| blob | e405a3bca74a6fb8c8d1f6d412fced183fee20f3 |
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 10.99 2000/07/21 17:35:05 skimo Exp $ (Berkeley) $Date: 2000/07/21 17:35:05 $";
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
69 ARG_CHAR_T prompt;
70 u_int flags;
71 {
72 /* Normally, we end up where we started. */
76 /* Initialize the map. */
80 /* Move to the last line. */
84 /* Don't update the modeline for now. */
87 /* Set the input flags. */
95 /* Do the input thing. */
99 /* Reenable the modeline updates. */
102 /* Clean up the map. */
109 /* Set the cursor to the resulting position. */
114 }
116 /*
117 * txt_map_init
118 * Initialize the screen map for colon command-line input.
119 */
120 static int
123 {
129 /*
130 * Fake like the user is doing input on the last line of the
131 * screen. This makes all of the scrolling work correctly,
132 * and allows us the use of the vi text editing routines, not
133 * to mention practically infinite length ex commands.
134 *
135 * Save the current location.
136 */
144 /*
145 * If it's a small screen, TMAP may be small for the screen.
146 * Fix it, filling in fake lines as we go.
147 */
154 }
156 /* Build the fake entry. */
163 /* Reset the screen information. */
165 }
167 }
169 /*
170 * txt_map_end
171 * Reset the screen map for colon command-line input.
172 */
173 static int
176 {
182 /* Restore the screen information. */
187 /*
188 * If it's a small screen, TMAP may be wrong. Clear any
189 * lines that might have been overwritten.
190 */
195 }
200 /*
201 * The map may be wrong if the user entered more than one
202 * (logical) line. Fix it. If the user entered a whole
203 * screen, this will be slow, but we probably don't care.
204 */
210 }
212 /*
213 * Invalidate the cursor and the line size cache, the line never
214 * really existed. This fixes bugs where the user searches for
215 * the last line on the screen + 1 and the refresh routine thinks
216 * that's where we just were.
217 */
222 }
224 /*
225 * If doing input mapping on the colon command line, may need to unmap
226 * based on the command.
227 */
228 #define UNMAP_TST \
229 FL_ISSET(ec_flags, EC_MAPINPUT) && LF_ISSET(TXT_INFOLINE)
231 /*
232 * Internally, we maintain tp->lno and tp->cno, externally, everyone uses
233 * sp->lno and sp->cno. Make them consistent as necessary.
234 */
235 #define UPDATE_POSITION(sp, tp) { \
236 (sp)->lno = (tp)->lno; \
237 (sp)->cno = (tp)->cno; \
238 }
240 /*
241 * v_txt --
242 * Vi text input.
243 *
244 * PUBLIC: int v_txt __P((SCR *, VICMD *, MARK *,
245 * PUBLIC: const CHAR_T *, size_t, ARG_CHAR_T, db_recno_t, u_long, u_int32_t));
246 */
247 int
258 {
277 u_int8_t is_flags;
289 /*
290 * Set the input flag, so tabs get displayed correctly
291 * and everyone knows that the text buffer is in use.
292 */
295 /*
296 * Get one TEXT structure with some initial buffer space, reusing
297 * the last one if it's big enough. (All TEXT bookkeeping fields
298 * default to 0 -- text_init() handles this.) If changing a line,
299 * copy it into the TEXT buffer.
300 */
307 }
318 }
320 /* Set default termination condition. */
323 /* Set the starting line, column. */
327 /*
328 * Set the insert and overwrite counts. If overwriting characters,
329 * do insertion afterward. If not overwriting characters, assume
330 * doing insertion. If change is to a mark, emphasize it with an
331 * CH_ENDMARK character.
332 */
342 }
344 /*
345 * Many of the special cases in text input are to handle autoindent
346 * support. Somebody decided that it would be a good idea if "^^D"
347 * and "0^D" deleted all of the autoindented characters. In an editor
348 * that takes single character input from the user, this beggars the
349 * imagination. Note also, "^^D" resets the next lines' autoindent,
350 * but "0^D" doesn't.
351 *
352 * We assume that autoindent only happens on empty lines, so insert
353 * and overwrite will be zero. If doing autoindent, figure out how
354 * much indentation we need and fill it in. Update input column and
355 * screen cursor as necessary.
356 */
362 /*
363 * The cc and S commands have a special feature -- leading
364 * <blank> characters are handled as autoindent characters.
365 * Beauty!
366 */
372 }
374 /* If getting a command buffer from the user, there may be a prompt. */
379 }
381 /*
382 * If appending after the end-of-line, add a space into the buffer
383 * and move the cursor right. This space is inserted, i.e. pushed
384 * along, and then deleted when the line is resolved. Assumes that
385 * the cursor is already positioned at the end of the line. This
386 * avoids the nastiness of having the cursor reside on a magical
387 * column, i.e. a column that doesn't really exist. The only down
388 * side is that we may wrap lines or scroll the screen before it's
389 * strictly necessary. Not a big deal.
390 */
396 }
398 /*
399 * Historic practice is that the wrapmargin value was a distance
400 * from the RIGHT-HAND margin, not the left. It's more useful to
401 * us as a distance from the left-hand margin, i.e. the same as
402 * the wraplen value. The wrapmargin option is historic practice.
403 * Nvi added the wraplen option so that it would be possible to
404 * edit files with consistent margins without knowing the number of
405 * columns in the window.
406 *
407 * XXX
408 * Setting margin causes a significant performance hit. Normally
409 * we don't update the screen if there are keys waiting, but we
410 * have to if margin is set, otherwise the screen routines don't
411 * know where the cursor is.
412 *
413 * !!!
414 * Abbreviated keys were affected by the wrapmargin option in the
415 * historic 4BSD vi. Mapped keys were usually, but sometimes not.
416 * See the comment in vi/v_text():set_txt_std for more information.
417 *
418 * !!!
419 * One more special case. If an inserted <blank> character causes
420 * wrapmargin to split the line, the next user entered character is
421 * discarded if it's a <space> character.
422 */
427 else
429 else
432 /* Initialize abbreviation checks. */
437 /*
438 * Set up the dot command. Dot commands are done by saving the actual
439 * characters and then reevaluating them so that things like wrapmargin
440 * can change between the insert and the replay.
441 *
442 * !!!
443 * Historically, vi did not remap or reabbreviate replayed input. (It
444 * did beep at you if you changed an abbreviation and then replayed the
445 * input. We're not that compatible.) We don't have to do anything to
446 * avoid remapping, as we're not getting characters from the terminal
447 * routines. Turn the abbreviation check off.
448 *
449 * XXX
450 * It would be nice if we could swallow backspaces and such, but it's
451 * not all that easy to do. What we can do is turn off the common
452 * error messages during the replay. Otherwise, when the user enters
453 * an illegal command, e.g., "Ia<erase><erase><erase><erase>b<escape>",
454 * and then does a '.', they get a list of error messages after command
455 * completion.
456 */
461 }
463 /* Other text input mode setup. */
470 /* Initialize input flags. */
473 /* Refresh the screen. */
478 /* If it's dot, just do it now. */
482 /* Get an event. */
487 /*
488 * If file completion overwrote part of the screen and nothing else has
489 * been displayed, clean up. We don't do this as part of the normal
490 * message resolution because we know the user is on the colon command
491 * line and there's no reason to enter explicit characters to continue.
492 */
503 }
505 /* Deal with all non-character events. */
514 /*
515 * !!!
516 * Historically, <interrupt> exited the user from text input
517 * mode or cancelled a colon command, and returned to command
518 * mode. It also beeped the terminal, but that seems a bit
519 * excessive.
520 */
527 /* <resize> interrupts the input mode. */
533 }
535 /*
536 * !!!
537 * If the first character of the input is a nul, replay the previous
538 * input. (Historically, it's okay to replay non-existent input.)
539 * This was not documented as far as I know, and is a great test of vi
540 * clones.
541 */
550 }
552 /*
553 * File name completion and colon command-line editing. We don't
554 * have enough meta characters, so we expect people to overload
555 * them. If the two characters are the same, then we do file name
556 * completion if the cursor is past the first column, and do colon
557 * command-line editing if it's not.
558 */
572 }
577 }
578 }
580 /* Abbreviation overflow check. See comment in txt_abbrev(). */
581 #define MAX_ABBREVIATION_EXPANSION 256
591 }
595 /* Check to see if the character fits into the replay buffers. */
600 }
606 }
608 /* Wrapmargin check for leading space. */
613 }
615 /* If quoted by someone else, simply insert the character. */
619 /*
620 * !!!
621 * If this character was quoted by a K_VLNEXT or a backslash, replace
622 * the placeholder (a carat or a backslash) with the new character.
623 * If it was quoted by a K_VLNEXT, we've already adjusted the cursor
624 * because it has to appear on top of the placeholder character. If
625 * it was quoted by a backslash, adjust the cursor now, the cursor
626 * doesn't appear on top of it. Historic practice in both cases.
627 *
628 * Skip tests for abbreviations; ":ab xa XA" followed by "ixa^V<space>"
629 * doesn't perform an abbreviation. Special case, ^V^J (not ^V^M) is
630 * the same as ^J, historically.
631 */
643 }
647 }
649 }
651 /*
652 * !!!
653 * Translate "<CH_HEX>[isxdigit()]*" to a character with a hex value:
654 * this test delimits the value by any non-hex character. Offset by
655 * one, we use 0 to mean that we've found <CH_HEX>.
656 */
661 }
666 /* Return in script windows and the command line. */
669 /*
670 * If this was a map, we may have not displayed
671 * the line. Display it, just in case.
672 *
673 * If a script window and not the colon line,
674 * push a <cr> so it gets executed.
675 */
682 /* Set term condition: if empty. */
685 /*
686 * Set term condition: if searching incrementally and
687 * the user entered a pattern, return a completed
688 * search, regardless if the entire pattern was found.
689 */
695 }
697 #define LINE_RESOLVE { \
698 /* \
699 * Handle abbreviations. If there was one, discard the \
700 * replay characters. \
702 if (abb == AB_INWORD && \
703 !LF_ISSET(TXT_REPLAY) && F_ISSET(gp, G_ABBREV)) { \
704 if (txt_abbrev(sp, tp, &evp->e_c, \
705 LF_ISSET(TXT_INFOLINE), &tmp, \
706 &ab_turnoff)) \
707 goto err; \
708 if (tmp) { \
709 if (LF_ISSET(TXT_RECORD)) \
710 rcol -= tmp + 1; \
711 goto resolve; \
712 } \
713 } \
714 if (abb != AB_NOTSET) \
715 abb = AB_NOTWORD; \
716 if (UNMAP_TST) \
717 txt_unmap(sp, tp, &ec_flags); \
718 /* \
719 * Delete any appended cursor. It's possible to get in \
720 * situations where TXT_APPENDEOL is set but tp->insert \
721 * is 0 when using the R command and all the characters \
722 * are tp->owrite characters. \
724 if (LF_ISSET(TXT_APPENDEOL) && tp->insert > 0) { \
725 --tp->len; \
726 --tp->insert; \
727 } \
728 }
729 LINE_RESOLVE;
731 /*
732 * Save the current line information for restoration in
733 * txt_backup(), and set the line final length.
734 */
739 /* Update the old line. */
743 /*
744 * Historic practice, when the autoindent edit option was set,
745 * was to delete <blank> characters following the inserted
746 * newline. This affected the 'R', 'c', and 's' commands; 'c'
747 * and 's' retained the insert characters only, 'R' moved the
748 * overwrite and insert characters into the next TEXT structure.
749 * We keep track of the number of characters erased for the 'R'
750 * command so that the final resolution of the line is correct.
751 */
767 }
769 /*
770 * !!!
771 * Create a new line and insert the new TEXT into the queue.
772 * DON'T insert until the old line has been updated, or the
773 * inserted line count in line.c:db_get() will be wrong.
774 */
780 /* Set up bookkeeping for the new line. */
785 /*
786 * Reset the autoindent line value. 0^D keeps the autoindent
787 * line from changing, ^D changes the level, even if there were
788 * no characters in the old line. Note, if using the current
789 * tp structure, use the cursor as the length, the autoindent
790 * characters may have been erased.
791 */
801 }
803 /* Reset the cursor. */
806 /*
807 * If we're here because wrapmargin was set and we've broken a
808 * line, there may be additional information (i.e. the start of
809 * a line) in the wmt structure.
810 */
814 #define WMTSPACE wmt.offset + wmt.owrite + wmt.insert
822 }
824 }
826 /* New lines are TXT_APPENDEOL. */
833 }
835 /* Swap old and new TEXT's, and update the new line. */
845 /* If we have a count, start replaying the input. */
854 /*
855 * Some commands (e.g. 'o') need a <newline> for each
856 * repetition.
857 */
861 /*
862 * The R command turns into the 'a' command after the
863 * first repetition.
864 */
869 }
871 }
873 /* Set term condition: if empty. */
876 /*
877 * Set term condition: if searching incrementally and the user
878 * entered a pattern, return a completed search, regardless if
879 * the entire pattern was found.
880 */
886 /*
887 * Clean up for the 'R' command, restoring overwrite
888 * characters, and making them into insert characters.
889 */
893 /*
894 * If there are any overwrite characters, copy down
895 * any insert characters, and decrement the length.
896 */
902 }
904 /*
905 * Optionally resolve the lines into the file. If not
906 * resolving the lines into the file, end the line with
907 * a nul. If the line is empty, then set the length to
908 * 0, the termination condition has already been set.
909 *
910 * XXX
911 * This is wrong, should pass back a length.
912 */
919 }
921 /*
922 * Set the return cursor position to rest on the last
923 * inserted character.
924 */
928 /* Update the last line. */
941 /*
942 * If in the first column or no characters to erase, ignore
943 * the ^D (this matches historic practice). If not doing
944 * autoindent or already inserted non-ai characters, it's a
945 * literal. The latter test is done in the switch, as the
946 * CARAT forms are N + 1, not N.
947 */
958 /* 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:
1465 }
1467 /*
1468 * txt_abbrev --
1469 * Handle abbreviations.
1470 */
1471 static int
1477 {
1483 /* Check to make sure we're not at the start of an append. */
1490 /*
1491 * Find the start of the "word".
1492 *
1493 * !!!
1494 * We match historic practice, which, as far as I can tell, had an
1495 * off-by-one error. The way this worked was that when the inserted
1496 * text switched from a "word" character to a non-word character,
1497 * vi would check for possible abbreviations. It would then take the
1498 * type (i.e. word/non-word) of the character entered TWO characters
1499 * ago, and move backward in the text until reaching a character that
1500 * was not that type, or the beginning of the insert, the line, or
1501 * the file. For example, in the string "abc<space>", when the <space>
1502 * character triggered the abbreviation check, the type of the 'b'
1503 * character was used for moving through the string. Maybe there's a
1504 * reason for not using the first (i.e. 'c') character, but I can't
1505 * think of one.
1506 *
1507 * Terminate at the beginning of the insert or the character after the
1508 * offset character -- both can be tested for using tp->offset.
1509 */
1520 }
1521 else
1527 }
1529 /*
1530 * !!!
1531 * Historic vi exploded abbreviations on the command line. This has
1532 * obvious problems in that unabbreviating the string can be extremely
1533 * tricky, particularly if the string has, say, an embedded escape
1534 * character. Personally, I think it's a stunningly bad idea. Other
1535 * examples of problems this caused in historic vi are:
1536 * :ab foo bar
1537 * :ab foo baz
1538 * results in "bar" being abbreviated to "baz", which wasn't what the
1539 * user had in mind at all. Also, the commands:
1540 * :ab foo bar
1541 * :unab foo<space>
1542 * resulted in an error message that "bar" wasn't mapped. Finally,
1543 * since the string was already exploded by the time the unabbreviate
1544 * command got it, all it knew was that an abbreviation had occurred.
1545 * Cleverly, it checked the replacement string for its unabbreviation
1546 * match, which meant that the commands:
1547 * :ab foo1 bar
1548 * :ab foo2 bar
1549 * :unab foo2
1550 * unabbreviate "foo1", and the commands:
1551 * :ab foo bar
1552 * :ab bar baz
1553 * unabbreviate "foo"!
1554 *
1555 * Anyway, people neglected to first ask my opinion before they wrote
1556 * macros that depend on this stuff, so, we make this work as follows.
1557 * When checking for an abbreviation on the command line, if we get a
1558 * string which is <blank> terminated and which starts at the beginning
1559 * of the line, we check to see it is the abbreviate or unabbreviate
1560 * commands. If it is, turn abbreviations off and return as if no
1561 * abbreviation was found. Note also, minor trickiness, so that if
1562 * the user erases the line and starts another command, we turn the
1563 * abbreviations back on.
1564 *
1565 * This makes the layering look like a Nachos Supreme.
1566 */
1574 else
1578 /* Check for any abbreviations. */
1582 /*
1583 * Push the abbreviation onto the tty stack. Historically, characters
1584 * resulting from an abbreviation expansion were themselves subject to
1585 * map expansions, O_SHOWMATCH matching etc. This means the expanded
1586 * characters will be re-tested for abbreviations. It's difficult to
1587 * know what historic practice in this case was, since abbreviations
1588 * were applied to :colon command lines, so entering abbreviations that
1589 * looped was tricky, although possible. In addition, obvious loops
1590 * didn't work as expected. (The command ':ab a b|ab b c|ab c a' will
1591 * silently only implement and/or display the last abbreviation.)
1592 *
1593 * This implementation doesn't recover well from such abbreviations.
1594 * The main input loop counts abbreviated characters, and, when it
1595 * reaches a limit, discards any abbreviated characters on the queue.
1596 * It's difficult to back up to the original position, as the replay
1597 * queue would have to be adjusted, and the line state when an initial
1598 * abbreviated character was received would have to be saved.
1599 */
1606 /*
1607 * If the size of the abbreviation is larger than or equal to the size
1608 * of the original text, move to the start of the replaced characters,
1609 * and add their length to the overwrite count.
1610 *
1611 * If the abbreviation is smaller than the original text, we have to
1612 * delete the additional overwrite characters and copy down any insert
1613 * characters.
1614 */
1624 }
1626 /*
1627 * We return the length of the abbreviated characters. This is so
1628 * the calling routine can replace the replay characters with the
1629 * abbreviation. This means that subsequent '.' commands will produce
1630 * the same text, regardless of intervening :[un]abbreviate commands.
1631 * This is historic practice.
1632 */
1635 }
1637 /*
1638 * txt_unmap --
1639 * Handle the unmap command.
1640 */
1641 static void
1646 {
1650 /* Find the beginning of this "word". */
1655 }
1659 }
1661 /*
1662 * !!!
1663 * Historic vi exploded input mappings on the command line. See the
1664 * txt_abbrev() routine for an explanation of the problems inherent
1665 * in this.
1666 *
1667 * We make this work as follows. If we get a string which is <blank>
1668 * terminated and which starts at the beginning of the line, we check
1669 * to see it is the unmap command. If it is, we return that the input
1670 * mapping should be turned off. Note also, minor trickiness, so that
1671 * if the user erases the line and starts another command, we go ahead
1672 * an turn mapping back on.
1673 */
1676 else
1678 }
1680 /*
1681 * txt_ai_resolve --
1682 * When a line is resolved by <esc>, review autoindent characters.
1683 */
1684 static void
1689 {
1690 u_long ts;
1697 /*
1698 * If the line is empty, has an offset, or no autoindent
1699 * characters, we're done.
1700 */
1704 /*
1705 * If the length is less than or equal to the autoindent
1706 * characters, delete them.
1707 */
1711 }
1713 /*
1714 * The autoindent characters plus any leading <blank> characters
1715 * in the line are resolved into the minimum number of characters.
1716 * Historic practice.
1717 */
1720 /* Figure out the last <blank> screen column. */
1730 }
1732 /*
1733 * If there are no spaces, or no tabs after spaces and less than
1734 * ts spaces, it's already minimal.
1735 */
1739 /* Count up spaces/tabs needed to get to the target. */
1744 /*
1745 * Figure out how many characters we're dropping -- if we're not
1746 * dropping any, it's already minimal, we're done.
1747 */
1753 /* Shift the rest of the characters down, adjust the counts. */
1759 /* Fill in space/tab characters. */
1765 }
1767 /*
1768 * v_txt_auto --
1769 * Handle autoindent. If aitp isn't NULL, use it, otherwise,
1770 * retrieve the line.
1771 *
1772 * PUBLIC: int v_txt_auto __P((SCR *, db_recno_t, TEXT *, size_t, TEXT *));
1773 */
1774 int
1777 db_recno_t lno;
1780 {
1785 /*
1786 * If the ex append command is executed with an address of 0,
1787 * it's possible to get here with a line number of 0. Return
1788 * an indent of 0.
1789 */
1793 }
1799 /* Count whitespace characters. */
1804 /* Set count, check for no indentation. */
1808 /* Make sure the buffer's big enough. */
1811 /* Copy the buffer's current contents up. */
1816 /* Copy the indentation into the new buffer. */
1819 /* Set the autoindent count. */
1822 }
1824 /*
1825 * txt_backup --
1826 * Back up to the previously edited line.
1827 */
1834 {
1838 /* Get a handle on the previous TEXT structure. */
1844 }
1846 /* Bookkeeping. */
1849 /* Handle appending to the line. */
1859 /* Release the current TEXT. */
1863 /* Update the old line on the screen. */
1867 /* Return the new/current TEXT. */
1869 }
1871 /*
1872 * Text indentation is truly strange. ^T and ^D do movements to the next or
1873 * previous shiftwidth value, i.e. for a 1-based numbering, with shiftwidth=3,
1874 * ^T moves a cursor on the 7th, 8th or 9th column to the 10th column, and ^D
1875 * moves it back.
1876 *
1877 * !!!
1878 * The ^T and ^D characters in historical vi had special meaning only when they
1879 * were the first characters entered after entering text input mode. As normal
1880 * erase characters couldn't erase autoindent characters (^T in this case), it
1881 * meant that inserting text into previously existing text was strange -- ^T
1882 * only worked if it was the first keystroke(s), and then could only be erased
1883 * using ^D. This implementation treats ^T specially anywhere it occurs in the
1884 * input, and permits the standard erase characters to erase the characters it
1885 * inserts.
1886 *
1887 * !!!
1888 * A fun test is to try:
1889 * :se sw=4 ai list
1890 * i<CR>^Tx<CR>^Tx<CR>^Tx<CR>^Dx<CR>^Dx<CR>^Dx<esc>
1891 * Historic vi loses some of the '$' marks on the line ends, but otherwise gets
1892 * it right.
1893 *
1894 * XXX
1895 * Technically, txt_dent should be part of the screen interface, as it requires
1896 * knowledge of character sizes, including <space>s, on the screen. It's here
1897 * because it's a complicated little beast, and I didn't want to shove it down
1898 * into the screen. It's probable that KEY_LEN will call into the screen once
1899 * there are screens with different character representations.
1900 *
1901 * txt_dent --
1902 * Handle ^T indents, ^D outdents.
1903 *
1904 * If anything changes here, check the ex version to see if it needs similar
1905 * changes.
1906 */
1907 static int
1912 {
1913 CHAR_T ch;
1921 /*
1922 * Since we don't know what precedes the character(s) being inserted
1923 * (or deleted), the preceding whitespace characters must be resolved.
1924 * An example is a <tab>, which doesn't need a full shiftwidth number
1925 * of columns because it's preceded by <space>s. This is easy to get
1926 * if the user sets shiftwidth to a value less than tabstop (or worse,
1927 * something for which tabstop isn't a multiple) and then uses ^T to
1928 * indent, and ^D to outdent.
1929 *
1930 * Figure out the current and target screen columns. In the historic
1931 * vi, the autoindent column was NOT determined using display widths
1932 * of characters as was the wrapmargin column. For that reason, we
1933 * can't use the vs_column() function, but have to calculate it here.
1934 * This is slow, but it's normally only on the first few characters of
1935 * a line.
1936 */
1944 else
1947 /*
1948 * The AI characters will be turned into overwrite characters if the
1949 * cursor immediately follows them. We test both the cursor position
1950 * and the indent flag because there's no single test. (^T can only
1951 * be detected by the cursor position, and while we know that the test
1952 * is always true for ^D, the cursor can be in more than one place, as
1953 * "0^D" and "^D" are different.)
1954 */
1957 /*
1958 * Back up over any previous <blank> characters, changing them into
1959 * overwrite characters (including any ai characters). Then figure
1960 * out the current screen column.
1961 */
1969 /*
1970 * If we didn't move up to or past the target, it's because there
1971 * weren't enough characters to delete, e.g. the first character
1972 * of the line was a tp->offset character, and the user entered
1973 * ^D to move to the beginning of a line. An example of this is:
1974 *
1975 * :set ai sw=4<cr>i<space>a<esc>i^T^D
1976 *
1977 * Otherwise, count up the total spaces/tabs needed to get from the
1978 * beginning of the line (or the last non-<blank> character) to the
1979 * target.
1980 */
1988 }
1990 /* If we overwrote ai characters, reset the ai count. */
1994 /*
1995 * Call txt_insch() to insert each character, so that we get the
1996 * correct effect when we add a <tab> to replace N <spaces>.
1997 */
2003 }
2005 /*
2006 * txt_fc --
2007 * File name completion.
2008 */
2009 static int
2014 {
2017 CHAR_T s_ch;
2018 EXCMD cmd;
2028 /*
2029 * Find the beginning of this "word" -- if we're at the beginning
2030 * of the line, it's a special case.
2031 */
2041 }
2045 }
2047 /*
2048 * Get enough space for a wildcard character.
2049 *
2050 * XXX
2051 * This won't work for "foo\", since the \ will escape the expansion
2052 * character. I'm not sure if that's a bug or not...
2053 */
2061 /* Build an ex command, and call the ex expansion routines. */
2066 }
2078 /* If something changed, do the exchange. */
2083 /* If haven't done a directory test, do it now. */
2090 }
2092 /* If nothing changed, period, ring the bell. */
2101 /* Find the length of the shortest match. */
2109 }
2111 }
2113 /* Overwrite the expanded text first. */
2117 /* If lost text, make the remaining old text overwrite characters. */
2121 }
2123 /* Overwrite any overwrite characters next. */
2127 /* Shift remaining text up, and move the cursor to the end. */
2140 }
2142 /* If a single match and it's a directory, retry it. */
2160 }
2162 }
2164 /*
2165 * txt_fc_col --
2166 * Display file names for file name completion.
2167 */
2168 static int
2173 {
2184 /* Trim any directory prefix common to all of the files. */
2193 prefix)) {
2196 }
2197 }
2199 /*
2200 * Figure out the column width for the longest name. Output is done on
2201 * 6 character "tab" boundaries for no particular reason. (Since we
2202 * don't output tab characters, we ignore the terminal's tab settings.)
2203 * Ignore the user's tab setting because we have no idea how reasonable
2204 * it is.
2205 */
2211 }
2214 /*
2215 * Writing to the bottom line of the screen is always turned off when
2216 * SC_TINPUT_INFO is set. Turn it back on, we know what we're doing.
2217 */
2224 #define CHK_INTR \
2225 if (F_ISSET(gp, G_INTERRUPTED)) \
2226 goto intr;
2228 /* If the largest file name is too large, just print them. */
2237 }
2240 CHK_INTR;
2242 /* Figure out the number of columns. */
2251 /* Display the files in sorted order. */
2260 CHK_INTR;
2265 CHK_INTR;
2266 }
2268 CHK_INTR;
2269 }
2271 CHK_INTR;
2272 }
2277 }
2282 }
2284 /*
2285 * txt_emark --
2286 * Set the end mark on the line.
2287 */
2288 static int
2293 {
2294 CHAR_T ch;
2301 /*
2302 * The end mark may not be the same size as the current character.
2303 * Don't let the line shift.
2304 */
2308 else
2311 /*
2312 * If the line got longer, well, it's weird, but it's easy. If
2313 * it's the same length, it's easy. If it got shorter, we have
2314 * to fix it up.
2315 */
2329 else
2333 }
2336 }
2338 /*
2339 * txt_err --
2340 * Handle an error during input processing.
2341 */
2342 static void
2346 {
2347 db_recno_t lno;
2349 /*
2350 * The problem with input processing is that the cursor is at an
2351 * indeterminate position since some input may have been lost due
2352 * to a malloc error. So, try to go back to the place from which
2353 * the cursor started, knowing that it may no longer be available.
2354 *
2355 * We depend on at least one line number being set in the text
2356 * chain.
2357 */
2364 /* Redraw the screen, just in case. */
2366 }
2368 /*
2369 * txt_hex --
2370 * Let the user insert any character value they want.
2371 *
2372 * !!!
2373 * This is an extension. The pattern "^X[0-9a-fA-F]*" is a way
2374 * for the user to specify a character value which their keyboard
2375 * may not be able to enter.
2376 */
2377 static int
2381 {
2382 CHAR_T savec;
2384 u_long value;
2389 /*
2390 * Null-terminate the string. Since nul isn't a legal hex value,
2391 * this should be okay, and lets us use a local routine, which
2392 * presumably understands the character set, to convert the value.
2393 */
2397 /* Find the previous CH_HEX character. */
2402 }
2403 /* Not on this line? Shouldn't happen. */
2406 }
2408 /* If length of 0, then it wasn't a hex value. */
2412 /* Get the value. */
2419 }
2421 /* Restore the original character. */
2424 /* Adjust the bookkeeping. */
2429 /* Copy down any overwrite characters. */
2434 /* Copy down any insert characters. */
2441 }
2443 /*
2444 * txt_insch --
2445 *
2446 * !!!
2447 * Historic vi did a special screen optimization for tab characters. As an
2448 * example, for the keystrokes "iabcd<esc>0C<tab>", the tab overwrote the
2449 * rest of the string when it was displayed.
2450 *
2451 * Because early versions of this implementation redisplayed the entire line
2452 * on each keystroke, the "bcd" was pushed to the right as it ignored that
2453 * the user had "promised" to change the rest of the characters. However,
2454 * the historic vi implementation had an even worse bug: given the keystrokes
2455 * "iabcd<esc>0R<tab><esc>", the "bcd" disappears, and magically reappears
2456 * on the second <esc> key.
2457 *
2458 * POSIX 1003.2 requires (will require) that this be fixed, specifying that
2459 * vi overwrite characters the user has committed to changing, on the basis
2460 * of the screen space they require, but that it not overwrite other characters.
2461 */
2462 static int
2467 u_int flags;
2468 {
2470 CHAR_T savech;
2474 /*
2475 * The 'R' command does one-for-one replacement, because there's
2476 * no way to know how many characters the user intends to replace.
2477 */
2483 }
2487 /*
2488 * If the old or new characters are tabs, then the length of the
2489 * display depends on the character position in the display. We
2490 * don't even try to handle this here, just ask the screen.
2491 */
2500 /*
2501 * Eat overwrite characters until we run out of them or we've
2502 * handled the length of the new character. If we only eat
2503 * part of an overwrite character, break it into its component
2504 * elements and display the remaining components.
2505 */
2512 else
2518 }
2535 else
2542 }
2543 }
2545 /*
2546 * If had to erase several characters, we adjust the total
2547 * count, and if there are any characters left, shift them
2548 * into position.
2549 */
2554 /* If we had enough overwrite characters, we're done. */
2558 }
2559 }
2561 /* Check to see if the character fits into the input buffer. */
2568 else
2571 }
2574 }
2576 /*
2577 * txt_isrch --
2578 * Do an incremental search.
2579 */
2580 static int
2586 {
2587 MARK start;
2588 db_recno_t lno;
2589 u_int sf;
2591 /* If it's a one-line screen, we don't do incrementals. */
2595 }
2597 /*
2598 * If the user erases back to the beginning of the buffer, there's
2599 * nothing to search for. Reset the cursor to the starting point.
2600 */
2604 }
2606 /*
2607 * If it's an RE quote character, and not quoted, ignore it until
2608 * we get another character.
2609 */
2614 /*
2615 * If it's a magic shell character, and not quoted, reset the cursor
2616 * to the starting point.
2617 */
2622 /*
2623 * If we see the search pattern termination character, then quit doing
2624 * an incremental search. There may be more, e.g., ":/foo/;/bar/",
2625 * and we can't handle that incrementally. Also, reset the cursor to
2626 * the original location, the ex search routines don't know anything
2627 * about incremental searches.
2628 */
2634 }
2636 /*
2637 * Remember the input line and discard the special input map,
2638 * but don't overwrite the input line on the screen.
2639 */
2646 /*
2647 * Specify a starting point and search. If we find a match, move to
2648 * it and refresh the screen. If we didn't find the match, then we
2649 * beep the screen. When searching from the original cursor position,
2650 * we have to move the cursor, otherwise, we don't want to move the
2651 * cursor in case the text at the current position continues to match.
2652 */
2659 }
2675 /* Reinstantiate the special input map. */
2681 /* Reset the line number of the input line. */
2684 /*
2685 * If the colon command-line moved, i.e. the screen scrolled,
2686 * refresh the input line.
2687 *
2688 * XXX
2689 * We shouldn't be calling vs_line, here -- we need dirty bits
2690 * on entries in the SMAP array.
2691 */
2696 }
2698 }
2700 /*
2701 * txt_resolve --
2702 * Resolve the input text chain into the file.
2703 */
2704 static int
2708 u_int32_t flags;
2709 {
2712 db_recno_t lno;
2715 /*
2716 * The first line replaces a current line, and all subsequent lines
2717 * are appended into the file. Resolve autoindented characters for
2718 * each line before committing it. If the latter causes the line to
2719 * change, we have to redisplay it, otherwise the information cached
2720 * about the line will be wrong.
2721 */
2727 else
2736 else
2741 }
2743 /*
2744 * Clear the input flag, the look-aside buffer is no longer valid.
2745 * Has to be done as part of text resolution, or upon return we'll
2746 * be looking at incorrect data.
2747 */
2751 }
2753 /*
2754 * txt_showmatch --
2755 * Show a character match.
2756 *
2757 * !!!
2758 * Historic vi tried to display matches even in the :colon command line.
2759 * I think not.
2760 */
2761 static int
2765 {
2767 VCS cs;
2768 MARK m;
2773 /*
2774 * Do a refresh first, in case we haven't done one in awhile,
2775 * so the user can see what we're complaining about.
2776 */
2781 /*
2782 * We don't display the match if it's not on the screen. Find
2783 * out what the first character on the screen is.
2784 */
2788 /* Initialize the getc() interface. */
2795 /* Search for the match. */
2804 }
2806 }
2811 }
2813 /* If the match is on the screen, move to it. */
2821 /* Wait for timeout or character arrival. */
2824 }
2826 /*
2827 * txt_margin --
2828 * Handle margin wrap.
2829 */
2830 static int
2835 u_int32_t flags;
2836 {
2841 /* Find the nearest previous blank. */
2846 }
2848 /*
2849 * If reach the start of the line, there's nowhere to break.
2850 *
2851 * !!!
2852 * Historic vi belled each time a character was entered after
2853 * crossing the margin until a space was entered which could
2854 * be used to break the line. I don't as it tends to wake the
2855 * cats.
2856 */
2860 }
2861 }
2863 /*
2864 * Store saved information about the rest of the line in the
2865 * wrapmargin TEXT structure.
2866 *
2867 * !!!
2868 * The offset field holds the length of the current characters
2869 * that the user entered, but which are getting split to the new
2870 * line -- it's going to be used to set the cursor value when we
2871 * move to the new line.
2872 */
2879 /* Correct current bookkeeping information. */
2887 }
2890 /*
2891 * !!!
2892 * Delete any trailing whitespace from the current line.
2893 */
2901 }
2904 }
2906 /*
2907 * txt_Rresolve --
2908 * Resolve the input line for the 'R' command.
2909 */
2910 static void
2916 {
2921 /*
2922 * Check to make sure that the cursor hasn't moved beyond
2923 * the end of the line.
2924 */
2928 /*
2929 * Calculate how many characters the user has entered,
2930 * plus the blanks erased by <carriage-return>/<newline>s.
2931 */
2936 }
2938 /*
2939 * If the user has entered less characters than the original line
2940 * was long, restore any overwriteable characters to the original
2941 * characters. These characters are entered as "insert characters",
2942 * because they're after the cursor and we don't want to lose them.
2943 * (This is okay because the R command has no insert characters.)
2944 * We set owrite to 0 so that the insert characters don't get copied
2945 * to somewhere else, which means that the line and the length have
2946 * to be adjusted here as well.
2947 *
2948 * We have to retrieve the original line because the original pinned
2949 * page has long since been discarded. If it doesn't exist, that's
2950 * okay, the user just extended the file.
2951 */
2961 }
2962 }
2964 /*
2965 * txt_nomorech --
2966 * No more characters message.
2967 */
2968 static void
2971 {
2973 }