You cannot select more than 25 topics
Topics must start with a letter or number, can include dashes ('-') and can be up to 35 characters long.
143 lines
6.4 KiB
C
143 lines
6.4 KiB
C
/* this header file comes from libowfat, http://www.fefe.de/libowfat/ */
|
|
#ifndef TEXTCODE_H
|
|
#define TEXTCODE_H
|
|
|
|
#include <stddef.h>
|
|
|
|
#ifdef __cplusplus
|
|
extern "C" {
|
|
#endif
|
|
|
|
/* These take len bytes from src and write them in encoded form to
|
|
* dest (if dest != NULL), returning the number of bytes written. */
|
|
|
|
/* Needs len/3*4 bytes */
|
|
size_t fmt_uuencoded(char* dest,const char* src,size_t len);
|
|
/* Needs len/3*4 bytes */
|
|
size_t fmt_base64(char* dest,const char* src,size_t len);
|
|
/* Worst case: len*3 */
|
|
size_t fmt_quotedprintable(char* dest,const char* src,size_t len);
|
|
/* Worst case: len*3 */
|
|
size_t fmt_quotedprintable2(char* dest,const char* src,size_t len,const char* escapeme);
|
|
/* Worst case: len*3 */
|
|
size_t fmt_urlencoded(char* dest,const char* src,size_t len);
|
|
/* Worst case: len*3 */
|
|
size_t fmt_urlencoded2(char* dest,const char* src,size_t len,const char* escapeme);
|
|
/* Worst case: len*2 */
|
|
size_t fmt_yenc(char* dest,const char* src,size_t len);
|
|
/* Needs len*2 bytes */
|
|
size_t fmt_hexdump(char* dest,const char* src,size_t len);
|
|
/* Change '<' to '<' and '&' to '&' and '\n' to '<br>'; worst case: len*5 */
|
|
/* This is meant for outputting text that goes between tags */
|
|
size_t fmt_html(char* dest,const char* src,size_t len);
|
|
/* Change '<' to '<' and '&' to '&' and '"' to '"'; worst case: len*6 */
|
|
/* This is meant for outputting text that goes in a tag argument between double quotes*/
|
|
size_t fmt_html_tagarg(char* dest,const char* src,size_t len);
|
|
/* Change '<' to '<' and '&' to '&'; worst case: len*5 */
|
|
size_t fmt_xml(char* dest,const char* src,size_t len);
|
|
|
|
/* Change '\' to "\\", '\n' to "\n", ^A to "\x01" etc; worst case: len*4 */
|
|
size_t fmt_cescape(char* dest,const char* src,size_t len);
|
|
/* Worst case: len*4 */
|
|
size_t fmt_cescape2(char* dest,const char* src,size_t len,const char* escapeme);
|
|
/* Fold whitespace to '_'; this is great for writing fields with
|
|
* white spaces to a log file and still allow awk to do log analysis */
|
|
/* Worst case: same size */
|
|
size_t fmt_foldwhitespace(char* dest,const char* src,size_t len);
|
|
/* Worst case: len*3 */
|
|
size_t fmt_ldapescape(char* dest,const char* src,size_t len);
|
|
size_t fmt_ldapescape2(char* dest,const char* src,size_t len,const char* escapeme);
|
|
/* Encode JSON string from UTF-8; will backslash-escape the bare minimum.
|
|
* Will not verify that the input is valid UTF-8!
|
|
* Worst case: len*6 */
|
|
size_t fmt_jsonescape(char* dest,const char* src,size_t len);
|
|
|
|
size_t fmt_base85(char* dest,const char* src,size_t len);
|
|
|
|
/* These read one line from src, decode it, and write the result to
|
|
* dest. The number of decoded bytes is written to destlen. dest
|
|
* should be able to hold strlen(src) bytes as a rule of thumb. */
|
|
size_t scan_uuencoded(const char *src,char *dest,size_t *destlen);
|
|
size_t scan_base64(const char *src,char *dest,size_t *destlen);
|
|
size_t scan_quotedprintable(const char *src,char *dest,size_t *destlen);
|
|
size_t scan_urlencoded(const char *src,char *dest,size_t *destlen);
|
|
size_t scan_urlencoded2(const char *src,char *dest,size_t *destlen);
|
|
size_t scan_yenc(const char *src,char *dest,size_t *destlen);
|
|
size_t scan_hexdump(const char *src,char *dest,size_t *destlen);
|
|
size_t scan_html(const char *src,char *dest,size_t *destlen);
|
|
size_t scan_cescape(const char *src,char *dest,size_t *destlen);
|
|
size_t scan_ldapescape(const char* src,char* dest,size_t *destlen);
|
|
size_t scan_jsonescape(const char* src,char* dest,size_t *destlen);
|
|
|
|
size_t scan_base85(const char* src,char* dest,size_t *destlen);
|
|
|
|
#ifdef STRALLOC_H
|
|
/* WARNING: these functions _append_ to the stralloc, not overwrite! */
|
|
/* stralloc wrappers; return 1 on success, 0 on failure */
|
|
/* arg 1 is one of the fmt_* functions from above */
|
|
int fmt_to_sa(size_t (*func)(char*,const char*,size_t),
|
|
stralloc* sa,const char* src,size_t len);
|
|
|
|
int fmt_to_sa2(size_t (*func)(char*,const char*,size_t,const char*),
|
|
stralloc* sa,const char* src,size_t len,const char* escapeme);
|
|
|
|
/* arg 1 is one of the scan_* functions from above */
|
|
/* return number of bytes scanned */
|
|
size_t scan_to_sa(size_t (*func)(const char*,char*,size_t*),
|
|
const char* src,stralloc* sa);
|
|
|
|
#define fmt_uuencoded_sa(sa,src,len) fmt_to_sa(fmt_uuencoded,sa,src,len)
|
|
#define fmt_base64_sa(sa,src,len) fmt_to_sa(fmt_base64,sa,src,len)
|
|
#define fmt_quotedprintable_sa(sa,src,len) fmt_to_sa(fmt_quotedprintable,sa,src,len)
|
|
#define fmt_urlencoded_sa(sa,src,len) fmt_to_sa(fmt_urlencoded,sa,src,len)
|
|
#define fmt_yenc_sa(sa,src,len) fmt_to_sa(fmt_yenc,sa,src,len)
|
|
#define fmt_hexdump_sa(sa,src,len) fmt_to_sa(fmt_hexdump,sa,src,len)
|
|
#define fmt_html_sa(sa,src,len) fmt_to_sa(fmt_html,sa,src,len)
|
|
#define fmt_cescape_sa(sa,src,len) fmt_to_sa(fmt_cescape,sa,src,len)
|
|
#define fmt_ldapescape_sa(sa,src,len) fmt_to_sa(fmt_ldapescape,sa,src,len)
|
|
#define fmt_jsonescape_sa(sa,src,len) fmt_to_sa(fmt_jsonescape,sa,src,len)
|
|
|
|
#define fmt_quotedprintable2_sa(sa,src,len,escapeme) fmt_to_sa2(fmt_quotedprintable2,sa,src,len,escapeme)
|
|
#define fmt_urlencoded2_sa(sa,src,len,escapeme) fmt_to_sa2(fmt_urlencoded2,sa,src,len,escapeme)
|
|
#define fmt_cescape2_sa(sa,src,len,escapeme) fmt_to_sa2(fmt_cescape2,sa,src,len,escapeme)
|
|
|
|
#define scan_uuencoded_sa(src,sa) scan_to_sa(scan_uuencoded,src,sa)
|
|
#define scan_base64_sa(src,sa) scan_to_sa(scan_base64,src,sa)
|
|
#define scan_quotedprintable_sa(src,sa) scan_to_sa(scan_quotedprintable,src,sa)
|
|
#define scan_urlencoded_sa(src,sa) scan_to_sa(scan_urlencoded,src,sa)
|
|
#define scan_yenc_sa(src,sa) scan_to_sa(scan_yenc,src,sa)
|
|
#define scan_hexdump_sa(src,sa) scan_to_sa(scan_hexdump,src,sa)
|
|
#define scan_html_sa(src,sa) scan_to_sa(scan_html,src,sa)
|
|
#define scan_cescape_sa(src,sa) scan_to_sa(scan_cescape,src,sa)
|
|
#define scan_ldapescape_sa(src,sa) scan_to_sa(scan_ldapescape,src,sa)
|
|
#define scan_jsonescape_sa(src,sa) scan_to_sa(scan_jsonescape,src,sa)
|
|
#endif
|
|
|
|
#ifdef ARRAY_H
|
|
void fmt_to_array(size_t (*func)(char*,const char*,size_t),
|
|
array* a,const char* src,size_t len);
|
|
|
|
void fmt_tofrom_array(size_t (*func)(char*,const char*,size_t),
|
|
array* dest,array* src);
|
|
|
|
void fmt_to_array2(size_t (*func)(char*,const char*,size_t,const char*),
|
|
array* a,const char* src,size_t len,const char* escapeme);
|
|
|
|
void fmt_tofrom_array2(size_t (*func)(char*,const char*,size_t,const char*),
|
|
array* dest,array* src,const char* escapeme);
|
|
|
|
size_t scan_to_array(size_t (*func)(const char*,char*,size_t*),
|
|
const char* src,array* dest);
|
|
|
|
size_t scan_tofrom_array(size_t (*func)(const char*,char*,size_t*),
|
|
array* src,array* dest);
|
|
#endif
|
|
|
|
extern const char base64[64];
|
|
|
|
#ifdef __cplusplus
|
|
}
|
|
#endif
|
|
|
|
#endif
|