107 lines
2.4 KiB
C
107 lines
2.4 KiB
C
/*
|
|
* Copyright (C) 1984-2015 Mark Nudelman
|
|
*
|
|
* You may distribute under the terms of either the GNU General Public
|
|
* License or the Less License, as specified in the README file.
|
|
*
|
|
* For more information, see the README file.
|
|
*/
|
|
|
|
/*
|
|
* Routines to convert text in various ways. Used by search.
|
|
*/
|
|
|
|
#include "less.h"
|
|
#include "charset.h"
|
|
|
|
extern int utf_mode;
|
|
|
|
/*
|
|
* Get the length of a buffer needed to convert a string.
|
|
*/
|
|
public int
|
|
cvt_length(int len, int ops)
|
|
{
|
|
if (utf_mode)
|
|
/*
|
|
* Just copying a string in UTF-8 mode can cause it to grow
|
|
* in length.
|
|
* Four output bytes for one input byte is the worst case.
|
|
*/
|
|
len *= 4;
|
|
return (len + 1);
|
|
}
|
|
|
|
/*
|
|
* Allocate a chpos array for use by cvt_text.
|
|
*/
|
|
public int *
|
|
cvt_alloc_chpos(int len)
|
|
{
|
|
int i;
|
|
int *chpos = (int *) ecalloc(sizeof(int), len);
|
|
/* Initialize all entries to an invalid position. */
|
|
for (i = 0; i < len; i++)
|
|
chpos[i] = -1;
|
|
return (chpos);
|
|
}
|
|
|
|
/*
|
|
* Convert text. Perform the transformations specified by ops.
|
|
* Returns converted text in odst. The original offset of each
|
|
* odst character (when it was in osrc) is returned in the chpos array.
|
|
*/
|
|
public void
|
|
cvt_text(char *odst, char *osrc, int *chpos, int *lenp, int ops)
|
|
{
|
|
char *dst;
|
|
char *edst = odst;
|
|
char *src;
|
|
char *src_end;
|
|
LWCHAR ch;
|
|
|
|
if (lenp != NULL)
|
|
src_end = osrc + *lenp;
|
|
else
|
|
src_end = osrc + strlen(osrc);
|
|
|
|
for (src = osrc, dst = odst; src < src_end; )
|
|
{
|
|
int src_pos = (int) (src - osrc);
|
|
int dst_pos = (int) (dst - odst);
|
|
ch = step_char((constant char **)&src, +1, src_end);
|
|
if ((ops & CVT_BS) && ch == '\b' && dst > odst)
|
|
{
|
|
/* Delete backspace and preceding char. */
|
|
do {
|
|
dst--;
|
|
} while (dst > odst &&
|
|
!IS_ASCII_OCTET(*dst) && !IS_UTF8_LEAD(*dst));
|
|
} else if ((ops & CVT_ANSI) && IS_CSI_START(ch))
|
|
{
|
|
/* Skip to end of ANSI escape sequence. */
|
|
src++; /* skip the CSI start char */
|
|
while (src < src_end)
|
|
if (!is_ansi_middle(*src++))
|
|
break;
|
|
} else
|
|
{
|
|
/* Just copy the char to the destination buffer. */
|
|
if ((ops & CVT_TO_LC) && IS_UPPER(ch))
|
|
ch = TO_LOWER(ch);
|
|
put_wchar(&dst, ch);
|
|
/* Record the original position of the char. */
|
|
if (chpos != NULL)
|
|
chpos[dst_pos] = src_pos;
|
|
}
|
|
if (dst > edst)
|
|
edst = dst;
|
|
}
|
|
if ((ops & CVT_CRLF) && edst > odst && edst[-1] == '\r')
|
|
edst--;
|
|
*edst = '\0';
|
|
if (lenp != NULL)
|
|
*lenp = (int) (edst - odst);
|
|
/* FIXME: why was this here? if (chpos != NULL) chpos[dst - odst] = src - osrc; */
|
|
}
|