freebsd-dev/contrib/dialog/inputstr.c
Devin Teske febdb46880 Update dialog to 1.2-20130923.
Approved by:	re (marius)
2013-09-24 14:52:43 +00:00

803 lines
18 KiB
C

/*
* $Id: inputstr.c,v 1.83 2013/09/23 23:19:26 tom Exp $
*
* inputstr.c -- functions for input/display of a string
*
* Copyright 2000-2012,2013 Thomas E. Dickey
*
* This program is free software; you can redistribute it and/or modify
* it under the terms of the GNU Lesser General Public License, version 2.1
* as published by the Free Software Foundation.
*
* This program is distributed in the hope that it will be useful, but
* WITHOUT ANY WARRANTY; without even the implied warranty of
* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
* Lesser General Public License for more details.
*
* You should have received a copy of the GNU Lesser General Public
* License along with this program; if not, write to
* Free Software Foundation, Inc.
* 51 Franklin St., Fifth Floor
* Boston, MA 02110, USA.
*/
#include <dialog.h>
#include <dlg_keys.h>
#include <errno.h>
#ifdef HAVE_SETLOCALE
#include <locale.h>
#endif
#if defined(HAVE_SEARCH_H) && defined(HAVE_TSEARCH)
#include <search.h>
#else
#undef HAVE_TSEARCH
#endif
#ifdef NEED_WCHAR_H
#include <wchar.h>
#endif
#if defined(USE_WIDE_CURSES)
#define USE_CACHING 1
#elif defined(HAVE_XDIALOG)
#define USE_CACHING 1 /* editbox really needs caching! */
#else
#define USE_CACHING 0
#endif
typedef struct _cache {
struct _cache *next;
#if USE_CACHING
int cache_num; /* tells what type of data is in list[] */
const char *string_at; /* unique: associate caches by char* */
#endif
size_t s_len; /* strlen(string) - we add 1 for EOS */
size_t i_len; /* length(list) - we add 1 for EOS */
char *string; /* a copy of the last-processed string */
int *list; /* indices into the string */
} CACHE;
#if USE_CACHING
#define SAME_CACHE(c,s,l) (c->string != 0 && memcmp(c->string,s,l) == 0)
static CACHE *cache_list;
typedef enum {
cInxCols
,cCntWideBytes
,cCntWideChars
,cInxWideChars
,cMAX
} CACHE_USED;
#ifdef HAVE_TSEARCH
static void *sorted_cache;
#endif
#ifdef USE_WIDE_CURSES
static int
have_locale(void)
{
static int result = -1;
if (result < 0) {
char *test = setlocale(LC_ALL, 0);
if (test == 0 || *test == 0) {
result = FALSE;
} else if (strcmp(test, "C") && strcmp(test, "POSIX")) {
result = TRUE;
} else {
result = FALSE;
}
}
return result;
}
#endif
#ifdef HAVE_TSEARCH
#if 0
static void
show_tsearch(const void *nodep, const VISIT which, const int depth)
{
const CACHE *p = *(CACHE * const *) nodep;
(void) depth;
if (which == postorder || which == leaf) {
dlg_trace_msg("\tcache %p %p:%s\n", p, p->string, p->string);
}
}
static void
trace_cache(const char *fn, int ln)
{
dlg_trace_msg("trace_cache %s@%d\n", fn, ln);
twalk(sorted_cache, show_tsearch);
}
#else
#define trace_cache(fn, ln) /* nothing */
#endif
static int
compare_cache(const void *a, const void *b)
{
const CACHE *p = (const CACHE *) a;
const CACHE *q = (const CACHE *) b;
int result = (p->cache_num - q->cache_num);
if (result == 0)
result = (int) (p->string_at - q->string_at);
return result;
}
#endif
static CACHE *
find_cache(int cache_num, const char *string)
{
CACHE *p;
#ifdef HAVE_TSEARCH
void *pp;
CACHE find;
memset(&find, 0, sizeof(find));
find.cache_num = cache_num;
find.string_at = string;
if ((pp = tfind(&find, &sorted_cache, compare_cache)) != 0) {
p = *(CACHE **) pp;
} else {
p = 0;
}
#else
for (p = cache_list; p != 0; p = p->next) {
if (p->string_at == string) {
break;
}
}
#endif
return p;
}
static CACHE *
make_cache(int cache_num, const char *string)
{
CACHE *p;
p = dlg_calloc(CACHE, 1);
assert_ptr(p, "load_cache");
p->next = cache_list;
cache_list = p;
p->cache_num = cache_num;
p->string_at = string;
#ifdef HAVE_TSEARCH
(void) tsearch(p, &sorted_cache, compare_cache);
#endif
return p;
}
static CACHE *
load_cache(int cache_num, const char *string)
{
CACHE *p;
if ((p = find_cache(cache_num, string)) == 0) {
p = make_cache(cache_num, string);
}
return p;
}
#else
static CACHE my_cache;
#define SAME_CACHE(c,s,l) (c->string != 0)
#define load_cache(cache, string) &my_cache
#endif /* USE_CACHING */
/*
* If the given string has not changed, we do not need to update the index.
* If we need to update the index, allocate enough memory for it.
*/
static bool
same_cache2(CACHE * cache, const char *string, unsigned i_len)
{
unsigned need;
size_t s_len = strlen(string);
bool result = TRUE;
if (cache->s_len == 0
|| cache->s_len < s_len
|| cache->list == 0
|| !SAME_CACHE(cache, string, (size_t) s_len)) {
need = (i_len + 1);
if (cache->list == 0) {
cache->list = dlg_malloc(int, need);
} else if (cache->i_len < i_len) {
cache->list = dlg_realloc(int, need, cache->list);
}
assert_ptr(cache->list, "load_cache");
cache->i_len = i_len;
if (cache->s_len >= s_len && cache->string != 0) {
strcpy(cache->string, string);
} else {
if (cache->string != 0)
free(cache->string);
cache->string = dlg_strclone(string);
}
cache->s_len = s_len;
result = FALSE;
}
return result;
}
#ifdef USE_WIDE_CURSES
/*
* Like same_cache2(), but we are only concerned about caching a copy of the
* string and its associated length.
*/
static bool
same_cache1(CACHE * cache, const char *string, size_t i_len)
{
size_t s_len = strlen(string);
bool result = TRUE;
if (cache->s_len != s_len
|| !SAME_CACHE(cache, string, (size_t) s_len)) {
if (cache->s_len >= s_len && cache->string != 0) {
strcpy(cache->string, string);
} else {
if (cache->string != 0)
free(cache->string);
cache->string = dlg_strclone(string);
}
cache->s_len = s_len;
cache->i_len = i_len;
result = FALSE;
}
return result;
}
#endif /* USE_CACHING */
/*
* Counts the number of bytes that make up complete wide-characters, up to byte
* 'len'. If there is no locale set, simply return the original length.
*/
#ifdef USE_WIDE_CURSES
static int
dlg_count_wcbytes(const char *string, size_t len)
{
int result;
if (have_locale()) {
CACHE *cache = load_cache(cCntWideBytes, string);
if (!same_cache1(cache, string, len)) {
while (len != 0) {
size_t code = 0;
const char *src = cache->string;
mbstate_t state;
char save = cache->string[len];
cache->string[len] = '\0';
memset(&state, 0, sizeof(state));
code = mbsrtowcs((wchar_t *) 0, &src, len, &state);
cache->string[len] = save;
if ((int) code >= 0) {
break;
}
--len;
}
cache->i_len = len;
}
result = (int) cache->i_len;
} else {
result = (int) len;
}
return result;
}
#endif /* USE_WIDE_CURSES */
/*
* Counts the number of wide-characters in the string.
*/
int
dlg_count_wchars(const char *string)
{
int result;
#ifdef USE_WIDE_CURSES
if (have_locale()) {
size_t len = strlen(string);
CACHE *cache = load_cache(cCntWideChars, string);
if (!same_cache1(cache, string, len)) {
const char *src = cache->string;
mbstate_t state;
int part = dlg_count_wcbytes(cache->string, len);
char save = cache->string[part];
size_t code;
wchar_t *temp = dlg_calloc(wchar_t, len + 1);
if (temp != 0) {
cache->string[part] = '\0';
memset(&state, 0, sizeof(state));
code = mbsrtowcs(temp, &src, (size_t) part, &state);
cache->i_len = ((int) code >= 0) ? wcslen(temp) : 0;
cache->string[part] = save;
free(temp);
} else {
cache->i_len = 0;
}
}
result = (int) cache->i_len;
} else
#endif /* USE_WIDE_CURSES */
{
result = (int) strlen(string);
}
return result;
}
/*
* Build an index of the wide-characters in the string, so we can easily tell
* which byte-offset begins a given wide-character.
*/
const int *
dlg_index_wchars(const char *string)
{
unsigned len = (unsigned) dlg_count_wchars(string);
unsigned inx;
CACHE *cache = load_cache(cInxWideChars, string);
if (!same_cache2(cache, string, len)) {
const char *current = string;
cache->list[0] = 0;
for (inx = 1; inx <= len; ++inx) {
#ifdef USE_WIDE_CURSES
if (have_locale()) {
mbstate_t state;
int width;
memset(&state, 0, sizeof(state));
width = (int) mbrlen(current, strlen(current), &state);
if (width <= 0)
width = 1; /* FIXME: what if we have a control-char? */
current += width;
cache->list[inx] = cache->list[inx - 1] + width;
} else
#endif /* USE_WIDE_CURSES */
{
(void) current;
cache->list[inx] = (int) inx;
}
}
}
return cache->list;
}
/*
* Given the character-offset to find in the list, return the corresponding
* array index.
*/
int
dlg_find_index(const int *list, int limit, int to_find)
{
int result;
for (result = 0; result <= limit; ++result) {
if (to_find == list[result]
|| result == limit
|| ((result < limit) && (to_find < list[result + 1]))) {
break;
}
}
return result;
}
/*
* Build a list of the display-columns for the given string's characters.
*/
const int *
dlg_index_columns(const char *string)
{
unsigned len = (unsigned) dlg_count_wchars(string);
unsigned inx;
CACHE *cache = load_cache(cInxCols, string);
if (!same_cache2(cache, string, len)) {
cache->list[0] = 0;
#ifdef USE_WIDE_CURSES
if (have_locale()) {
size_t num_bytes = strlen(string);
const int *inx_wchars = dlg_index_wchars(string);
mbstate_t state;
for (inx = 0; inx < len; ++inx) {
wchar_t temp[2];
size_t check;
int result;
if (string[inx_wchars[inx]] == TAB) {
result = ((cache->list[inx] | 7) + 1) - cache->list[inx];
} else {
memset(&state, 0, sizeof(state));
memset(temp, 0, sizeof(temp));
check = mbrtowc(temp,
string + inx_wchars[inx],
num_bytes - (size_t) inx_wchars[inx],
&state);
if ((int) check <= 0) {
result = 1;
} else {
result = wcwidth(temp[0]);
}
if (result < 0) {
const wchar_t *printable;
cchar_t temp2, *temp2p = &temp2;
setcchar(temp2p, temp, 0, 0, 0);
printable = wunctrl(temp2p);
result = printable ? (int) wcslen(printable) : 1;
}
}
cache->list[inx + 1] = result;
if (inx != 0)
cache->list[inx + 1] += cache->list[inx];
}
} else
#endif /* USE_WIDE_CURSES */
{
for (inx = 0; inx < len; ++inx) {
chtype ch = UCH(string[inx]);
if (ch == TAB)
cache->list[inx + 1] =
((cache->list[inx] | 7) + 1) - cache->list[inx];
else if (isprint(ch))
cache->list[inx + 1] = 1;
else {
const char *printable;
printable = unctrl(ch);
cache->list[inx + 1] = (printable
? (int) strlen(printable)
: 1);
}
if (inx != 0)
cache->list[inx + 1] += cache->list[inx];
}
}
}
return cache->list;
}
/*
* Returns the number of columns used for a string. That happens to be the
* end-value of the cols[] array.
*/
int
dlg_count_columns(const char *string)
{
int result = 0;
int limit = dlg_count_wchars(string);
if (limit > 0) {
const int *cols = dlg_index_columns(string);
result = cols[limit];
} else {
result = (int) strlen(string);
}
dlg_finish_string(string);
return result;
}
/*
* Given a column limit, count the number of wide characters that can fit
* into that limit. The offset is used to skip over a leading character
* that was already written.
*/
int
dlg_limit_columns(const char *string, int limit, int offset)
{
const int *cols = dlg_index_columns(string);
int result = dlg_count_wchars(string);
while (result > 0 && (cols[result] - cols[offset]) > limit)
--result;
return result;
}
/*
* Updates the string and character-offset, given various editing characters
* or literal characters which are inserted at the character-offset.
*/
bool
dlg_edit_string(char *string, int *chr_offset, int key, int fkey, bool force)
{
int i;
int len = (int) strlen(string);
int limit = dlg_count_wchars(string);
const int *indx = dlg_index_wchars(string);
int offset = dlg_find_index(indx, limit, *chr_offset);
int max_len = dlg_max_input(MAX_LEN);
bool edit = TRUE;
/* transform editing characters into equivalent function-keys */
if (!fkey) {
fkey = TRUE; /* assume we transform */
switch (key) {
case 0:
break;
case ESC:
case TAB:
fkey = FALSE; /* this is used for navigation */
break;
default:
fkey = FALSE; /* ...no, we did not transform */
break;
}
}
if (fkey) {
switch (key) {
case 0: /* special case for loop entry */
edit = force;
break;
case DLGK_GRID_LEFT:
if (*chr_offset && offset > 0)
*chr_offset = indx[offset - 1];
break;
case DLGK_GRID_RIGHT:
if (offset < limit)
*chr_offset = indx[offset + 1];
break;
case DLGK_BEGIN:
if (*chr_offset)
*chr_offset = 0;
break;
case DLGK_FINAL:
if (offset < limit)
*chr_offset = indx[limit];
break;
case DLGK_DELETE_LEFT:
if (offset) {
int gap = indx[offset] - indx[offset - 1];
*chr_offset = indx[offset - 1];
if (gap > 0) {
for (i = *chr_offset;
(string[i] = string[i + gap]) != '\0';
i++) {
;
}
}
}
break;
case DLGK_DELETE_RIGHT:
if (limit) {
if (--limit == 0) {
string[*chr_offset = 0] = '\0';
} else {
int gap = ((offset <= limit)
? (indx[offset + 1] - indx[offset])
: 0);
if (gap > 0) {
for (i = indx[offset];
(string[i] = string[i + gap]) != '\0';
i++) {
;
}
} else if (offset > 0) {
string[indx[offset - 1]] = '\0';
}
if (*chr_offset > indx[limit])
*chr_offset = indx[limit];
}
}
break;
case DLGK_DELETE_ALL:
string[*chr_offset = 0] = '\0';
break;
case DLGK_ENTER:
edit = 0;
break;
#ifdef KEY_RESIZE
case KEY_RESIZE:
edit = 0;
break;
#endif
case DLGK_GRID_UP:
case DLGK_GRID_DOWN:
case DLGK_FIELD_NEXT:
case DLGK_FIELD_PREV:
edit = 0;
break;
case ERR:
edit = 0;
break;
default:
beep();
break;
}
} else {
if (key == ESC || key == ERR) {
edit = 0;
} else {
if (len < max_len) {
for (i = ++len; i > *chr_offset; i--)
string[i] = string[i - 1];
string[*chr_offset] = (char) key;
*chr_offset += 1;
} else {
(void) beep();
}
}
}
return edit;
}
static void
compute_edit_offset(const char *string,
int chr_offset,
int x_last,
int *p_dpy_column,
int *p_scroll_amt)
{
const int *cols = dlg_index_columns(string);
const int *indx = dlg_index_wchars(string);
int limit = dlg_count_wchars(string);
int offset = dlg_find_index(indx, limit, chr_offset);
int offset2;
int dpy_column;
int n;
for (n = offset2 = 0; n <= offset; ++n) {
if ((cols[offset] - cols[n]) < x_last
&& (offset == limit || (cols[offset + 1] - cols[n]) < x_last)) {
offset2 = n;
break;
}
}
dpy_column = cols[offset] - cols[offset2];
if (p_dpy_column != 0)
*p_dpy_column = dpy_column;
if (p_scroll_amt != 0)
*p_scroll_amt = offset2;
}
/*
* Given the character-offset in the string, returns the display-offset where
* we will position the cursor.
*/
int
dlg_edit_offset(char *string, int chr_offset, int x_last)
{
int result;
compute_edit_offset(string, chr_offset, x_last, &result, 0);
return result;
}
/*
* Displays the string, shifted as necessary, to fit within the box and show
* the current character-offset.
*/
void
dlg_show_string(WINDOW *win,
const char *string, /* string to display (may be multibyte) */
int chr_offset, /* character (not bytes) offset */
chtype attr, /* window-attributes */
int y_base, /* beginning row on screen */
int x_base, /* beginning column on screen */
int x_last, /* number of columns on screen */
bool hidden, /* if true, do not echo */
bool force) /* if true, force repaint */
{
x_last = MIN(x_last + x_base, getmaxx(win)) - x_base;
if (hidden && !dialog_vars.insecure) {
if (force) {
(void) wmove(win, y_base, x_base);
wrefresh(win);
}
} else {
const int *cols = dlg_index_columns(string);
const int *indx = dlg_index_wchars(string);
int limit = dlg_count_wchars(string);
int i, j, k;
int input_x;
int scrollamt;
compute_edit_offset(string, chr_offset, x_last, &input_x, &scrollamt);
(void) wattrset(win, attr);
(void) wmove(win, y_base, x_base);
for (i = scrollamt, k = 0; i < limit && k < x_last; ++i) {
int check = cols[i + 1] - cols[scrollamt];
if (check <= x_last) {
for (j = indx[i]; j < indx[i + 1]; ++j) {
chtype ch = UCH(string[j]);
if (hidden && dialog_vars.insecure) {
waddch(win, '*');
} else if (ch == TAB) {
int count = cols[i + 1] - cols[i];
while (--count >= 0)
waddch(win, ' ');
} else {
waddch(win, ch);
}
}
k = check;
} else {
break;
}
}
while (k++ < x_last)
waddch(win, ' ');
(void) wmove(win, y_base, x_base + input_x);
wrefresh(win);
}
}
/*
* Discard cached data for the given string.
*/
void
dlg_finish_string(const char *string)
{
#if USE_CACHING
if ((string != 0) && dialog_state.finish_string) {
CACHE *p = cache_list;
CACHE *q = 0;
CACHE *r;
while (p != 0) {
if (p->string_at == string) {
#ifdef HAVE_TSEARCH
if (tdelete(p, &sorted_cache, compare_cache) == 0) {
continue;
}
trace_cache(__FILE__, __LINE__);
#endif
if (p->string != 0)
free(p->string);
if (p->list != 0)
free(p->list);
if (p == cache_list) {
cache_list = p->next;
r = cache_list;
} else {
q->next = p->next;
r = q;
}
free(p);
p = r;
} else {
q = p;
p = p->next;
}
}
}
#else
(void) string;
#endif
}
#ifdef NO_LEAKS
void
_dlg_inputstr_leaks(void)
{
#if USE_CACHING
dialog_state.finish_string = TRUE;
trace_cache(__FILE__, __LINE__);
while (cache_list != 0) {
dlg_finish_string(cache_list->string_at);
}
#endif /* USE_CACHING */
}
#endif /* NO_LEAKS */