| blob | 862a2c825e9c1e449a9427bb20c82dfc54674d39 |
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 10.94 2000/06/30 19:46:03 skimo Exp $ (Berkeley) $Date: 2000/06/30 19:46:03 $";
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 *, 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. */
668 /*
669 * If this was a map, we may have not displayed
670 * the line. Display it, just in case.
671 *
672 * If a script window and not the colon line,
673 * push a <cr> so it gets executed.
674 */
681 /* Set term condition: if empty. */
684 /*
685 * Set term condition: if searching incrementally and
686 * the user entered a pattern, return a completed
687 * search, regardless if the entire pattern was found.
688 */
694 }
696 #define LINE_RESOLVE { \
697 /* \
698 * Handle abbreviations. If there was one, discard the \
699 * replay characters. \
701 if (abb == AB_INWORD && \
702 !LF_ISSET(TXT_REPLAY) && F_ISSET(gp, G_ABBREV)) { \
703 if (txt_abbrev(sp, tp, &evp->e_c, \
704 LF_ISSET(TXT_INFOLINE), &tmp, \
705 &ab_turnoff)) \
706 goto err; \
707 if (tmp) { \
708 if (LF_ISSET(TXT_RECORD)) \
709 rcol -= tmp + 1; \
710 goto resolve; \
711 } \
712 } \
713 if (abb != AB_NOTSET) \
714 abb = AB_NOTWORD; \
715 if (UNMAP_TST) \
716 txt_unmap(sp, tp, &ec_flags); \
717 /* \
718 * Delete any appended cursor. It's possible to get in \
719 * situations where TXT_APPENDEOL is set but tp->insert \
720 * is 0 when using the R command and all the characters \
721 * are tp->owrite characters. \
723 if (LF_ISSET(TXT_APPENDEOL) && tp->insert > 0) { \
724 --tp->len; \
725 --tp->insert; \
726 } \
727 }
728 LINE_RESOLVE;
730 /*
731 * Save the current line information for restoration in
732 * txt_backup(), and set the line final length.
733 */
738 /* Update the old line. */
742 /*
743 * Historic practice, when the autoindent edit option was set,
744 * was to delete <blank> characters following the inserted
745 * newline. This affected the 'R', 'c', and 's' commands; 'c'
746 * and 's' retained the insert characters only, 'R' moved the
747 * overwrite and insert characters into the next TEXT structure.
748 * We keep track of the number of characters erased for the 'R'
749 * command so that the final resolution of the line is correct.
750 */
766 }
768 /*
769 * !!!
770 * Create a new line and insert the new TEXT into the queue.
771 * DON'T insert until the old line has been updated, or the
772 * inserted line count in line.c:db_get() will be wrong.
773 */
779 /* Set up bookkeeping for the new line. */
784 /*
785 * Reset the autoindent line value. 0^D keeps the autoindent
786 * line from changing, ^D changes the level, even if there were
787 * no characters in the old line. Note, if using the current
788 * tp structure, use the cursor as the length, the autoindent
789 * characters may have been erased.
790 */
800 }
802 /* Reset the cursor. */
805 /*
806 * If we're here because wrapmargin was set and we've broken a
807 * line, there may be additional information (i.e. the start of
808 * a line) in the wmt structure.
809 */
813 #define WMTSPACE wmt.offset + wmt.owrite + wmt.insert
821 }
823 }
825 /* New lines are TXT_APPENDEOL. */
832 }
834 /* Swap old and new TEXT's, and update the new line. */
844 /* 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. */
984 }
987 /* If can erase over the prompt, return. */
991 }
993 /*
994 * If at the beginning of the line, try and drop back to a
995 * previously inserted line.
996 */
1003 }
1005 /* If nothing to erase, bell the user. */
1010 }
1012 /* Drop back one character. */
1015 /*
1016 * Historically, vi didn't replace the erased characters with
1017 * <blank>s, presumably because it's easier to fix a minor
1018 * typing mistake and continue on if the previous letters are
1019 * already there. This is a problem for incremental searching,
1020 * because the user can no longer tell where they are in the
1021 * colon command line because the cursor is at the last search
1022 * point in the screen. So, if incrementally searching, erase
1023 * the erased characters from the screen.
1024 */
1028 /*
1029 * Increment overwrite, decrement ai if deleted.
1030 *
1031 * !!!
1032 * Historic vi did not permit users to use erase characters
1033 * to delete autoindent characters. We do. Eat hot death,
1034 * POSIX.
1035 */
1040 /* Reset if we deleted an incremental search character. */
1045 /*
1046 * If at the beginning of the line, try and drop back to a
1047 * previously inserted line.
1048 */
1054 }
1056 /*
1057 * If at offset, nothing to erase so bell the user.
1058 */
1063 }
1065 /*
1066 * The first werase goes back to any autoindent column and the
1067 * second werase goes back to the offset.
1068 *
1069 * !!!
1070 * Historic vi did not permit users to use erase characters to
1071 * delete autoindent characters.
1072 */
1078 }
1080 /* Skip over trailing space characters. */
1084 }
1087 /*
1088 * There are three types of word erase found on UNIX systems.
1089 * They can be identified by how the string /a/b/c is treated
1090 * -- as 1, 3, or 6 words. Historic vi had two classes of
1091 * characters, and strings were delimited by them and
1092 * <blank>'s, so, 6 words. The historic tty interface used
1093 * <blank>'s to delimit strings, so, 1 word. The algorithm
1094 * offered in the 4.4BSD tty interface (as stty altwerase)
1095 * treats it as 3 words -- there are two classes of
1096 * characters, and strings are delimited by them and
1097 * <blank>'s. The difference is that the type of the first
1098 * erased character erased is ignored, which is exactly right
1099 * when erasing pathname components. The edit options
1100 * TXT_ALTWERASE and TXT_TTYWERASE specify the 4.4BSD tty
1101 * interface and the historic tty driver behavior,
1102 * respectively, and the default is the same as the historic
1103 * vi behavior.
1104 *
1105 * Overwrite erased characters if doing incremental search;
1106 * see comment above.
1107 */
1116 }
1125 }
1136 }
1137 }
1139 /* Reset if we deleted an incremental search character. */
1144 /*
1145 * !!!
1146 * If at the beginning of the line, try and drop back to a
1147 * previously inserted line. Historic vi did not permit
1148 * users to go back to previous lines.
1149 */
1155 }
1157 /* If at offset, nothing to erase so bell the user. */
1162 }
1164 /*
1165 * First kill goes back to any autoindent and second kill goes
1166 * back to the offset.
1167 *
1168 * !!!
1169 * Historic vi did not permit users to use erase characters to
1170 * delete autoindent characters.
1171 */
1177 }
1180 /*
1181 * Overwrite erased characters if doing incremental search;
1182 * see comment above.
1183 */
1188 else
1191 /* Reset if we deleted an incremental search character. */
1207 /*
1208 * !!!
1209 * Historic vi tried to make abbreviations after a backslash
1210 * escape work. If you did ":ab x y", and inserted "x\^H",
1211 * (assuming the erase character was ^H) you got "x^H", and
1212 * no abbreviation was done. If you inserted "x\z", however,
1213 * it tried to back up and do the abbreviation, i.e. replace
1214 * 'x' with 'y'. The problem was it got it wrong, and you
1215 * ended up with "zy\".
1216 *
1217 * This is really hard to do (you have to remember the
1218 * word/non-word state, for example), and doesn't make any
1219 * sense to me. Both backslash and the characters it
1220 * (usually) escapes will individually trigger the
1221 * abbreviation, so I don't see why the combination of them
1222 * wouldn't. I don't expect to get caught on this one,
1223 * particularly since it never worked right, but I've been
1224 * wrong before.
1225 *
1226 * Do the tests for abbreviations, so ":ab xa XA",
1227 * "ixa\<K_VERASE>" performs the abbreviation.
1228 */
1234 /*
1235 * Turn on the quote flag so that the underlying routines
1236 * quote the next character where it's possible. Turn off
1237 * the input mapbiting flag so that we don't remap the next
1238 * character.
1239 */
1243 /*
1244 * !!!
1245 * Skip the tests for abbreviations, so ":ab xa XA",
1246 * "ixa^V<space>" doesn't perform the abbreviation.
1247 */
1254 * Historically, vi eliminated nul's out of hand. If the
1255 * beautify option was set, it also deleted any unknown
1256 * ASCII value less than space (040) and the del character
1257 * (0177), except for tabs. Unknown is a key word here.
1258 * Most vi documentation claims that it deleted everything
1259 * but <tab>, <nl> and <ff>, as that's what the original
1260 * 4BSD documentation said. This is obviously wrong,
1261 * however, as <esc> would be included in that list. What
1262 * we do is eliminate any unquoted, iscntrl() character that
1263 * wasn't a replay and wasn't handled specially, except
1264 * <tab> or <ff>.
1265 */
1273 }
1276 * If entering a non-word character after a word, check for
1277 * abbreviations. If there was one, discard replay characters.
1278 * If entering a blank character, check for unmap commands,
1279 * as well.
1280 */
1291 }
1292 }
1295 }
1302 /*
1303 * If we're using K_VLNEXT to quote the next character, then
1304 * we want the cursor to position itself on the ^ placeholder
1305 * we're displaying, to match historic practice.
1306 */
1310 }
1312 /*
1313 * !!!
1314 * Translate "<CH_HEX>[isxdigit()]*" to a character with
1315 * a hex value: this test delimits the value by the max
1316 * number of hex bytes. Offset by one, we use 0 to mean
1317 * that we've found <CH_HEX>.
1318 */
1323 }
1325 /*
1326 * Check to see if we've crossed the margin.
1327 *
1328 * !!!
1329 * In the historic vi, the wrapmargin value was figured out
1330 * using the display widths of the characters, i.e. <tab>
1331 * characters were counted as two characters if the list edit
1332 * option is set, but as the tabstop edit option number of
1333 * characters otherwise. That's what the vs_column() function
1334 * gives us, so we use it.
1335 */
1347 }
1348 }
1349 }
1351 /*
1352 * If we've reached the end of the buffer, then we need to
1353 * switch into insert mode. This happens when there's a
1354 * change to a mark and the user puts in more characters than
1355 * the length of the motion.
1356 */
1364 }
1366 /* Step the quote state forward. */
1372 }
1374 }
1376 #ifdef DEBUG
1384 }
1385 #endif
1388 * 1: If we don't need to know where the cursor really is and we're
1389 * replaying text, keep going.
1390 */
1394 /*
1395 * 2: Reset the line. Don't bother unless we're about to wait on
1396 * a character or we need to know where the cursor really is.
1397 * We have to do this before showing matching characters so the
1398 * user can see what they're matching.
1399 */
1404 /*
1405 * 3: If there aren't keys waiting, display the matching character.
1406 * We have to do this before resolving any messages, otherwise
1407 * the error message from a missing match won't appear correctly.
1408 */
1413 }
1415 /*
1416 * 4: If there have been messages and we're not editing on the colon
1417 * command line or doing file name completion, resolve them.
1418 */
1424 /*
1425 * 5: Refresh the screen if we're about to wait on a character or we
1426 * need to know where the cursor really is.
1427 */
1432 }
1434 /* 6: Proceed with the incremental search. */
1438 /* 7: Next character... */
1446 /* If recording for playback, save it. */
1450 /*
1451 * If not working on the colon command line, set the final cursor
1452 * position.
1453 */
1457 }
1460 err:
1461 alloc_err:
1464 }
1466 /*
1467 * txt_abbrev --
1468 * Handle abbreviations.
1469 */
1470 static int
1476 {
1482 /* Check to make sure we're not at the start of an append. */
1489 /*
1490 * Find the start of the "word".
1491 *
1492 * !!!
1493 * We match historic practice, which, as far as I can tell, had an
1494 * off-by-one error. The way this worked was that when the inserted
1495 * text switched from a "word" character to a non-word character,
1496 * vi would check for possible abbreviations. It would then take the
1497 * type (i.e. word/non-word) of the character entered TWO characters
1498 * ago, and move backward in the text until reaching a character that
1499 * was not that type, or the beginning of the insert, the line, or
1500 * the file. For example, in the string "abc<space>", when the <space>
1501 * character triggered the abbreviation check, the type of the 'b'
1502 * character was used for moving through the string. Maybe there's a
1503 * reason for not using the first (i.e. 'c') character, but I can't
1504 * think of one.
1505 *
1506 * Terminate at the beginning of the insert or the character after the
1507 * offset character -- both can be tested for using tp->offset.
1508 */
1519 }
1520 else
1526 }
1528 /*
1529 * !!!
1530 * Historic vi exploded abbreviations on the command line. This has
1531 * obvious problems in that unabbreviating the string can be extremely
1532 * tricky, particularly if the string has, say, an embedded escape
1533 * character. Personally, I think it's a stunningly bad idea. Other
1534 * examples of problems this caused in historic vi are:
1535 * :ab foo bar
1536 * :ab foo baz
1537 * results in "bar" being abbreviated to "baz", which wasn't what the
1538 * user had in mind at all. Also, the commands:
1539 * :ab foo bar
1540 * :unab foo<space>
1541 * resulted in an error message that "bar" wasn't mapped. Finally,
1542 * since the string was already exploded by the time the unabbreviate
1543 * command got it, all it knew was that an abbreviation had occurred.
1544 * Cleverly, it checked the replacement string for its unabbreviation
1545 * match, which meant that the commands:
1546 * :ab foo1 bar
1547 * :ab foo2 bar
1548 * :unab foo2
1549 * unabbreviate "foo1", and the commands:
1550 * :ab foo bar
1551 * :ab bar baz
1552 * unabbreviate "foo"!
1553 *
1554 * Anyway, people neglected to first ask my opinion before they wrote
1555 * macros that depend on this stuff, so, we make this work as follows.
1556 * When checking for an abbreviation on the command line, if we get a
1557 * string which is <blank> terminated and which starts at the beginning
1558 * of the line, we check to see it is the abbreviate or unabbreviate
1559 * commands. If it is, turn abbreviations off and return as if no
1560 * abbreviation was found. Note also, minor trickiness, so that if
1561 * the user erases the line and starts another command, we turn the
1562 * abbreviations back on.
1563 *
1564 * This makes the layering look like a Nachos Supreme.
1565 */
1573 else
1577 /* Check for any abbreviations. */
1581 /*
1582 * Push the abbreviation onto the tty stack. Historically, characters
1583 * resulting from an abbreviation expansion were themselves subject to
1584 * map expansions, O_SHOWMATCH matching etc. This means the expanded
1585 * characters will be re-tested for abbreviations. It's difficult to
1586 * know what historic practice in this case was, since abbreviations
1587 * were applied to :colon command lines, so entering abbreviations that
1588 * looped was tricky, although possible. In addition, obvious loops
1589 * didn't work as expected. (The command ':ab a b|ab b c|ab c a' will
1590 * silently only implement and/or display the last abbreviation.)
1591 *
1592 * This implementation doesn't recover well from such abbreviations.
1593 * The main input loop counts abbreviated characters, and, when it
1594 * reaches a limit, discards any abbreviated characters on the queue.
1595 * It's difficult to back up to the original position, as the replay
1596 * queue would have to be adjusted, and the line state when an initial
1597 * abbreviated character was received would have to be saved.
1598 */
1605 /*
1606 * If the size of the abbreviation is larger than or equal to the size
1607 * of the original text, move to the start of the replaced characters,
1608 * and add their length to the overwrite count.
1609 *
1610 * If the abbreviation is smaller than the original text, we have to
1611 * delete the additional overwrite characters and copy down any insert
1612 * characters.
1613 */
1623 }
1625 /*
1626 * We return the length of the abbreviated characters. This is so
1627 * the calling routine can replace the replay characters with the
1628 * abbreviation. This means that subsequent '.' commands will produce
1629 * the same text, regardless of intervening :[un]abbreviate commands.
1630 * This is historic practice.
1631 */
1634 }
1636 /*
1637 * txt_unmap --
1638 * Handle the unmap command.
1639 */
1640 static void
1645 {
1649 /* Find the beginning of this "word". */
1654 }
1658 }
1660 /*
1661 * !!!
1662 * Historic vi exploded input mappings on the command line. See the
1663 * txt_abbrev() routine for an explanation of the problems inherent
1664 * in this.
1665 *
1666 * We make this work as follows. If we get a string which is <blank>
1667 * terminated and which starts at the beginning of the line, we check
1668 * to see it is the unmap command. If it is, we return that the input
1669 * mapping should be turned off. Note also, minor trickiness, so that
1670 * if the user erases the line and starts another command, we go ahead
1671 * an turn mapping back on.
1672 */
1675 else
1677 }
1679 /*
1680 * txt_ai_resolve --
1681 * When a line is resolved by <esc>, review autoindent characters.
1682 */
1683 static void
1688 {
1689 u_long ts;
1696 /*
1697 * If the line is empty, has an offset, or no autoindent
1698 * characters, we're done.
1699 */
1703 /*
1704 * If the length is less than or equal to the autoindent
1705 * characters, delete them.
1706 */
1710 }
1712 /*
1713 * The autoindent characters plus any leading <blank> characters
1714 * in the line are resolved into the minimum number of characters.
1715 * Historic practice.
1716 */
1719 /* Figure out the last <blank> screen column. */
1729 }
1731 /*
1732 * If there are no spaces, or no tabs after spaces and less than
1733 * ts spaces, it's already minimal.
1734 */
1738 /* Count up spaces/tabs needed to get to the target. */
1743 /*
1744 * Figure out how many characters we're dropping -- if we're not
1745 * dropping any, it's already minimal, we're done.
1746 */
1752 /* Shift the rest of the characters down, adjust the counts. */
1758 /* Fill in space/tab characters. */
1764 }
1766 /*
1767 * v_txt_auto --
1768 * Handle autoindent. If aitp isn't NULL, use it, otherwise,
1769 * retrieve the line.
1770 *
1771 * PUBLIC: int v_txt_auto __P((SCR *, db_recno_t, TEXT *, size_t, TEXT *));
1772 */
1773 int
1776 db_recno_t lno;
1779 {
1784 /*
1785 * If the ex append command is executed with an address of 0,
1786 * it's possible to get here with a line number of 0. Return
1787 * an indent of 0.
1788 */
1792 }
1798 /* Count whitespace characters. */
1803 /* Set count, check for no indentation. */
1807 /* Make sure the buffer's big enough. */
1810 /* Copy the buffer's current contents up. */
1815 /* Copy the indentation into the new buffer. */
1818 /* Set the autoindent count. */
1821 }
1823 /*
1824 * txt_backup --
1825 * Back up to the previously edited line.
1826 */
1833 {
1837 /* Get a handle on the previous TEXT structure. */
1843 }
1845 /* Bookkeeping. */
1848 /* Handle appending to the line. */
1858 /* Release the current TEXT. */
1862 /* Update the old line on the screen. */
1866 /* Return the new/current TEXT. */
1868 }
1870 /*
1871 * Text indentation is truly strange. ^T and ^D do movements to the next or
1872 * previous shiftwidth value, i.e. for a 1-based numbering, with shiftwidth=3,
1873 * ^T moves a cursor on the 7th, 8th or 9th column to the 10th column, and ^D
1874 * moves it back.
1875 *
1876 * !!!
1877 * The ^T and ^D characters in historical vi had special meaning only when they
1878 * were the first characters entered after entering text input mode. As normal
1879 * erase characters couldn't erase autoindent characters (^T in this case), it
1880 * meant that inserting text into previously existing text was strange -- ^T
1881 * only worked if it was the first keystroke(s), and then could only be erased
1882 * using ^D. This implementation treats ^T specially anywhere it occurs in the
1883 * input, and permits the standard erase characters to erase the characters it
1884 * inserts.
1885 *
1886 * !!!
1887 * A fun test is to try:
1888 * :se sw=4 ai list
1889 * i<CR>^Tx<CR>^Tx<CR>^Tx<CR>^Dx<CR>^Dx<CR>^Dx<esc>
1890 * Historic vi loses some of the '$' marks on the line ends, but otherwise gets
1891 * it right.
1892 *
1893 * XXX
1894 * Technically, txt_dent should be part of the screen interface, as it requires
1895 * knowledge of character sizes, including <space>s, on the screen. It's here
1896 * because it's a complicated little beast, and I didn't want to shove it down
1897 * into the screen. It's probable that KEY_LEN will call into the screen once
1898 * there are screens with different character representations.
1899 *
1900 * txt_dent --
1901 * Handle ^T indents, ^D outdents.
1902 *
1903 * If anything changes here, check the ex version to see if it needs similar
1904 * changes.
1905 */
1906 static int
1911 {
1912 CHAR_T ch;
1920 /*
1921 * Since we don't know what precedes the character(s) being inserted
1922 * (or deleted), the preceding whitespace characters must be resolved.
1923 * An example is a <tab>, which doesn't need a full shiftwidth number
1924 * of columns because it's preceded by <space>s. This is easy to get
1925 * if the user sets shiftwidth to a value less than tabstop (or worse,
1926 * something for which tabstop isn't a multiple) and then uses ^T to
1927 * indent, and ^D to outdent.
1928 *
1929 * Figure out the current and target screen columns. In the historic
1930 * vi, the autoindent column was NOT determined using display widths
1931 * of characters as was the wrapmargin column. For that reason, we
1932 * can't use the vs_column() function, but have to calculate it here.
1933 * This is slow, but it's normally only on the first few characters of
1934 * a line.
1935 */
1943 else
1946 /*
1947 * The AI characters will be turned into overwrite characters if the
1948 * cursor immediately follows them. We test both the cursor position
1949 * and the indent flag because there's no single test. (^T can only
1950 * be detected by the cursor position, and while we know that the test
1951 * is always true for ^D, the cursor can be in more than one place, as
1952 * "0^D" and "^D" are different.)
1953 */
1956 /*
1957 * Back up over any previous <blank> characters, changing them into
1958 * overwrite characters (including any ai characters). Then figure
1959 * out the current screen column.
1960 */
1968 /*
1969 * If we didn't move up to or past the target, it's because there
1970 * weren't enough characters to delete, e.g. the first character
1971 * of the line was a tp->offset character, and the user entered
1972 * ^D to move to the beginning of a line. An example of this is:
1973 *
1974 * :set ai sw=4<cr>i<space>a<esc>i^T^D
1975 *
1976 * Otherwise, count up the total spaces/tabs needed to get from the
1977 * beginning of the line (or the last non-<blank> character) to the
1978 * target.
1979 */
1987 }
1989 /* If we overwrote ai characters, reset the ai count. */
1993 /*
1994 * Call txt_insch() to insert each character, so that we get the
1995 * correct effect when we add a <tab> to replace N <spaces>.
1996 */
2002 }
2004 /*
2005 * txt_fc --
2006 * File name completion.
2007 */
2008 static int
2013 {
2016 CHAR_T s_ch;
2017 EXCMD cmd;
2025 /*
2026 * Find the beginning of this "word" -- if we're at the beginning
2027 * of the line, it's a special case.
2028 */
2038 }
2042 }
2044 /*
2045 * Get enough space for a wildcard character.
2046 *
2047 * XXX
2048 * This won't work for "foo\", since the \ will escape the expansion
2049 * character. I'm not sure if that's a bug or not...
2050 */
2058 /* Build an ex command, and call the ex expansion routines. */
2063 }
2075 /* If something changed, do the exchange. */
2080 /* If haven't done a directory test, do it now. */
2085 }
2087 /* If nothing changed, period, ring the bell. */
2096 /* Find the length of the shortest match. */
2104 }
2106 }
2108 /* Overwrite the expanded text first. */
2112 /* If lost text, make the remaining old text overwrite characters. */
2116 }
2118 /* Overwrite any overwrite characters next. */
2122 /* Shift remaining text up, and move the cursor to the end. */
2135 }
2137 /* If a single match and it's a directory, retry it. */
2154 }
2156 }
2158 /*
2159 * txt_fc_col --
2160 * Display file names for file name completion.
2161 */
2162 static int
2167 {
2176 /* Trim any directory prefix common to all of the files. */
2186 }
2187 }
2189 /*
2190 * Figure out the column width for the longest name. Output is done on
2191 * 6 character "tab" boundaries for no particular reason. (Since we
2192 * don't output tab characters, we ignore the terminal's tab settings.)
2193 * Ignore the user's tab setting because we have no idea how reasonable
2194 * it is.
2195 */
2201 }
2204 /*
2205 * Writing to the bottom line of the screen is always turned off when
2206 * SC_TINPUT_INFO is set. Turn it back on, we know what we're doing.
2207 */
2214 #define CHK_INTR \
2215 if (F_ISSET(gp, G_INTERRUPTED)) \
2216 goto intr;
2218 /* If the largest file name is too large, just print them. */
2225 }
2228 CHK_INTR;
2230 /* Figure out the number of columns. */
2239 /* Display the files in sorted order. */
2246 CHK_INTR;
2251 CHK_INTR;
2252 }
2254 CHK_INTR;
2255 }
2257 CHK_INTR;
2258 }
2263 }
2268 }
2270 /*
2271 * txt_emark --
2272 * Set the end mark on the line.
2273 */
2274 static int
2279 {
2286 /*
2287 * The end mark may not be the same size as the current character.
2288 * Don't let the line shift.
2289 */
2293 else
2296 /*
2297 * If the line got longer, well, it's weird, but it's easy. If
2298 * it's the same length, it's easy. If it got shorter, we have
2299 * to fix it up.
2300 */
2314 else
2318 }
2321 }
2323 /*
2324 * txt_err --
2325 * Handle an error during input processing.
2326 */
2327 static void
2331 {
2332 db_recno_t lno;
2334 /*
2335 * The problem with input processing is that the cursor is at an
2336 * indeterminate position since some input may have been lost due
2337 * to a malloc error. So, try to go back to the place from which
2338 * the cursor started, knowing that it may no longer be available.
2339 *
2340 * We depend on at least one line number being set in the text
2341 * chain.
2342 */
2349 /* Redraw the screen, just in case. */
2351 }
2353 /*
2354 * txt_hex --
2355 * Let the user insert any character value they want.
2356 *
2357 * !!!
2358 * This is an extension. The pattern "^X[0-9a-fA-F]*" is a way
2359 * for the user to specify a character value which their keyboard
2360 * may not be able to enter.
2361 */
2362 static int
2366 {
2367 CHAR_T savec;
2369 u_long value;
2372 /*
2373 * Null-terminate the string. Since nul isn't a legal hex value,
2374 * this should be okay, and lets us use a local routine, which
2375 * presumably understands the character set, to convert the value.
2376 */
2380 /* Find the previous CH_HEX character. */
2385 }
2386 /* Not on this line? Shouldn't happen. */
2389 }
2391 /* If length of 0, then it wasn't a hex value. */
2395 /* Get the value. */
2401 }
2403 /* Restore the original character. */
2406 /* Adjust the bookkeeping. */
2411 /* Copy down any overwrite characters. */
2415 /* Copy down any insert characters. */
2421 }
2423 /*
2424 * txt_insch --
2425 *
2426 * !!!
2427 * Historic vi did a special screen optimization for tab characters. As an
2428 * example, for the keystrokes "iabcd<esc>0C<tab>", the tab overwrote the
2429 * rest of the string when it was displayed.
2430 *
2431 * Because early versions of this implementation redisplayed the entire line
2432 * on each keystroke, the "bcd" was pushed to the right as it ignored that
2433 * the user had "promised" to change the rest of the characters. However,
2434 * the historic vi implementation had an even worse bug: given the keystrokes
2435 * "iabcd<esc>0R<tab><esc>", the "bcd" disappears, and magically reappears
2436 * on the second <esc> key.
2437 *
2438 * POSIX 1003.2 requires (will require) that this be fixed, specifying that
2439 * vi overwrite characters the user has committed to changing, on the basis
2440 * of the screen space they require, but that it not overwrite other characters.
2441 */
2442 static int
2447 u_int flags;
2448 {
2453 /*
2454 * The 'R' command does one-for-one replacement, because there's
2455 * no way to know how many characters the user intends to replace.
2456 */
2462 }
2466 /*
2467 * If the old or new characters are tabs, then the length of the
2468 * display depends on the character position in the display. We
2469 * don't even try to handle this here, just ask the screen.
2470 */
2479 /*
2480 * Eat overwrite characters until we run out of them or we've
2481 * handled the length of the new character. If we only eat
2482 * part of an overwrite character, break it into its component
2483 * elements and display the remaining components.
2484 */
2491 else
2497 }
2513 else
2520 }
2521 }
2523 /*
2524 * If had to erase several characters, we adjust the total
2525 * count, and if there are any characters left, shift them
2526 * into position.
2527 */
2532 /* If we had enough overwrite characters, we're done. */
2536 }
2537 }
2539 /* Check to see if the character fits into the input buffer. */
2546 else
2549 }
2552 }
2554 /*
2555 * txt_isrch --
2556 * Do an incremental search.
2557 */
2558 static int
2564 {
2565 MARK start;
2566 db_recno_t lno;
2567 u_int sf;
2569 /* If it's a one-line screen, we don't do incrementals. */
2573 }
2575 /*
2576 * If the user erases back to the beginning of the buffer, there's
2577 * nothing to search for. Reset the cursor to the starting point.
2578 */
2582 }
2584 /*
2585 * If it's an RE quote character, and not quoted, ignore it until
2586 * we get another character.
2587 */
2592 /*
2593 * If it's a magic shell character, and not quoted, reset the cursor
2594 * to the starting point.
2595 */
2600 /*
2601 * If we see the search pattern termination character, then quit doing
2602 * an incremental search. There may be more, e.g., ":/foo/;/bar/",
2603 * and we can't handle that incrementally. Also, reset the cursor to
2604 * the original location, the ex search routines don't know anything
2605 * about incremental searches.
2606 */
2612 }
2614 /*
2615 * Remember the input line and discard the special input map,
2616 * but don't overwrite the input line on the screen.
2617 */
2624 /*
2625 * Specify a starting point and search. If we find a match, move to
2626 * it and refresh the screen. If we didn't find the match, then we
2627 * beep the screen. When searching from the original cursor position,
2628 * we have to move the cursor, otherwise, we don't want to move the
2629 * cursor in case the text at the current position continues to match.
2630 */
2637 }
2653 /* Reinstantiate the special input map. */
2659 /* Reset the line number of the input line. */
2662 /*
2663 * If the colon command-line moved, i.e. the screen scrolled,
2664 * refresh the input line.
2665 *
2666 * XXX
2667 * We shouldn't be calling vs_line, here -- we need dirty bits
2668 * on entries in the SMAP array.
2669 */
2674 }
2676 }
2678 /*
2679 * txt_resolve --
2680 * Resolve the input text chain into the file.
2681 */
2682 static int
2686 u_int32_t flags;
2687 {
2690 db_recno_t lno;
2693 /*
2694 * The first line replaces a current line, and all subsequent lines
2695 * are appended into the file. Resolve autoindented characters for
2696 * each line before committing it. If the latter causes the line to
2697 * change, we have to redisplay it, otherwise the information cached
2698 * about the line will be wrong.
2699 */
2705 else
2714 else
2719 }
2721 /*
2722 * Clear the input flag, the look-aside buffer is no longer valid.
2723 * Has to be done as part of text resolution, or upon return we'll
2724 * be looking at incorrect data.
2725 */
2729 }
2731 /*
2732 * txt_showmatch --
2733 * Show a character match.
2734 *
2735 * !!!
2736 * Historic vi tried to display matches even in the :colon command line.
2737 * I think not.
2738 */
2739 static int
2743 {
2745 VCS cs;
2746 MARK m;
2751 /*
2752 * Do a refresh first, in case we haven't done one in awhile,
2753 * so the user can see what we're complaining about.
2754 */
2759 /*
2760 * We don't display the match if it's not on the screen. Find
2761 * out what the first character on the screen is.
2762 */
2766 /* Initialize the getc() interface. */
2773 /* Search for the match. */
2782 }
2784 }
2789 }
2791 /* If the match is on the screen, move to it. */
2799 /* Wait for timeout or character arrival. */
2802 }
2804 /*
2805 * txt_margin --
2806 * Handle margin wrap.
2807 */
2808 static int
2813 u_int32_t flags;
2814 {
2819 /* Find the nearest previous blank. */
2824 }
2826 /*
2827 * If reach the start of the line, there's nowhere to break.
2828 *
2829 * !!!
2830 * Historic vi belled each time a character was entered after
2831 * crossing the margin until a space was entered which could
2832 * be used to break the line. I don't as it tends to wake the
2833 * cats.
2834 */
2838 }
2839 }
2841 /*
2842 * Store saved information about the rest of the line in the
2843 * wrapmargin TEXT structure.
2844 *
2845 * !!!
2846 * The offset field holds the length of the current characters
2847 * that the user entered, but which are getting split to the new
2848 * line -- it's going to be used to set the cursor value when we
2849 * move to the new line.
2850 */
2857 /* Correct current bookkeeping information. */
2865 }
2868 /*
2869 * !!!
2870 * Delete any trailing whitespace from the current line.
2871 */
2879 }
2882 }
2884 /*
2885 * txt_Rresolve --
2886 * Resolve the input line for the 'R' command.
2887 */
2888 static void
2894 {
2899 /*
2900 * Check to make sure that the cursor hasn't moved beyond
2901 * the end of the line.
2902 */
2906 /*
2907 * Calculate how many characters the user has entered,
2908 * plus the blanks erased by <carriage-return>/<newline>s.
2909 */
2914 }
2916 /*
2917 * If the user has entered less characters than the original line
2918 * was long, restore any overwriteable characters to the original
2919 * characters. These characters are entered as "insert characters",
2920 * because they're after the cursor and we don't want to lose them.
2921 * (This is okay because the R command has no insert characters.)
2922 * We set owrite to 0 so that the insert characters don't get copied
2923 * to somewhere else, which means that the line and the length have
2924 * to be adjusted here as well.
2925 *
2926 * We have to retrieve the original line because the original pinned
2927 * page has long since been discarded. If it doesn't exist, that's
2928 * okay, the user just extended the file.
2929 */
2939 }
2940 }
2942 /*
2943 * txt_nomorech --
2944 * No more characters message.
2945 */
2946 static void
2949 {
2951 }