| blob | 5580f4d4ede944fb1daf07947f11e2b30250b773 |
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 10.100 2000/08/30 18:03:44 skimo Exp $ (Berkeley) $Date: 2000/08/30 18:03:44 $";
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 }
319 }
321 /* Set default termination condition. */
324 /* Set the starting line, column. */
328 /*
329 * Set the insert and overwrite counts. If overwriting characters,
330 * do insertion afterward. If not overwriting characters, assume
331 * doing insertion. If change is to a mark, emphasize it with an
332 * CH_ENDMARK character.
333 */
343 }
345 /*
346 * Many of the special cases in text input are to handle autoindent
347 * support. Somebody decided that it would be a good idea if "^^D"
348 * and "0^D" deleted all of the autoindented characters. In an editor
349 * that takes single character input from the user, this beggars the
350 * imagination. Note also, "^^D" resets the next lines' autoindent,
351 * but "0^D" doesn't.
352 *
353 * We assume that autoindent only happens on empty lines, so insert
354 * and overwrite will be zero. If doing autoindent, figure out how
355 * much indentation we need and fill it in. Update input column and
356 * screen cursor as necessary.
357 */
363 /*
364 * The cc and S commands have a special feature -- leading
365 * <blank> characters are handled as autoindent characters.
366 * Beauty!
367 */
373 }
375 /* If getting a command buffer from the user, there may be a prompt. */
380 }
382 /*
383 * If appending after the end-of-line, add a space into the buffer
384 * and move the cursor right. This space is inserted, i.e. pushed
385 * along, and then deleted when the line is resolved. Assumes that
386 * the cursor is already positioned at the end of the line. This
387 * avoids the nastiness of having the cursor reside on a magical
388 * column, i.e. a column that doesn't really exist. The only down
389 * side is that we may wrap lines or scroll the screen before it's
390 * strictly necessary. Not a big deal.
391 */
397 }
399 /*
400 * Historic practice is that the wrapmargin value was a distance
401 * from the RIGHT-HAND margin, not the left. It's more useful to
402 * us as a distance from the left-hand margin, i.e. the same as
403 * the wraplen value. The wrapmargin option is historic practice.
404 * Nvi added the wraplen option so that it would be possible to
405 * edit files with consistent margins without knowing the number of
406 * columns in the window.
407 *
408 * XXX
409 * Setting margin causes a significant performance hit. Normally
410 * we don't update the screen if there are keys waiting, but we
411 * have to if margin is set, otherwise the screen routines don't
412 * know where the cursor is.
413 *
414 * !!!
415 * Abbreviated keys were affected by the wrapmargin option in the
416 * historic 4BSD vi. Mapped keys were usually, but sometimes not.
417 * See the comment in vi/v_text():set_txt_std for more information.
418 *
419 * !!!
420 * One more special case. If an inserted <blank> character causes
421 * wrapmargin to split the line, the next user entered character is
422 * discarded if it's a <space> character.
423 */
428 else
430 else
433 /* Initialize abbreviation checks. */
438 /*
439 * Set up the dot command. Dot commands are done by saving the actual
440 * characters and then reevaluating them so that things like wrapmargin
441 * can change between the insert and the replay.
442 *
443 * !!!
444 * Historically, vi did not remap or reabbreviate replayed input. (It
445 * did beep at you if you changed an abbreviation and then replayed the
446 * input. We're not that compatible.) We don't have to do anything to
447 * avoid remapping, as we're not getting characters from the terminal
448 * routines. Turn the abbreviation check off.
449 *
450 * XXX
451 * It would be nice if we could swallow backspaces and such, but it's
452 * not all that easy to do. What we can do is turn off the common
453 * error messages during the replay. Otherwise, when the user enters
454 * an illegal command, e.g., "Ia<erase><erase><erase><erase>b<escape>",
455 * and then does a '.', they get a list of error messages after command
456 * completion.
457 */
462 }
464 /* Other text input mode setup. */
471 /* Initialize input flags. */
474 /* Refresh the screen. */
479 /* If it's dot, just do it now. */
483 /* Get an event. */
488 /*
489 * If file completion overwrote part of the screen and nothing else has
490 * been displayed, clean up. We don't do this as part of the normal
491 * message resolution because we know the user is on the colon command
492 * line and there's no reason to enter explicit characters to continue.
493 */
504 }
506 /* Deal with all non-character events. */
515 /*
516 * !!!
517 * Historically, <interrupt> exited the user from text input
518 * mode or cancelled a colon command, and returned to command
519 * mode. It also beeped the terminal, but that seems a bit
520 * excessive.
521 */
528 /* <resize> interrupts the input mode. */
534 }
536 /*
537 * !!!
538 * If the first character of the input is a nul, replay the previous
539 * input. (Historically, it's okay to replay non-existent input.)
540 * This was not documented as far as I know, and is a great test of vi
541 * clones.
542 */
551 }
553 /*
554 * File name completion and colon command-line editing. We don't
555 * have enough meta characters, so we expect people to overload
556 * them. If the two characters are the same, then we do file name
557 * completion if the cursor is past the first column, and do colon
558 * command-line editing if it's not.
559 */
573 }
578 }
579 }
581 /* Abbreviation overflow check. See comment in txt_abbrev(). */
582 #define MAX_ABBREVIATION_EXPANSION 256
592 }
596 /* Check to see if the character fits into the replay buffers. */
601 }
607 }
609 /* Wrapmargin check for leading space. */
614 }
616 /* If quoted by someone else, simply insert the character. */
620 /*
621 * !!!
622 * If this character was quoted by a K_VLNEXT or a backslash, replace
623 * the placeholder (a carat or a backslash) with the new character.
624 * If it was quoted by a K_VLNEXT, we've already adjusted the cursor
625 * because it has to appear on top of the placeholder character. If
626 * it was quoted by a backslash, adjust the cursor now, the cursor
627 * doesn't appear on top of it. Historic practice in both cases.
628 *
629 * Skip tests for abbreviations; ":ab xa XA" followed by "ixa^V<space>"
630 * doesn't perform an abbreviation. Special case, ^V^J (not ^V^M) is
631 * the same as ^J, historically.
632 */
644 }
648 }
650 }
652 /*
653 * !!!
654 * Translate "<CH_HEX>[isxdigit()]*" to a character with a hex value:
655 * this test delimits the value by any non-hex character. Offset by
656 * one, we use 0 to mean that we've found <CH_HEX>.
657 */
662 }
667 /* Return in script windows and the command line. */
670 /*
671 * If this was a map, we may have not displayed
672 * the line. Display it, just in case.
673 *
674 * If a script window and not the colon line,
675 * push a <cr> so it gets executed.
676 */
683 /* Set term condition: if empty. */
686 /*
687 * Set term condition: if searching incrementally and
688 * the user entered a pattern, return a completed
689 * search, regardless if the entire pattern was found.
690 */
696 }
698 #define LINE_RESOLVE { \
699 /* \
700 * Handle abbreviations. If there was one, discard the \
701 * replay characters. \
703 if (abb == AB_INWORD && \
704 !LF_ISSET(TXT_REPLAY) && F_ISSET(gp, G_ABBREV)) { \
705 if (txt_abbrev(sp, tp, &evp->e_c, \
706 LF_ISSET(TXT_INFOLINE), &tmp, \
707 &ab_turnoff)) \
708 goto err; \
709 if (tmp) { \
710 if (LF_ISSET(TXT_RECORD)) \
711 rcol -= tmp + 1; \
712 goto resolve; \
713 } \
714 } \
715 if (abb != AB_NOTSET) \
716 abb = AB_NOTWORD; \
717 if (UNMAP_TST) \
718 txt_unmap(sp, tp, &ec_flags); \
719 /* \
720 * Delete any appended cursor. It's possible to get in \
721 * situations where TXT_APPENDEOL is set but tp->insert \
722 * is 0 when using the R command and all the characters \
723 * are tp->owrite characters. \
725 if (LF_ISSET(TXT_APPENDEOL) && tp->insert > 0) { \
726 --tp->len; \
727 --tp->insert; \
728 } \
729 }
730 LINE_RESOLVE;
732 /*
733 * Save the current line information for restoration in
734 * txt_backup(), and set the line final length.
735 */
740 /* Update the old line. */
744 /*
745 * Historic practice, when the autoindent edit option was set,
746 * was to delete <blank> characters following the inserted
747 * newline. This affected the 'R', 'c', and 's' commands; 'c'
748 * and 's' retained the insert characters only, 'R' moved the
749 * overwrite and insert characters into the next TEXT structure.
750 * We keep track of the number of characters erased for the 'R'
751 * command so that the final resolution of the line is correct.
752 */
768 }
770 /*
771 * !!!
772 * Create a new line and insert the new TEXT into the queue.
773 * DON'T insert until the old line has been updated, or the
774 * inserted line count in line.c:db_get() will be wrong.
775 */
781 /* Set up bookkeeping for the new line. */
786 /*
787 * Reset the autoindent line value. 0^D keeps the autoindent
788 * line from changing, ^D changes the level, even if there were
789 * no characters in the old line. Note, if using the current
790 * tp structure, use the cursor as the length, the autoindent
791 * characters may have been erased.
792 */
802 }
804 /* Reset the cursor. */
807 /*
808 * If we're here because wrapmargin was set and we've broken a
809 * line, there may be additional information (i.e. the start of
810 * a line) in the wmt structure.
811 */
815 #define WMTSPACE wmt.offset + wmt.owrite + wmt.insert
823 }
825 }
827 /* New lines are TXT_APPENDEOL. */
834 }
836 /* Swap old and new TEXT's, and update the new line. */
846 /* If we have a count, start replaying the input. */
855 /*
856 * Some commands (e.g. 'o') need a <newline> for each
857 * repetition.
858 */
862 /*
863 * The R command turns into the 'a' command after the
864 * first repetition.
865 */
870 }
872 }
874 /* Set term condition: if empty. */
877 /*
878 * Set term condition: if searching incrementally and the user
879 * entered a pattern, return a completed search, regardless if
880 * the entire pattern was found.
881 */
887 /*
888 * Clean up for the 'R' command, restoring overwrite
889 * characters, and making them into insert characters.
890 */
894 /*
895 * If there are any overwrite characters, copy down
896 * any insert characters, and decrement the length.
897 */
903 }
905 /*
906 * Optionally resolve the lines into the file. If not
907 * resolving the lines into the file, end the line with
908 * a nul. If the line is empty, then set the length to
909 * 0, the termination condition has already been set.
910 *
911 * XXX
912 * This is wrong, should pass back a length.
913 */
920 }
922 /*
923 * Set the return cursor position to rest on the last
924 * inserted character.
925 */
929 /* Update the last line. */
942 /*
943 * If in the first column or no characters to erase, ignore
944 * the ^D (this matches historic practice). If not doing
945 * autoindent or already inserted non-ai characters, it's a
946 * literal. The latter test is done in the switch, as the
947 * CARAT forms are N + 1, not N.
948 */
959 /* Save the ai string for later. */
986 }
989 /* If can erase over the prompt, return. */
993 }
995 /*
996 * If at the beginning of the line, try and drop back to a
997 * previously inserted line.
998 */
1005 }
1007 /* If nothing to erase, bell the user. */
1012 }
1014 /* Drop back one character. */
1017 /*
1018 * Historically, vi didn't replace the erased characters with
1019 * <blank>s, presumably because it's easier to fix a minor
1020 * typing mistake and continue on if the previous letters are
1021 * already there. This is a problem for incremental searching,
1022 * because the user can no longer tell where they are in the
1023 * colon command line because the cursor is at the last search
1024 * point in the screen. So, if incrementally searching, erase
1025 * the erased characters from the screen.
1026 */
1030 /*
1031 * Increment overwrite, decrement ai if deleted.
1032 *
1033 * !!!
1034 * Historic vi did not permit users to use erase characters
1035 * to delete autoindent characters. We do. Eat hot death,
1036 * POSIX.
1037 */
1042 /* Reset if we deleted an incremental search character. */
1047 /*
1048 * If at the beginning of the line, try and drop back to a
1049 * previously inserted line.
1050 */
1056 }
1058 /*
1059 * If at offset, nothing to erase so bell the user.
1060 */
1065 }
1067 /*
1068 * The first werase goes back to any autoindent column and the
1069 * second werase goes back to the offset.
1070 *
1071 * !!!
1072 * Historic vi did not permit users to use erase characters to
1073 * delete autoindent characters.
1074 */
1080 }
1082 /* Skip over trailing space characters. */
1086 }
1089 /*
1090 * There are three types of word erase found on UNIX systems.
1091 * They can be identified by how the string /a/b/c is treated
1092 * -- as 1, 3, or 6 words. Historic vi had two classes of
1093 * characters, and strings were delimited by them and
1094 * <blank>'s, so, 6 words. The historic tty interface used
1095 * <blank>'s to delimit strings, so, 1 word. The algorithm
1096 * offered in the 4.4BSD tty interface (as stty altwerase)
1097 * treats it as 3 words -- there are two classes of
1098 * characters, and strings are delimited by them and
1099 * <blank>'s. The difference is that the type of the first
1100 * erased character erased is ignored, which is exactly right
1101 * when erasing pathname components. The edit options
1102 * TXT_ALTWERASE and TXT_TTYWERASE specify the 4.4BSD tty
1103 * interface and the historic tty driver behavior,
1104 * respectively, and the default is the same as the historic
1105 * vi behavior.
1106 *
1107 * Overwrite erased characters if doing incremental search;
1108 * see comment above.
1109 */
1118 }
1127 }
1138 }
1139 }
1141 /* Reset if we deleted an incremental search character. */
1146 /*
1147 * !!!
1148 * If at the beginning of the line, try and drop back to a
1149 * previously inserted line. Historic vi did not permit
1150 * users to go back to previous lines.
1151 */
1157 }
1159 /* If at offset, nothing to erase so bell the user. */
1164 }
1166 /*
1167 * First kill goes back to any autoindent and second kill goes
1168 * back to the offset.
1169 *
1170 * !!!
1171 * Historic vi did not permit users to use erase characters to
1172 * delete autoindent characters.
1173 */
1179 }
1182 /*
1183 * Overwrite erased characters if doing incremental search;
1184 * see comment above.
1185 */
1190 else
1193 /* Reset if we deleted an incremental search character. */
1209 /*
1210 * !!!
1211 * Historic vi tried to make abbreviations after a backslash
1212 * escape work. If you did ":ab x y", and inserted "x\^H",
1213 * (assuming the erase character was ^H) you got "x^H", and
1214 * no abbreviation was done. If you inserted "x\z", however,
1215 * it tried to back up and do the abbreviation, i.e. replace
1216 * 'x' with 'y'. The problem was it got it wrong, and you
1217 * ended up with "zy\".
1218 *
1219 * This is really hard to do (you have to remember the
1220 * word/non-word state, for example), and doesn't make any
1221 * sense to me. Both backslash and the characters it
1222 * (usually) escapes will individually trigger the
1223 * abbreviation, so I don't see why the combination of them
1224 * wouldn't. I don't expect to get caught on this one,
1225 * particularly since it never worked right, but I've been
1226 * wrong before.
1227 *
1228 * Do the tests for abbreviations, so ":ab xa XA",
1229 * "ixa\<K_VERASE>" performs the abbreviation.
1230 */
1236 /*
1237 * Turn on the quote flag so that the underlying routines
1238 * quote the next character where it's possible. Turn off
1239 * the input mapbiting flag so that we don't remap the next
1240 * character.
1241 */
1245 /*
1246 * !!!
1247 * Skip the tests for abbreviations, so ":ab xa XA",
1248 * "ixa^V<space>" doesn't perform the abbreviation.
1249 */
1256 * Historically, vi eliminated nul's out of hand. If the
1257 * beautify option was set, it also deleted any unknown
1258 * ASCII value less than space (040) and the del character
1259 * (0177), except for tabs. Unknown is a key word here.
1260 * Most vi documentation claims that it deleted everything
1261 * but <tab>, <nl> and <ff>, as that's what the original
1262 * 4BSD documentation said. This is obviously wrong,
1263 * however, as <esc> would be included in that list. What
1264 * we do is eliminate any unquoted, iscntrl() character that
1265 * wasn't a replay and wasn't handled specially, except
1266 * <tab> or <ff>.
1267 */
1275 }
1278 * If entering a non-word character after a word, check for
1279 * abbreviations. If there was one, discard replay characters.
1280 * If entering a blank character, check for unmap commands,
1281 * as well.
1282 */
1293 }
1294 }
1297 }
1304 /*
1305 * If we're using K_VLNEXT to quote the next character, then
1306 * we want the cursor to position itself on the ^ placeholder
1307 * we're displaying, to match historic practice.
1308 */
1312 }
1314 /*
1315 * !!!
1316 * Translate "<CH_HEX>[isxdigit()]*" to a character with
1317 * a hex value: this test delimits the value by the max
1318 * number of hex bytes. Offset by one, we use 0 to mean
1319 * that we've found <CH_HEX>.
1320 */
1325 }
1327 /*
1328 * Check to see if we've crossed the margin.
1329 *
1330 * !!!
1331 * In the historic vi, the wrapmargin value was figured out
1332 * using the display widths of the characters, i.e. <tab>
1333 * characters were counted as two characters if the list edit
1334 * option is set, but as the tabstop edit option number of
1335 * characters otherwise. That's what the vs_column() function
1336 * gives us, so we use it.
1337 */
1349 }
1350 }
1351 }
1353 /*
1354 * If we've reached the end of the buffer, then we need to
1355 * switch into insert mode. This happens when there's a
1356 * change to a mark and the user puts in more characters than
1357 * the length of the motion.
1358 */
1366 }
1368 /* Step the quote state forward. */
1374 }
1376 }
1378 #ifdef DEBUG
1386 }
1387 #endif
1390 * 1: If we don't need to know where the cursor really is and we're
1391 * replaying text, keep going.
1392 */
1396 /*
1397 * 2: Reset the line. Don't bother unless we're about to wait on
1398 * a character or we need to know where the cursor really is.
1399 * We have to do this before showing matching characters so the
1400 * user can see what they're matching.
1401 */
1406 /*
1407 * 3: If there aren't keys waiting, display the matching character.
1408 * We have to do this before resolving any messages, otherwise
1409 * the error message from a missing match won't appear correctly.
1410 */
1415 }
1417 /*
1418 * 4: If there have been messages and we're not editing on the colon
1419 * command line or doing file name completion, resolve them.
1420 */
1426 /*
1427 * 5: Refresh the screen if we're about to wait on a character or we
1428 * need to know where the cursor really is.
1429 */
1434 }
1436 /* 6: Proceed with the incremental search. */
1440 /* 7: Next character... */
1448 /* If recording for playback, save it. */
1452 /*
1453 * If not working on the colon command line, set the final cursor
1454 * position.
1455 */
1459 }
1462 err:
1463 alloc_err:
1466 }
1468 /*
1469 * txt_abbrev --
1470 * Handle abbreviations.
1471 */
1472 static int
1478 {
1484 /* Check to make sure we're not at the start of an append. */
1491 /*
1492 * Find the start of the "word".
1493 *
1494 * !!!
1495 * We match historic practice, which, as far as I can tell, had an
1496 * off-by-one error. The way this worked was that when the inserted
1497 * text switched from a "word" character to a non-word character,
1498 * vi would check for possible abbreviations. It would then take the
1499 * type (i.e. word/non-word) of the character entered TWO characters
1500 * ago, and move backward in the text until reaching a character that
1501 * was not that type, or the beginning of the insert, the line, or
1502 * the file. For example, in the string "abc<space>", when the <space>
1503 * character triggered the abbreviation check, the type of the 'b'
1504 * character was used for moving through the string. Maybe there's a
1505 * reason for not using the first (i.e. 'c') character, but I can't
1506 * think of one.
1507 *
1508 * Terminate at the beginning of the insert or the character after the
1509 * offset character -- both can be tested for using tp->offset.
1510 */
1521 }
1522 else
1528 }
1530 /*
1531 * !!!
1532 * Historic vi exploded abbreviations on the command line. This has
1533 * obvious problems in that unabbreviating the string can be extremely
1534 * tricky, particularly if the string has, say, an embedded escape
1535 * character. Personally, I think it's a stunningly bad idea. Other
1536 * examples of problems this caused in historic vi are:
1537 * :ab foo bar
1538 * :ab foo baz
1539 * results in "bar" being abbreviated to "baz", which wasn't what the
1540 * user had in mind at all. Also, the commands:
1541 * :ab foo bar
1542 * :unab foo<space>
1543 * resulted in an error message that "bar" wasn't mapped. Finally,
1544 * since the string was already exploded by the time the unabbreviate
1545 * command got it, all it knew was that an abbreviation had occurred.
1546 * Cleverly, it checked the replacement string for its unabbreviation
1547 * match, which meant that the commands:
1548 * :ab foo1 bar
1549 * :ab foo2 bar
1550 * :unab foo2
1551 * unabbreviate "foo1", and the commands:
1552 * :ab foo bar
1553 * :ab bar baz
1554 * unabbreviate "foo"!
1555 *
1556 * Anyway, people neglected to first ask my opinion before they wrote
1557 * macros that depend on this stuff, so, we make this work as follows.
1558 * When checking for an abbreviation on the command line, if we get a
1559 * string which is <blank> terminated and which starts at the beginning
1560 * of the line, we check to see it is the abbreviate or unabbreviate
1561 * commands. If it is, turn abbreviations off and return as if no
1562 * abbreviation was found. Note also, minor trickiness, so that if
1563 * the user erases the line and starts another command, we turn the
1564 * abbreviations back on.
1565 *
1566 * This makes the layering look like a Nachos Supreme.
1567 */
1575 else
1579 /* Check for any abbreviations. */
1583 /*
1584 * Push the abbreviation onto the tty stack. Historically, characters
1585 * resulting from an abbreviation expansion were themselves subject to
1586 * map expansions, O_SHOWMATCH matching etc. This means the expanded
1587 * characters will be re-tested for abbreviations. It's difficult to
1588 * know what historic practice in this case was, since abbreviations
1589 * were applied to :colon command lines, so entering abbreviations that
1590 * looped was tricky, although possible. In addition, obvious loops
1591 * didn't work as expected. (The command ':ab a b|ab b c|ab c a' will
1592 * silently only implement and/or display the last abbreviation.)
1593 *
1594 * This implementation doesn't recover well from such abbreviations.
1595 * The main input loop counts abbreviated characters, and, when it
1596 * reaches a limit, discards any abbreviated characters on the queue.
1597 * It's difficult to back up to the original position, as the replay
1598 * queue would have to be adjusted, and the line state when an initial
1599 * abbreviated character was received would have to be saved.
1600 */
1607 /*
1608 * If the size of the abbreviation is larger than or equal to the size
1609 * of the original text, move to the start of the replaced characters,
1610 * and add their length to the overwrite count.
1611 *
1612 * If the abbreviation is smaller than the original text, we have to
1613 * delete the additional overwrite characters and copy down any insert
1614 * characters.
1615 */
1625 }
1627 /*
1628 * We return the length of the abbreviated characters. This is so
1629 * the calling routine can replace the replay characters with the
1630 * abbreviation. This means that subsequent '.' commands will produce
1631 * the same text, regardless of intervening :[un]abbreviate commands.
1632 * This is historic practice.
1633 */
1636 }
1638 /*
1639 * txt_unmap --
1640 * Handle the unmap command.
1641 */
1642 static void
1647 {
1651 /* Find the beginning of this "word". */
1656 }
1660 }
1662 /*
1663 * !!!
1664 * Historic vi exploded input mappings on the command line. See the
1665 * txt_abbrev() routine for an explanation of the problems inherent
1666 * in this.
1667 *
1668 * We make this work as follows. If we get a string which is <blank>
1669 * terminated and which starts at the beginning of the line, we check
1670 * to see it is the unmap command. If it is, we return that the input
1671 * mapping should be turned off. Note also, minor trickiness, so that
1672 * if the user erases the line and starts another command, we go ahead
1673 * an turn mapping back on.
1674 */
1677 else
1679 }
1681 /*
1682 * txt_ai_resolve --
1683 * When a line is resolved by <esc>, review autoindent characters.
1684 */
1685 static void
1690 {
1691 u_long ts;
1698 /*
1699 * If the line is empty, has an offset, or no autoindent
1700 * characters, we're done.
1701 */
1705 /*
1706 * If the length is less than or equal to the autoindent
1707 * characters, delete them.
1708 */
1712 }
1714 /*
1715 * The autoindent characters plus any leading <blank> characters
1716 * in the line are resolved into the minimum number of characters.
1717 * Historic practice.
1718 */
1721 /* Figure out the last <blank> screen column. */
1731 }
1733 /*
1734 * If there are no spaces, or no tabs after spaces and less than
1735 * ts spaces, it's already minimal.
1736 */
1740 /* Count up spaces/tabs needed to get to the target. */
1745 /*
1746 * Figure out how many characters we're dropping -- if we're not
1747 * dropping any, it's already minimal, we're done.
1748 */
1754 /* Shift the rest of the characters down, adjust the counts. */
1760 /* Fill in space/tab characters. */
1766 }
1768 /*
1769 * v_txt_auto --
1770 * Handle autoindent. If aitp isn't NULL, use it, otherwise,
1771 * retrieve the line.
1772 *
1773 * PUBLIC: int v_txt_auto __P((SCR *, db_recno_t, TEXT *, size_t, TEXT *));
1774 */
1775 int
1778 db_recno_t lno;
1781 {
1786 /*
1787 * If the ex append command is executed with an address of 0,
1788 * it's possible to get here with a line number of 0. Return
1789 * an indent of 0.
1790 */
1794 }
1800 /* Count whitespace characters. */
1805 /* Set count, check for no indentation. */
1809 /* Make sure the buffer's big enough. */
1812 /* Copy the buffer's current contents up. */
1817 /* Copy the indentation into the new buffer. */
1820 /* Set the autoindent count. */
1823 }
1825 /*
1826 * txt_backup --
1827 * Back up to the previously edited line.
1828 */
1835 {
1839 /* Get a handle on the previous TEXT structure. */
1845 }
1847 /* Bookkeeping. */
1850 /* Handle appending to the line. */
1860 /* Release the current TEXT. */
1864 /* Update the old line on the screen. */
1868 /* Return the new/current TEXT. */
1870 }
1872 /*
1873 * Text indentation is truly strange. ^T and ^D do movements to the next or
1874 * previous shiftwidth value, i.e. for a 1-based numbering, with shiftwidth=3,
1875 * ^T moves a cursor on the 7th, 8th or 9th column to the 10th column, and ^D
1876 * moves it back.
1877 *
1878 * !!!
1879 * The ^T and ^D characters in historical vi had special meaning only when they
1880 * were the first characters entered after entering text input mode. As normal
1881 * erase characters couldn't erase autoindent characters (^T in this case), it
1882 * meant that inserting text into previously existing text was strange -- ^T
1883 * only worked if it was the first keystroke(s), and then could only be erased
1884 * using ^D. This implementation treats ^T specially anywhere it occurs in the
1885 * input, and permits the standard erase characters to erase the characters it
1886 * inserts.
1887 *
1888 * !!!
1889 * A fun test is to try:
1890 * :se sw=4 ai list
1891 * i<CR>^Tx<CR>^Tx<CR>^Tx<CR>^Dx<CR>^Dx<CR>^Dx<esc>
1892 * Historic vi loses some of the '$' marks on the line ends, but otherwise gets
1893 * it right.
1894 *
1895 * XXX
1896 * Technically, txt_dent should be part of the screen interface, as it requires
1897 * knowledge of character sizes, including <space>s, on the screen. It's here
1898 * because it's a complicated little beast, and I didn't want to shove it down
1899 * into the screen. It's probable that KEY_LEN will call into the screen once
1900 * there are screens with different character representations.
1901 *
1902 * txt_dent --
1903 * Handle ^T indents, ^D outdents.
1904 *
1905 * If anything changes here, check the ex version to see if it needs similar
1906 * changes.
1907 */
1908 static int
1913 {
1914 CHAR_T ch;
1922 /*
1923 * Since we don't know what precedes the character(s) being inserted
1924 * (or deleted), the preceding whitespace characters must be resolved.
1925 * An example is a <tab>, which doesn't need a full shiftwidth number
1926 * of columns because it's preceded by <space>s. This is easy to get
1927 * if the user sets shiftwidth to a value less than tabstop (or worse,
1928 * something for which tabstop isn't a multiple) and then uses ^T to
1929 * indent, and ^D to outdent.
1930 *
1931 * Figure out the current and target screen columns. In the historic
1932 * vi, the autoindent column was NOT determined using display widths
1933 * of characters as was the wrapmargin column. For that reason, we
1934 * can't use the vs_column() function, but have to calculate it here.
1935 * This is slow, but it's normally only on the first few characters of
1936 * a line.
1937 */
1945 else
1948 /*
1949 * The AI characters will be turned into overwrite characters if the
1950 * cursor immediately follows them. We test both the cursor position
1951 * and the indent flag because there's no single test. (^T can only
1952 * be detected by the cursor position, and while we know that the test
1953 * is always true for ^D, the cursor can be in more than one place, as
1954 * "0^D" and "^D" are different.)
1955 */
1958 /*
1959 * Back up over any previous <blank> characters, changing them into
1960 * overwrite characters (including any ai characters). Then figure
1961 * out the current screen column.
1962 */
1970 /*
1971 * If we didn't move up to or past the target, it's because there
1972 * weren't enough characters to delete, e.g. the first character
1973 * of the line was a tp->offset character, and the user entered
1974 * ^D to move to the beginning of a line. An example of this is:
1975 *
1976 * :set ai sw=4<cr>i<space>a<esc>i^T^D
1977 *
1978 * Otherwise, count up the total spaces/tabs needed to get from the
1979 * beginning of the line (or the last non-<blank> character) to the
1980 * target.
1981 */
1989 }
1991 /* If we overwrote ai characters, reset the ai count. */
1995 /*
1996 * Call txt_insch() to insert each character, so that we get the
1997 * correct effect when we add a <tab> to replace N <spaces>.
1998 */
2004 }
2006 /*
2007 * txt_fc --
2008 * File name completion.
2009 */
2010 static int
2015 {
2018 CHAR_T s_ch;
2019 EXCMD cmd;
2029 /*
2030 * Find the beginning of this "word" -- if we're at the beginning
2031 * of the line, it's a special case.
2032 */
2042 }
2046 }
2048 /*
2049 * Get enough space for a wildcard character.
2050 *
2051 * XXX
2052 * This won't work for "foo\", since the \ will escape the expansion
2053 * character. I'm not sure if that's a bug or not...
2054 */
2062 /* Build an ex command, and call the ex expansion routines. */
2067 }
2079 /* If something changed, do the exchange. */
2084 /* If haven't done a directory test, do it now. */
2091 }
2093 /* If nothing changed, period, ring the bell. */
2102 /* Find the length of the shortest match. */
2110 }
2112 }
2114 /* Overwrite the expanded text first. */
2118 /* If lost text, make the remaining old text overwrite characters. */
2122 }
2124 /* Overwrite any overwrite characters next. */
2128 /* Shift remaining text up, and move the cursor to the end. */
2141 }
2143 /* If a single match and it's a directory, retry it. */
2161 }
2163 }
2165 /*
2166 * txt_fc_col --
2167 * Display file names for file name completion.
2168 */
2169 static int
2174 {
2185 /* Trim any directory prefix common to all of the files. */
2194 prefix)) {
2197 }
2198 }
2200 /*
2201 * Figure out the column width for the longest name. Output is done on
2202 * 6 character "tab" boundaries for no particular reason. (Since we
2203 * don't output tab characters, we ignore the terminal's tab settings.)
2204 * Ignore the user's tab setting because we have no idea how reasonable
2205 * it is.
2206 */
2212 }
2215 /*
2216 * Writing to the bottom line of the screen is always turned off when
2217 * SC_TINPUT_INFO is set. Turn it back on, we know what we're doing.
2218 */
2225 #define CHK_INTR \
2226 if (F_ISSET(gp, G_INTERRUPTED)) \
2227 goto intr;
2229 /* If the largest file name is too large, just print them. */
2238 }
2241 CHK_INTR;
2243 /* Figure out the number of columns. */
2252 /* Display the files in sorted order. */
2261 CHK_INTR;
2266 CHK_INTR;
2267 }
2269 CHK_INTR;
2270 }
2272 CHK_INTR;
2273 }
2278 }
2283 }
2285 /*
2286 * txt_emark --
2287 * Set the end mark on the line.
2288 */
2289 static int
2294 {
2295 CHAR_T ch;
2302 /*
2303 * The end mark may not be the same size as the current character.
2304 * Don't let the line shift.
2305 */
2309 else
2312 /*
2313 * If the line got longer, well, it's weird, but it's easy. If
2314 * it's the same length, it's easy. If it got shorter, we have
2315 * to fix it up.
2316 */
2330 else
2334 }
2337 }
2339 /*
2340 * txt_err --
2341 * Handle an error during input processing.
2342 */
2343 static void
2347 {
2348 db_recno_t lno;
2350 /*
2351 * The problem with input processing is that the cursor is at an
2352 * indeterminate position since some input may have been lost due
2353 * to a malloc error. So, try to go back to the place from which
2354 * the cursor started, knowing that it may no longer be available.
2355 *
2356 * We depend on at least one line number being set in the text
2357 * chain.
2358 */
2365 /* Redraw the screen, just in case. */
2367 }
2369 /*
2370 * txt_hex --
2371 * Let the user insert any character value they want.
2372 *
2373 * !!!
2374 * This is an extension. The pattern "^X[0-9a-fA-F]*" is a way
2375 * for the user to specify a character value which their keyboard
2376 * may not be able to enter.
2377 */
2378 static int
2382 {
2383 CHAR_T savec;
2385 u_long value;
2390 /*
2391 * Null-terminate the string. Since nul isn't a legal hex value,
2392 * this should be okay, and lets us use a local routine, which
2393 * presumably understands the character set, to convert the value.
2394 */
2398 /* Find the previous CH_HEX character. */
2403 }
2404 /* Not on this line? Shouldn't happen. */
2407 }
2409 /* If length of 0, then it wasn't a hex value. */
2413 /* Get the value. */
2420 }
2422 /* Restore the original character. */
2425 /* Adjust the bookkeeping. */
2430 /* Copy down any overwrite characters. */
2435 /* Copy down any insert characters. */
2442 }
2444 /*
2445 * txt_insch --
2446 *
2447 * !!!
2448 * Historic vi did a special screen optimization for tab characters. As an
2449 * example, for the keystrokes "iabcd<esc>0C<tab>", the tab overwrote the
2450 * rest of the string when it was displayed.
2451 *
2452 * Because early versions of this implementation redisplayed the entire line
2453 * on each keystroke, the "bcd" was pushed to the right as it ignored that
2454 * the user had "promised" to change the rest of the characters. However,
2455 * the historic vi implementation had an even worse bug: given the keystrokes
2456 * "iabcd<esc>0R<tab><esc>", the "bcd" disappears, and magically reappears
2457 * on the second <esc> key.
2458 *
2459 * POSIX 1003.2 requires (will require) that this be fixed, specifying that
2460 * vi overwrite characters the user has committed to changing, on the basis
2461 * of the screen space they require, but that it not overwrite other characters.
2462 */
2463 static int
2468 u_int flags;
2469 {
2471 CHAR_T savech;
2475 /*
2476 * The 'R' command does one-for-one replacement, because there's
2477 * no way to know how many characters the user intends to replace.
2478 */
2484 }
2488 /*
2489 * If the old or new characters are tabs, then the length of the
2490 * display depends on the character position in the display. We
2491 * don't even try to handle this here, just ask the screen.
2492 */
2501 /*
2502 * Eat overwrite characters until we run out of them or we've
2503 * handled the length of the new character. If we only eat
2504 * part of an overwrite character, break it into its component
2505 * elements and display the remaining components.
2506 */
2513 else
2519 }
2536 else
2543 }
2544 }
2546 /*
2547 * If had to erase several characters, we adjust the total
2548 * count, and if there are any characters left, shift them
2549 * into position.
2550 */
2555 /* If we had enough overwrite characters, we're done. */
2559 }
2560 }
2562 /* Check to see if the character fits into the input buffer. */
2569 else
2572 }
2575 }
2577 /*
2578 * txt_isrch --
2579 * Do an incremental search.
2580 */
2581 static int
2587 {
2588 MARK start;
2589 db_recno_t lno;
2590 u_int sf;
2592 /* If it's a one-line screen, we don't do incrementals. */
2596 }
2598 /*
2599 * If the user erases back to the beginning of the buffer, there's
2600 * nothing to search for. Reset the cursor to the starting point.
2601 */
2605 }
2607 /*
2608 * If it's an RE quote character, and not quoted, ignore it until
2609 * we get another character.
2610 */
2615 /*
2616 * If it's a magic shell character, and not quoted, reset the cursor
2617 * to the starting point.
2618 */
2623 /*
2624 * If we see the search pattern termination character, then quit doing
2625 * an incremental search. There may be more, e.g., ":/foo/;/bar/",
2626 * and we can't handle that incrementally. Also, reset the cursor to
2627 * the original location, the ex search routines don't know anything
2628 * about incremental searches.
2629 */
2635 }
2637 /*
2638 * Remember the input line and discard the special input map,
2639 * but don't overwrite the input line on the screen.
2640 */
2647 /*
2648 * Specify a starting point and search. If we find a match, move to
2649 * it and refresh the screen. If we didn't find the match, then we
2650 * beep the screen. When searching from the original cursor position,
2651 * we have to move the cursor, otherwise, we don't want to move the
2652 * cursor in case the text at the current position continues to match.
2653 */
2660 }
2676 /* Reinstantiate the special input map. */
2682 /* Reset the line number of the input line. */
2685 /*
2686 * If the colon command-line moved, i.e. the screen scrolled,
2687 * refresh the input line.
2688 *
2689 * XXX
2690 * We shouldn't be calling vs_line, here -- we need dirty bits
2691 * on entries in the SMAP array.
2692 */
2697 }
2699 }
2701 /*
2702 * txt_resolve --
2703 * Resolve the input text chain into the file.
2704 */
2705 static int
2709 u_int32_t flags;
2710 {
2713 db_recno_t lno;
2716 /*
2717 * The first line replaces a current line, and all subsequent lines
2718 * are appended into the file. Resolve autoindented characters for
2719 * each line before committing it. If the latter causes the line to
2720 * change, we have to redisplay it, otherwise the information cached
2721 * about the line will be wrong.
2722 */
2728 else
2737 else
2742 }
2744 /*
2745 * Clear the input flag, the look-aside buffer is no longer valid.
2746 * Has to be done as part of text resolution, or upon return we'll
2747 * be looking at incorrect data.
2748 */
2752 }
2754 /*
2755 * txt_showmatch --
2756 * Show a character match.
2757 *
2758 * !!!
2759 * Historic vi tried to display matches even in the :colon command line.
2760 * I think not.
2761 */
2762 static int
2766 {
2768 VCS cs;
2769 MARK m;
2774 /*
2775 * Do a refresh first, in case we haven't done one in awhile,
2776 * so the user can see what we're complaining about.
2777 */
2782 /*
2783 * We don't display the match if it's not on the screen. Find
2784 * out what the first character on the screen is.
2785 */
2789 /* Initialize the getc() interface. */
2796 /* Search for the match. */
2805 }
2807 }
2812 }
2814 /* If the match is on the screen, move to it. */
2822 /* Wait for timeout or character arrival. */
2825 }
2827 /*
2828 * txt_margin --
2829 * Handle margin wrap.
2830 */
2831 static int
2836 u_int32_t flags;
2837 {
2842 /* Find the nearest previous blank. */
2847 }
2849 /*
2850 * If reach the start of the line, there's nowhere to break.
2851 *
2852 * !!!
2853 * Historic vi belled each time a character was entered after
2854 * crossing the margin until a space was entered which could
2855 * be used to break the line. I don't as it tends to wake the
2856 * cats.
2857 */
2861 }
2862 }
2864 /*
2865 * Store saved information about the rest of the line in the
2866 * wrapmargin TEXT structure.
2867 *
2868 * !!!
2869 * The offset field holds the length of the current characters
2870 * that the user entered, but which are getting split to the new
2871 * line -- it's going to be used to set the cursor value when we
2872 * move to the new line.
2873 */
2880 /* Correct current bookkeeping information. */
2888 }
2891 /*
2892 * !!!
2893 * Delete any trailing whitespace from the current line.
2894 */
2902 }
2905 }
2907 /*
2908 * txt_Rresolve --
2909 * Resolve the input line for the 'R' command.
2910 */
2911 static void
2917 {
2922 /*
2923 * Check to make sure that the cursor hasn't moved beyond
2924 * the end of the line.
2925 */
2929 /*
2930 * Calculate how many characters the user has entered,
2931 * plus the blanks erased by <carriage-return>/<newline>s.
2932 */
2937 }
2939 /*
2940 * If the user has entered less characters than the original line
2941 * was long, restore any overwriteable characters to the original
2942 * characters. These characters are entered as "insert characters",
2943 * because they're after the cursor and we don't want to lose them.
2944 * (This is okay because the R command has no insert characters.)
2945 * We set owrite to 0 so that the insert characters don't get copied
2946 * to somewhere else, which means that the line and the length have
2947 * to be adjusted here as well.
2948 *
2949 * We have to retrieve the original line because the original pinned
2950 * page has long since been discarded. If it doesn't exist, that's
2951 * okay, the user just extended the file.
2952 */
2962 }
2963 }
2965 /*
2966 * txt_nomorech --
2967 * No more characters message.
2968 */
2969 static void
2972 {
2974 }