_recno_t, u_long, u_int32_t));
240 */
241 int
242 v_txt(SCR *sp, VICMD *vp, MARK *tm, const CHAR_T *lp, size_t len, ARG_CHAR_T prompt, db_recno_t ai_line, u_long rcount, u_int32_t flags)
245 /* To MARK. */
246 /* Input line. */
247 /* Input line length. */
248 /* Prompt to display. */
249 /* Line number to use for autoindent count. */
250 /* Replay count. */
251 /* TXT_* flags. */
252 {
271 u_int8_t is_flags;
284 /*
285 * Set the input flag, so tabs get displayed correctly
286 * and everyone knows that the text buffer is in use.
287 */
290 /*
291 * Get one TEXT structure with some initial buffer space, reusing
292 * the last one if it's big enough. (All TEXT bookkeeping fields
293 * default to 0 -- text_init() handles this.) If changing a line,
294 * copy it into the TEXT buffer.
295 */
302 }
314 }
316 /* Set default termination condition. */
319 /* Set the starting line, column. */
323 /*
324 * Set the insert and overwrite counts. If overwriting characters,
325 * do insertion afterward. If not overwriting characters, assume
326 * doing insertion. If change is to a mark, emphasize it with an
327 * CH_ENDMARK character.
328 */
338 }
340 /*
341 * Many of the special cases in text input are to handle autoindent
342 * support. Somebody decided that it would be a good idea if "^^D"
343 * and "0^D" deleted all of the autoindented characters. In an editor
344 * that takes single character input from the user, this beggars the
345 * imagination. Note also, "^^D" resets the next lines' autoindent,
346 * but "0^D" doesn't.
347 *
348 * We assume that autoindent only happens on empty lines, so insert
349 * and overwrite will be zero. If doing autoindent, figure out how
350 * much indentation we need and fill it in. Update input column and
351 * screen cursor as necessary.
352 */
358 /*
359 * The cc and S commands have a special feature -- leading
360 * <blank> characters are handled as autoindent characters.
361 * Beauty!
362 */
368 }
370 /* If getting a command buffer from the user, there may be a prompt. */
375 }
377 /*
378 * If appending after the end-of-line, add a space into the buffer
379 * and move the cursor right. This space is inserted, i.e. pushed
380 * along, and then deleted when the line is resolved. Assumes that
381 * the cursor is already positioned at the end of the line. This
382 * avoids the nastiness of having the cursor reside on a magical
383 * column, i.e. a column that doesn't really exist. The only down
384 * side is that we may wrap lines or scroll the screen before it's
385 * strictly necessary. Not a big deal.
386 */
392 }
394 /*
395 * Historic practice is that the wrapmargin value was a distance
396 * from the RIGHT-HAND margin, not the left. It's more useful to
397 * us as a distance from the left-hand margin, i.e. the same as
398 * the wraplen value. The wrapmargin option is historic practice.
399 * Nvi added the wraplen option so that it would be possible to
400 * edit files with consistent margins without knowing the number of
401 * columns in the window.
402 *
403 * XXX
404 * Setting margin causes a significant performance hit. Normally
405 * we don't update the screen if there are keys waiting, but we
406 * have to if margin is set, otherwise the screen routines don't
407 * know where the cursor is.
408 *
409 * !!!
410 * Abbreviated keys were affected by the wrapmargin option in the
411 * historic 4BSD vi. Mapped keys were usually, but sometimes not.
412 * See the comment in vi/v_text():set_txt_std for more information.
413 *
414 * !!!
415 * One more special case. If an inserted <blank> character causes
416 * wrapmargin to split the line, the next user entered character is
417 * discarded if it's a <space> character.
418 */
423 else
425 else
428 /* Initialize abbreviation checks. */
433 /*
434 * Set up the dot command. Dot commands are done by saving the actual
435 * characters and then reevaluating them so that things like wrapmargin
436 * can change between the insert and the replay.
437 *
438 * !!!
439 * Historically, vi did not remap or reabbreviate replayed input. (It
440 * did beep at you if you changed an abbreviation and then replayed the
441 * input. We're not that compatible.) We don't have to do anything to
442 * avoid remapping, as we're not getting characters from the terminal
443 * routines. Turn the abbreviation check off.
444 *
445 * XXX
446 * It would be nice if we could swallow backspaces and such, but it's
447 * not all that easy to do. What we can do is turn off the common
448 * error messages during the replay. Otherwise, when the user enters
449 * an illegal command, e.g., "Ia<erase><erase><erase><erase>b<escape>",
450 * and then does a '.', they get a list of error messages after command
451 * completion.
452 */
457 }
459 /* Other text input mode setup. */
467 /* Initialize input flags. */
470 /* Refresh the screen. */
475 /* If it's dot, just do it now. */
479 /* Get an event. */
484 /*
485 * If file completion overwrote part of the screen and nothing else has
486 * been displayed, clean up. We don't do this as part of the normal
487 * message resolution because we know the user is on the colon command
488 * line and there's no reason to enter explicit characters to continue.
489 */
500 }
502 /* Deal with all non-character events. */
511 /*
512 * !!!
513 * Historically, <interrupt> exited the user from text input
514 * mode or cancelled a colon command, and returned to command
515 * mode. It also beeped the terminal, but that seems a bit
516 * excessive.
517 */
524 /* <resize> interrupts the input mode. */
530 }
532 /*
533 * !!!
534 * If the first character of the input is a nul, replay the previous
535 * input. (Historically, it's okay to replay non-existent input.)
536 * This was not documented as far as I know, and is a great test of vi
537 * clones.
538 */
547 }
549 /*
550 * File name completion and colon command-line editing. We don't
551 * have enough meta characters, so we expect people to overload
552 * them. If the two characters are the same, then we do file name
553 * completion if the cursor is past the first column, and do colon
554 * command-line editing if it's not.
555 */
569 }
574 }
575 }
577 /* Abbreviation overflow check. See comment in txt_abbrev(). */
578 #define MAX_ABBREVIATION_EXPANSION 256
588 }
592 /* Check to see if the character fits into the replay buffers. */
597 }
603 }
605 /* Wrapmargin check for leading space. */
610 }
612 /* If quoted by someone else, simply insert the character. */
616 /*
617 * !!!
618 * If this character was quoted by a K_VLNEXT or a backslash, replace
619 * the placeholder (a carat or a backslash) with the new character.
620 * If it was quoted by a K_VLNEXT, we've already adjusted the cursor
621 * because it has to appear on top of the placeholder character. If
622 * it was quoted by a backslash, adjust the cursor now, the cursor
623 * doesn't appear on top of it. Historic practice in both cases.
624 *
625 * Skip tests for abbreviations; ":ab xa XA" followed by "ixa^V<space>"
626 * doesn't perform an abbreviation. Special case, ^V^J (not ^V^M) is
627 * the same as ^J, historically.
628 */
640 }
644 }
646 }
648 /*
649 * !!!
650 * Translate "<CH_HEX>[isxdigit()]*" to a character with a hex value:
651 * this test delimits the value by any non-hex character. Offset by
652 * one, we use 0 to mean that we've found <CH_HEX>.
653 */
658 }
663 /* Return in script windows and the command line. */
666 /*
667 * If this was a map, we may have not displayed
668 * the line. Display it, just in case.
669 *
670 * If a script window and not the colon line,
671 * push a <cr> so it gets executed.
672 */
679 /* Set term condition: if empty. */
682 /*
683 * Set term condition: if searching incrementally and
684 * the user entered a pattern, return a completed
685 * search, regardless if the entire pattern was found.
686 */
692 }
694 #define LINE_RESOLVE { \
695 /* \
696 * Handle abbreviations. If there was one, discard the \
697 * replay characters. \
699 if (abb == AB_INWORD && \
700 !LF_ISSET(TXT_REPLAY) && F_ISSET(gp, G_ABBREV)) { \
701 if (txt_abbrev(sp, tp, &evp->e_c, \
702 LF_ISSET(TXT_INFOLINE), &tmp, \
703 &ab_turnoff)) \
704 goto err; \
705 if (tmp) { \
706 if (LF_ISSET(TXT_RECORD)) \
707 rcol -= tmp + 1; \
708 goto resolve; \
709 } \
710 } \
711 if (abb != AB_NOTSET) \
712 abb = AB_NOTWORD; \
713 if (UNMAP_TST) \
714 txt_unmap(sp, tp, &ec_flags); \
715 /* \
716 * Delete any appended cursor. It's possible to get in \
717 * situations where TXT_APPENDEOL is set but tp->insert \
718 * is 0 when using the R command and all the characters \
719 * are tp->owrite characters. \
721 if (LF_ISSET(TXT_APPENDEOL) && tp->insert > 0) { \
722 --tp->len; \
723 --tp->insert; \
724 } \
725 }
726 LINE_RESOLVE;
728 /*
729 * Save the current line information for restoration in
730 * txt_backup(), and set the line final length.
731 */
736 /* Update the old line. */
740 /*
741 * Historic practice, when the autoindent edit option was set,
742 * was to delete <blank> characters following the inserted
743 * newline. This affected the 'R', 'c', and 's' commands; 'c'
744 * and 's' retained the insert characters only, 'R' moved the
745 * overwrite and insert characters into the next TEXT structure.
746 * We keep track of the number of characters erased for the 'R'
747 * command so that the final resolution of the line is correct.
748 */
764 }
766 /*
767 * !!!
768 * Create a new line and insert the new TEXT into the queue.
769 * DON'T insert until the old line has been updated, or the
770 * inserted line count in line.c:db_get() will be wrong.
771 */
777 /* Set up bookkeeping for the new line. */
782 /*
783 * Reset the autoindent line value. 0^D keeps the autoindent
784 * line from changing, ^D changes the level, even if there were
785 * no characters in the old line. Note, if using the current
786 * tp structure, use the cursor as the length, the autoindent
787 * characters may have been erased.
788 */
799 }
801 /* Reset the cursor. */
804 /*
805 * If we're here because wrapmargin was set and we've broken a
806 * line, there may be additional information (i.e. the start of
807 * a line) in the wmt structure.
808 */
812 #define WMTSPACE wmt.offset + wmt.owrite + wmt.insert
820 }
822 }
824 /* New lines are TXT_APPENDEOL. */
831 }
833 /* Swap old and new TEXT's, and update the new line. */
843 /* If we have a count, start replaying the input. */
853 /*
854 * Some commands (e.g. 'o') need a <newline> for each
855 * repetition.
856 */
860 /*
861 * The R command turns into the 'a' command after the
862 * first repetition.
863 */
868 }
870 }
872 /* Set term condition: if empty. */
875 /*
876 * Set term condition: if searching incrementally and the user
877 * entered a pattern, return a completed search, regardless if
878 * the entire pattern was found.
879 */
885 /*
886 * Clean up for the 'R' command, restoring overwrite
887 * characters, and making them into insert characters.
888 */
892 /*
893 * If there are any overwrite characters, copy down
894 * any insert characters, and decrement the length.
895 */
901 }
903 /*
904 * Optionally resolve the lines into the file. If not
905 * resolving the lines into the file, end the line with
906 * a nul. If the line is empty, then set the length to
907 * 0, the termination condition has already been set.
908 *
909 * XXX
910 * This is wrong, should pass back a length.
911 */
918 }
920 /*
921 * Set the return cursor position to rest on the last
922 * inserted character.
923 */
927 /* Update the last line. */
940 /*
941 * If in the first column or no characters to erase, ignore
942 * the ^D (this matches historic practice). If not doing
943 * autoindent or already inserted non-ai characters, it's a
944 * literal. The latter test is done in the switch, as the
945 * CARAT forms are N + 1, not N.
946 */
957 /* Save the ai string for later. */
985 }
988 /* If can erase over the prompt, return. */
992 }
994 /*
995 * If at the beginning of the line, try and drop back to a
996 * previously inserted line.
997 */
1004 }
1006 /* If nothing to erase, bell the user. */
1011 }
1013 /* Drop back one character. */
1016 /*
1017 * Historically, vi didn't replace the erased characters with
1018 * <blank>s, presumably because it's easier to fix a minor
1019 * typing mistake and continue on if the previous letters are
1020 * already there. This is a problem for incremental searching,
1021 * because the user can no longer tell where they are in the
1022 * colon command line because the cursor is at the last search
1023 * point in the screen. So, if incrementally searching, erase
1024 * the erased characters from the screen.
1025 */
1029 /*
1030 * Increment overwrite, decrement ai if deleted.
1031 *
1032 * !!!
1033 * Historic vi did not permit users to use erase characters
1034 * to delete autoindent characters. We do. Eat hot death,
1035 * POSIX.
1036 */
1041 /* Reset if we deleted an incremental search character. */
1046 /*
1047 * If at the beginning of the line, try and drop back to a
1048 * previously inserted line.
1049 */
1055 }
1057 /*
1058 * If at offset, nothing to erase so bell the user.
1059 */
1064 }
1066 /*
1067 * The first werase goes back to any autoindent column and the
1068 * second werase goes back to the offset.
1069 *
1070 * !!!
1071 * Historic vi did not permit users to use erase characters to
1072 * delete autoindent characters.
1073 */
1079 }
1081 /* Skip over trailing space characters. */
1085 }
1088 /*
1089 * There are three types of word erase found on UNIX systems.
1090 * They can be identified by how the string /a/b/c is treated
1091 * -- as 1, 3, or 6 words. Historic vi had two classes of
1092 * characters, and strings were delimited by them and
1093 * <blank>'s, so, 6 words. The historic tty interface used
1094 * <blank>'s to delimit strings, so, 1 word. The algorithm
1095 * offered in the 4.4BSD tty interface (as stty altwerase)
1096 * treats it as 3 words -- there are two classes of
1097 * characters, and strings are delimited by them and
1098 * <blank>'s. The difference is that the type of the first
1099 * erased character erased is ignored, which is exactly right
1100 * when erasing pathname components. The edit options
1101 * TXT_ALTWERASE and TXT_TTYWERASE specify the 4.4BSD tty
1102 * interface and the historic tty driver behavior,
1103 * respectively, and the default is the same as the historic
1104 * vi behavior.
1105 *
1106 * Overwrite erased characters if doing incremental search;
1107 * see comment above.
1108 */
1117 }
1126 }
1137 }
1138 }
1140 /* Reset if we deleted an incremental search character. */
1145 /*
1146 * !!!
1147 * If at the beginning of the line, try and drop back to a
1148 * previously inserted line. Historic vi did not permit
1149 * users to go back to previous lines.
1150 */
1156 }
1158 /* If at offset, nothing to erase so bell the user. */
1163 }
1165 /*
1166 * First kill goes back to any autoindent and second kill goes
1167 * back to the offset.
1168 *
1169 * !!!
1170 * Historic vi did not permit users to use erase characters to
1171 * delete autoindent characters.
1172 */
1178 }
1181 /*
1182 * Overwrite erased characters if doing incremental search;
1183 * see comment above.
1184 */
1189 else
1192 /* Reset if we deleted an incremental search character. */
1208 /*
1209 * !!!
1210 * Historic vi tried to make abbreviations after a backslash
1211 * escape work. If you did ":ab x y", and inserted "x\^H",
1212 * (assuming the erase character was ^H) you got "x^H", and
1213 * no abbreviation was done. If you inserted "x\z", however,
1214 * it tried to back up and do the abbreviation, i.e. replace
1215 * 'x' with 'y'. The problem was it got it wrong, and you
1216 * ended up with "zy\".
1217 *
1218 * This is really hard to do (you have to remember the
1219 * word/non-word state, for example), and doesn't make any
1220 * sense to me. Both backslash and the characters it
1221 * (usually) escapes will individually trigger the
1222 * abbreviation, so I don't see why the combination of them
1223 * wouldn't. I don't expect to get caught on this one,
1224 * particularly since it never worked right, but I've been
1225 * wrong before.
1226 *
1227 * Do the tests for abbreviations, so ":ab xa XA",
1228 * "ixa\<K_VERASE>" performs the abbreviation.
1229 */
1235 /*
1236 * Turn on the quote flag so that the underlying routines
1237 * quote the next character where it's possible. Turn off
1238 * the input mapbiting flag so that we don't remap the next
1239 * character.
1240 */
1244 /*
1245 * !!!
1246 * Skip the tests for abbreviations, so ":ab xa XA",
1247 * "ixa^V<space>" doesn't perform the abbreviation.
1248 */
1255 * Historically, vi eliminated nul's out of hand. If the
1256 * beautify option was set, it also deleted any unknown
1257 * ASCII value less than space (040) and the del character
1258 * (0177), except for tabs. Unknown is a key word here.
1259 * Most vi documentation claims that it deleted everything
1260 * but <tab>, <nl> and <ff>, as that's what the original
1261 * 4BSD documentation said. This is obviously wrong,
1262 * however, as <esc> would be included in that list. What
1263 * we do is eliminate any unquoted, iscntrl() character that
1264 * wasn't a replay and wasn't handled specially, except
1265 * <tab> or <ff>.
1266 */
1274 }
1277 * If entering a non-word character after a word, check for
1278 * abbreviations. If there was one, discard replay characters.
1279 * If entering a blank character, check for unmap commands,
1280 * as well.
1281 */
1292 }
1293 }
1296 }
1303 /*
1304 * If we're using K_VLNEXT to quote the next character, then
1305 * we want the cursor to position itself on the ^ placeholder
1306 * we're displaying, to match historic practice.
1307 */
1311 }
1313 /*
1314 * !!!
1315 * Translate "<CH_HEX>[isxdigit()]*" to a character with
1316 * a hex value: this test delimits the value by the max
1317 * number of hex bytes. Offset by one, we use 0 to mean
1318 * that we've found <CH_HEX>.
1319 */
1324 }
1326 /*
1327 * Check to see if we've crossed the margin.
1328 *
1329 * !!!
1330 * In the historic vi, the wrapmargin value was figured out
1331 * using the display widths of the characters, i.e. <tab>
1332 * characters were counted as two characters if the list edit
1333 * option is set, but as the tabstop edit option number of
1334 * characters otherwise. That's what the vs_column() function
1335 * gives us, so we use it.
1336 */
1348 }
1349 }
1350 }
1352 /*
1353 * If we've reached the end of the buffer, then we need to
1354 * switch into insert mode. This happens when there's a
1355 * change to a mark and the user puts in more characters than
1356 * the length of the motion.
1357 */
1365 }
1367 /* Step the quote state forward. */
1373 }
1375 }
1377 #ifdef DEBUG
1385 }
1386 #endif
1389 * 1: If we don't need to know where the cursor really is and we're
1390 * replaying text, keep going.
1391 */
1395 /*
1396 * 2: Reset the line. Don't bother unless we're about to wait on
1397 * a character or we need to know where the cursor really is.
1398 * We have to do this before showing matching characters so the
1399 * user can see what they're matching.
1400 */
1405 /*
1406 * 3: If there aren't keys waiting, display the matching character.
1407 * We have to do this before resolving any messages, otherwise
1408 * the error message from a missing match won't appear correctly.
1409 */
1414 }
1416 /*
1417 * 4: If there have been messages and we're not editing on the colon
1418 * command line or doing file name completion, resolve them.
1419 */
1425 /*
1426 * 5: Refresh the screen if we're about to wait on a character or we
1427 * need to know where the cursor really is.
1428 */
1433 }
1435 /* 6: Proceed with the incremental search. */
1439 /* 7: Next character... */
1447 /* If recording for playback, save it. */
1451 /*
1452 * If not working on the colon command line, set the final cursor
1453 * position.
1454 */
1458 }
1461 err:
1462 alloc_err:
1466 }
1468 /*
1469 * txt_abbrev --
1470 * Handle abbreviations.
1471 */
1472 static int
1474 {
1480 /* Check to make sure we're not at the start of an append. */
1487 /*
1488 * Find the start of the "word".
1489 *
1490 * !!!
1491 * We match historic practice, which, as far as I can tell, had an
1492 * off-by-one error. The way this worked was that when the inserted
1493 * text switched from a "word" character to a non-word character,
1494 * vi would check for possible abbreviations. It would then take the
1495 * type (i.e. word/non-word) of the character entered TWO characters
1496 * ago, and move backward in the text until reaching a character that
1497 * was not that type, or the beginning of the insert, the line, or
1498 * the file. For example, in the string "abc<space>", when the <space>
1499 * character triggered the abbreviation check, the type of the 'b'
1500 * character was used for moving through the string. Maybe there's a
1501 * reason for not using the first (i.e. 'c') character, but I can't
1502 * think of one.
1503 *
1504 * Terminate at the beginning of the insert or the character after the
1505 * offset character -- both can be tested for using tp->offset.
1506 */
1517 }
1518 else
1524 }
1526 /*
1527 * !!!
1528 * Historic vi exploded abbreviations on the command line. This has
1529 * obvious problems in that unabbreviating the string can be extremely
1530 * tricky, particularly if the string has, say, an embedded escape
1531 * character. Personally, I think it's a stunningly bad idea. Other
1532 * examples of problems this caused in historic vi are:
1533 * :ab foo bar
1534 * :ab foo baz
1535 * results in "bar" being abbreviated to "baz", which wasn't what the
1536 * user had in mind at all. Also, the commands:
1537 * :ab foo bar
1538 * :unab foo<space>
1539 * resulted in an error message that "bar" wasn't mapped. Finally,
1540 * since the string was already exploded by the time the unabbreviate
1541 * command got it, all it knew was that an abbreviation had occurred.
1542 * Cleverly, it checked the replacement string for its unabbreviation
1543 * match, which meant that the commands:
1544 * :ab foo1 bar
1545 * :ab foo2 bar
1546 * :unab foo2
1547 * unabbreviate "foo1", and the commands:
1548 * :ab foo bar
1549 * :ab bar baz
1550 * unabbreviate "foo"!
1551 *
1552 * Anyway, people neglected to first ask my opinion before they wrote
1553 * macros that depend on this stuff, so, we make this work as follows.
1554 * When checking for an abbreviation on the command line, if we get a
1555 * string which is <blank> terminated and which starts at the beginning
1556 * of the line, we check to see it is the abbreviate or unabbreviate
1557 * commands. If it is, turn abbreviations off and return as if no
1558 * abbreviation was found. Note also, minor trickiness, so that if
1559 * the user erases the line and starts another command, we turn the
1560 * abbreviations back on.
1561 *
1562 * This makes the layering look like a Nachos Supreme.
1563 */
1571 else
1575 /* Check for any abbreviations. */
1579 /*
1580 * Push the abbreviation onto the tty stack. Historically, characters
1581 * resulting from an abbreviation expansion were themselves subject to
1582 * map expansions, O_SHOWMATCH matching etc. This means the expanded
1583 * characters will be re-tested for abbreviations. It's difficult to
1584 * know what historic practice in this case was, since abbreviations
1585 * were applied to :colon command lines, so entering abbreviations that
1586 * looped was tricky, although possible. In addition, obvious loops
1587 * didn't work as expected. (The command ':ab a b|ab b c|ab c a' will
1588 * silently only implement and/or display the last abbreviation.)
1589 *
1590 * This implementation doesn't recover well from such abbreviations.
1591 * The main input loop counts abbreviated characters, and, when it
1592 * reaches a limit, discards any abbreviated characters on the queue.
1593 * It's difficult to back up to the original position, as the replay
1594 * queue would have to be adjusted, and the line state when an initial
1595 * abbreviated character was received would have to be saved.
1596 */
1603 /*
1604 * If the size of the abbreviation is larger than or equal to the size
1605 * of the original text, move to the start of the replaced characters,
1606 * and add their length to the overwrite count.
1607 *
1608 * If the abbreviation is smaller than the original text, we have to
1609 * delete the additional overwrite characters and copy down any insert
1610 * characters.
1611 */
1621 }
1623 /*
1624 * We return the length of the abbreviated characters. This is so
1625 * the calling routine can replace the replay characters with the
1626 * abbreviation. This means that subsequent '.' commands will produce
1627 * the same text, regardless of intervening :[un]abbreviate commands.
1628 * This is historic practice.
1629 */
1632 }
1634 /*
1635 * txt_unmap --
1636 * Handle the unmap command.
1637 */
1638 static void
1640 {
1644 /* Find the beginning of this "word". */
1649 }
1653 }
1655 /*
1656 * !!!
1657 * Historic vi exploded input mappings on the command line. See the
1658 * txt_abbrev() routine for an explanation of the problems inherent
1659 * in this.
1660 *
1661 * We make this work as follows. If we get a string which is <blank>
1662 * terminated and which starts at the beginning of the line, we check
1663 * to see it is the unmap command. If it is, we return that the input
1664 * mapping should be turned off. Note also, minor trickiness, so that
1665 * if the user erases the line and starts another command, we go ahead
1666 * an turn mapping back on.
1667 */
1670 else
1672 }
1674 /*
1675 * txt_ai_resolve --
1676 * When a line is resolved by <esc>, review autoindent characters.
1677 */
1678 static void
1680 {
1681 u_long ts;
1688 /*
1689 * If the line is empty, has an offset, or no autoindent
1690 * characters, we're done.
1691 */
1695 /*
1696 * If the length is less than or equal to the autoindent
1697 * characters, delete them.
1698 */
1702 }
1704 /*
1705 * The autoindent characters plus any leading <blank> characters
1706 * in the line are resolved into the minimum number of characters.
1707 * Historic practice.
1708 */
1711 /* Figure out the last <blank> screen column. */
1721 }
1723 /*
1724 * If there are no spaces, or no tabs after spaces and less than
1725 * ts spaces, it's already minimal.
1726 */
1730 /* Count up spaces/tabs needed to get to the target. */
1735 /*
1736 * Figure out how many characters we're dropping -- if we're not
1737 * dropping any, it's already minimal, we're done.
1738 */
1744 /* Shift the rest of the characters down, adjust the counts. */
1750 /* Fill in space/tab characters. */
1756 }
1758 /*
1759 * v_txt_auto --
1760 * Handle autoindent. If aitp isn't NULL, use it, otherwise,
1761 * retrieve the line.
1762 *
1763 * PUBLIC: int v_txt_auto __P((SCR *, db_recno_t, TEXT *, size_t, TEXT *));
1764 */
1765 int
1767 {
1772 /*
1773 * If the ex append command is executed with an address of 0,
1774 * it's possible to get here with a line number of 0. Return
1775 * an indent of 0.
1776 */
1780 }
1786 /* Count whitespace characters. */
1791 /* Set count, check for no indentation. */
1795 /* Make sure the buffer's big enough. */
1798 /* Copy the buffer's current contents up. */
1803 /* Copy the indentation into the new buffer. */
1806 /* Set the autoindent count. */
1809 }
1811 /*
1812 * txt_backup --
1813 * Back up to the previously edited line.
1814 */
1817 {
1821 /* Get a handle on the previous TEXT structure. */
1827 }
1829 /* Bookkeeping. */
1832 /* Handle appending to the line. */
1842 /* Release the current TEXT. */
1846 /* Update the old line on the screen. */
1850 /* Return the new/current TEXT. */
1852 }
1854 /*
1855 * Text indentation is truly strange. ^T and ^D do movements to the next or
1856 * previous shiftwidth value, i.e. for a 1-based numbering, with shiftwidth=3,
1857 * ^T moves a cursor on the 7th, 8th or 9th column to the 10th column, and ^D
1858 * moves it back.
1859 *
1860 * !!!
1861 * The ^T and ^D characters in historical vi had special meaning only when they
1862 * were the first characters entered after entering text input mode. As normal
1863 * erase characters couldn't erase autoindent characters (^T in this case), it
1864 * meant that inserting text into previously existing text was strange -- ^T
1865 * only worked if it was the first keystroke(s), and then could only be erased
1866 * using ^D. This implementation treats ^T specially anywhere it occurs in the
1867 * input, and permits the standard erase characters to erase the characters it
1868 * inserts.
1869 *
1870 * !!!
1871 * A fun test is to try:
1872 * :se sw=4 ai list
1873 * i<CR>^Tx<CR>^Tx<CR>^Tx<CR>^Dx<CR>^Dx<CR>^Dx<esc>
1874 * Historic vi loses some of the '$' marks on the line ends, but otherwise gets
1875 * it right.
1876 *
1877 * XXX
1878 * Technically, txt_dent should be part of the screen interface, as it requires
1879 * knowledge of character sizes, including <space>s, on the screen. It's here
1880 * because it's a complicated little beast, and I didn't want to shove it down
1881 * into the screen. It's probable that KEY_LEN will call into the screen once
1882 * there are screens with different character representations.
1883 *
1884 * txt_dent --
1885 * Handle ^T indents, ^D outdents.
1886 *
1887 * If anything changes here, check the ex version to see if it needs similar
1888 * changes.
1889 */
1890 static int
1892 {
1893 CHAR_T ch;
1901 /*
1902 * Since we don't know what precedes the character(s) being inserted
1903 * (or deleted), the preceding whitespace characters must be resolved.
1904 * An example is a <tab>, which doesn't need a full shiftwidth number
1905 * of columns because it's preceded by <space>s. This is easy to get
1906 * if the user sets shiftwidth to a value less than tabstop (or worse,
1907 * something for which tabstop isn't a multiple) and then uses ^T to
1908 * indent, and ^D to outdent.
1909 *
1910 * Figure out the current and target screen columns. In the historic
1911 * vi, the autoindent column was NOT determined using display widths
1912 * of characters as was the wrapmargin column. For that reason, we
1913 * can't use the vs_column() function, but have to calculate it here.
1914 * This is slow, but it's normally only on the first few characters of
1915 * a line.
1916 */
1924 else
1927 /*
1928 * The AI characters will be turned into overwrite characters if the
1929 * cursor immediately follows them. We test both the cursor position
1930 * and the indent flag because there's no single test. (^T can only
1931 * be detected by the cursor position, and while we know that the test
1932 * is always true for ^D, the cursor can be in more than one place, as
1933 * "0^D" and "^D" are different.)
1934 */
1937 /*
1938 * Back up over any previous <blank> characters, changing them into
1939 * overwrite characters (including any ai characters). Then figure
1940 * out the current screen column.
1941 */
1949 /*
1950 * If we didn't move up to or past the target, it's because there
1951 * weren't enough characters to delete, e.g. the first character
1952 * of the line was a tp->offset character, and the user entered
1953 * ^D to move to the beginning of a line. An example of this is:
1954 *
1955 * :set ai sw=4<cr>i<space>a<esc>i^T^D
1956 *
1957 * Otherwise, count up the total spaces/tabs needed to get from the
1958 * beginning of the line (or the last non-<blank> character) to the
1959 * target.
1960 */
1968 }
1970 /* If we overwrote ai characters, reset the ai count. */
1974 /*
1975 * Call txt_insch() to insert each character, so that we get the
1976 * correct effect when we add a <tab> to replace N <spaces>.
1977 */
1983 }
1985 /*
1986 * txt_fc --
1987 * File name completion.
1988 */
1989 static int
1991 {
1994 CHAR_T s_ch;
1995 EXCMD cmd;
2005 /*
2006 * Find the beginning of this "word" -- if we're at the beginning
2007 * of the line, it's a special case.
2008 */
2018 }
2022 }
2024 /*
2025 * Get enough space for a wildcard character.
2026 *
2027 * XXX
2028 * This won't work for "foo\", since the \ will escape the expansion
2029 * character. I'm not sure if that's a bug or not...
2030 */
2038 /* Build an ex command, and call the ex expansion routines. */
2043 }
2055 /* If something changed, do the exchange. */
2060 /* If haven't done a directory test, do it now. */
2067 }
2069 /* If nothing changed, period, ring the bell. */
2078 /* Find the length of the shortest match. */
2086 }
2088 }
2090 /* Overwrite the expanded text first. */
2094 /* If lost text, make the remaining old text overwrite characters. */
2098 }
2100 /* Overwrite any overwrite characters next. */
2104 /* Shift remaining text up, and move the cursor to the end. */
2117 }
2119 /* If a single match and it's a directory, retry it. */
2137 }
2139 }
2141 /*
2142 * txt_fc_col --
2143 * Display file names for file name completion.
2144 */
2145 static int
2147 {
2158 /* Trim any directory prefix common to all of the files. */
2167 prefix)) {
2170 }
2171 }
2173 /*
2174 * Figure out the column width for the longest name. Output is done on
2175 * 6 character "tab" boundaries for no particular reason. (Since we
2176 * don't output tab characters, we ignore the terminal's tab settings.)
2177 * Ignore the user's tab setting because we have no idea how reasonable
2178 * it is.
2179 */
2185 }
2188 /*
2189 * Writing to the bottom line of the screen is always turned off when
2190 * SC_TINPUT_INFO is set. Turn it back on, we know what we're doing.
2191 */
2198 #define CHK_INTR \
2199 if (F_ISSET(gp, G_INTERRUPTED)) \
2200 goto intr;
2202 /* If the largest file name is too large, just print them. */
2211 }
2214 CHK_INTR;
2216 /* Figure out the number of columns. */
2225 /* Display the files in sorted order. */
2234 CHK_INTR;
2239 CHK_INTR;
2240 }
2242 CHK_INTR;
2243 }
2245 CHK_INTR;
2246 }
2251 }
2256 }
2258 /*
2259 * txt_emark --
2260 * Set the end mark on the line.
2261 */
2262 static int
2264 {
2265 CHAR_T ch;
2272 /*
2273 * The end mark may not be the same size as the current character.
2274 * Don't let the line shift.
2275 */
2279 else
2282 /*
2283 * If the line got longer, well, it's weird, but it's easy. If
2284 * it's the same length, it's easy. If it got shorter, we have
2285 * to fix it up.
2286 */
2300 else
2304 }
2307 }
2309 /*
2310 * txt_err --
2311 * Handle an error during input processing.
2312 */
2313 static void
2315 {
2316 db_recno_t lno;
2318 /*
2319 * The problem with input processing is that the cursor is at an
2320 * indeterminate position since some input may have been lost due
2321 * to a malloc error. So, try to go back to the place from which
2322 * the cursor started, knowing that it may no longer be available.
2323 *
2324 * We depend on at least one line number being set in the text
2325 * chain.
2326 */
2333 /* Redraw the screen, just in case. */
2335 }
2337 /*
2338 * txt_hex --
2339 * Let the user insert any character value they want.
2340 *
2341 * !!!
2342 * This is an extension. The pattern "^X[0-9a-fA-F]*" is a way
2343 * for the user to specify a character value which their keyboard
2344 * may not be able to enter.
2345 */
2346 static int
2348 {
2349 CHAR_T savec;
2351 u_long value;
2354 /*
2355 * Null-terminate the string. Since nul isn't a legal hex value,
2356 * this should be okay, and lets us use a local routine, which
2357 * presumably understands the character set, to convert the value.
2358 */
2362 /* Find the previous CH_HEX character. */
2367 }
2368 /* Not on this line? Shouldn't happen. */
2371 }
2373 /* If length of 0, then it wasn't a hex value. */
2377 /* Get the value. */
2383 }
2385 /* Restore the original character. */
2388 /* Adjust the bookkeeping. */
2393 /* Copy down any overwrite characters. */
2398 /* Copy down any insert characters. */
2405 }
2407 /*
2408 * txt_insch --
2409 *
2410 * !!!
2411 * Historic vi did a special screen optimization for tab characters. As an
2412 * example, for the keystrokes "iabcd<esc>0C<tab>", the tab overwrote the
2413 * rest of the string when it was displayed.
2414 *
2415 * Because early versions of this implementation redisplayed the entire line
2416 * on each keystroke, the "bcd" was pushed to the right as it ignored that
2417 * the user had "promised" to change the rest of the characters. However,
2418 * the historic vi implementation had an even worse bug: given the keystrokes
2419 * "iabcd<esc>0R<tab><esc>", the "bcd" disappears, and magically reappears
2420 * on the second <esc> key.
2421 *
2422 * POSIX 1003.2 requires (will require) that this be fixed, specifying that
2423 * vi overwrite characters the user has committed to changing, on the basis
2424 * of the screen space they require, but that it not overwrite other characters.
2425 */
2426 static int
2428 {
2430 CHAR_T savech;
2434 /*
2435 * The 'R' command does one-for-one replacement, because there's
2436 * no way to know how many characters the user intends to replace.
2437 */
2443 }
2447 /*
2448 * If the old or new characters are tabs, then the length of the
2449 * display depends on the character position in the display. We
2450 * don't even try to handle this here, just ask the screen.
2451 */
2460 /*
2461 * Eat overwrite characters until we run out of them or we've
2462 * handled the length of the new character. If we only eat
2463 * part of an overwrite character, break it into its component
2464 * elements and display the remaining components.
2465 */
2472 else
2478 }
2495 else
2502 }
2503 }
2505 /*
2506 * If had to erase several characters, we adjust the total
2507 * count, and if there are any characters left, shift them
2508 * into position.
2509 */
2514 /* If we had enough overwrite characters, we're done. */
2518 }
2519 }
2521 /* Check to see if the character fits into the input buffer. */
2528 else
2531 }
2534 }
2536 /*
2537 * txt_isrch --
2538 * Do an incremental search.
2539 */
2540 static int
2542 {
2543 MARK start;
2544 db_recno_t lno;
2545 u_int sf;
2547 /* If it's a one-line screen, we don't do incrementals. */
2551 }
2553 /*
2554 * If the user erases back to the beginning of the buffer, there's
2555 * nothing to search for. Reset the cursor to the starting point.
2556 */
2560 }
2562 /*
2563 * If it's an RE quote character, and not quoted, ignore it until
2564 * we get another character.
2565 */
2570 /*
2571 * If it's a magic shell character, and not quoted, reset the cursor
2572 * to the starting point.
2573 */
2578 /*
2579 * If we see the search pattern termination character, then quit doing
2580 * an incremental search. There may be more, e.g., ":/foo/;/bar/",
2581 * and we can't handle that incrementally. Also, reset the cursor to
2582 * the original location, the ex search routines don't know anything
2583 * about incremental searches.
2584 */
2590 }
2592 /*
2593 * Remember the input line and discard the special input map,
2594 * but don't overwrite the input line on the screen.
2595 */
2602 /*
2603 * Specify a starting point and search. If we find a match, move to
2604 * it and refresh the screen. If we didn't find the match, then we
2605 * beep the screen. When searching from the original cursor position,
2606 * we have to move the cursor, otherwise, we don't want to move the
2607 * cursor in case the text at the current position continues to match.
2608 */
2615 }
2631 /* Reinstantiate the special input map. */
2637 /* Reset the line number of the input line. */
2640 /*
2641 * If the colon command-line moved, i.e. the screen scrolled,
2642 * refresh the input line.
2643 *
2644 * XXX
2645 * We shouldn't be calling vs_line, here -- we need dirty bits
2646 * on entries in the SMAP array.
2647 */
2652 }
2654 }
2656 /*
2657 * txt_resolve --
2658 * Resolve the input text chain into the file.
2659 */
2660 static int
2662 {
2665 db_recno_t lno;
2668 /*
2669 * The first line replaces a current line, and all subsequent lines
2670 * are appended into the file. Resolve autoindented characters for
2671 * each line before committing it. If the latter causes the line to
2672 * change, we have to redisplay it, otherwise the information cached
2673 * about the line will be wrong.
2674 */
2680 else
2689 else
2694 }
2696 /*
2697 * Clear the input flag, the look-aside buffer is no longer valid.
2698 * Has to be done as part of text resolution, or upon return we'll
2699 * be looking at incorrect data.
2700 */
2704 }
2706 /*
2707 * txt_showmatch --
2708 * Show a character match.
2709 *
2710 * !!!
2711 * Historic vi tried to display matches even in the :colon command line.
2712 * I think not.
2713 */
2714 static int
2716 {
2718 VCS cs;
2719 MARK m;
2724 /*
2725 * Do a refresh first, in case we haven't done one in awhile,
2726 * so the user can see what we're complaining about.
2727 */
2732 /*
2733 * We don't display the match if it's not on the screen. Find
2734 * out what the first character on the screen is.
2735 */
2739 /* Initialize the getc() interface. */
2746 /* Search for the match. */
2755 }
2757 }
2762 }
2764 /* If the match is on the screen, move to it. */
2772 /* Wait for timeout or character arrival. */
2775 }
2777 /*
2778 * txt_margin --
2779 * Handle margin wrap.
2780 */
2781 static int
2783 {
2788 /* Find the nearest previous blank. */
2793 }
2795 /*
2796 * If reach the start of the line, there's nowhere to break.
2797 *
2798 * !!!
2799 * Historic vi belled each time a character was entered after
2800 * crossing the margin until a space was entered which could
2801 * be used to break the line. I don't as it tends to wake the
2802 * cats.
2803 */
2807 }
2808 }
2810 /*
2811 * Store saved information about the rest of the line in the
2812 * wrapmargin TEXT structure.
2813 *
2814 * !!!
2815 * The offset field holds the length of the current characters
2816 * that the user entered, but which are getting split to the new
2817 * line -- it's going to be used to set the cursor value when we
2818 * move to the new line.
2819 */
2826 /* Correct current bookkeeping information. */
2834 }
2837 /*
2838 * !!!
2839 * Delete any trailing whitespace from the current line.
2840 */
2848 }
2851 }
2853 /*
2854 * txt_Rresolve --
2855 * Resolve the input line for the 'R' command.
2856 */
2857 static void
2859 {
2864 /*
2865 * Check to make sure that the cursor hasn't moved beyond
2866 * the end of the line.
2867 */
2871 /*
2872 * Calculate how many characters the user has entered,
2873 * plus the blanks erased by <carriage-return>/<newline>s.
2874 */
2879 }
2881 /*
2882 * If the user has entered less characters than the original line
2883 * was long, restore any overwriteable characters to the original
2884 * characters. These characters are entered as "insert characters",
2885 * because they're after the cursor and we don't want to lose them.
2886 * (This is okay because the R command has no insert characters.)
2887 * We set owrite to 0 so that the insert characters don't get copied
2888 * to somewhere else, which means that the line and the length have
2889 * to be adjusted here as well.
2890 *
2891 * We have to retrieve the original line because the original pinned
2892 * page has long since been discarded. If it doesn't exist, that's
2893 * okay, the user just extended the file.
2894 */
2904 }
2905 }
2907 /*
2908 * txt_nomorech --
2909 * No more characters message.
2910 */
2911 static void
2913 {
2915 }