summaryrefslogtreecommitdiff
path: root/console-tools/loadacm.c
diff options
context:
space:
mode:
Diffstat (limited to 'console-tools/loadacm.c')
-rw-r--r--console-tools/loadacm.c354
1 files changed, 0 insertions, 354 deletions
diff --git a/console-tools/loadacm.c b/console-tools/loadacm.c
deleted file mode 100644
index edaf51a..0000000
--- a/console-tools/loadacm.c
+++ /dev/null
@@ -1,354 +0,0 @@
-/* vi: set sw=4 ts=4: */
-/*
- * Derived from
- * mapscrn.c - version 0.92
- *
- * Was taken from console-tools and adapted by
- * Peter Novodvorsky <petya@logic.ru>
- */
-
-#include <stdio.h>
-#include <stdlib.h>
-#include <memory.h>
-#include <string.h>
-#include <unistd.h>
-#include <fcntl.h>
-#include <assert.h>
-#include <errno.h>
-#include <signal.h>
-#include <sys/kd.h>
-#include <sys/types.h>
-#include <sys/ioctl.h>
-#include "busybox.h"
-
-typedef unsigned short unicode;
-
-static long int ctoi(unsigned char *s, int *is_unicode);
-static int old_screen_map_read_ascii(FILE * fp, unsigned char buf[]);
-static int uni_screen_map_read_ascii(FILE * fp, unicode buf[], int *is_unicode);
-static unicode utf8_to_ucs2(char *buf);
-static int screen_map_load(int fd, FILE * fp);
-
-int loadacm_main(int argc, char **argv)
-{
- int fd;
-
- if (argc>=2 && *argv[1]=='-') {
- bb_show_usage();
- }
-
- fd = bb_xopen(CURRENT_VC, O_RDWR);
-
- if (screen_map_load(fd, stdin)) {
- bb_perror_msg_and_die("Error loading acm");
- }
-
- write(fd, "\033(K", 3);
-
- return EXIT_SUCCESS;
-}
-
-static int screen_map_load(int fd, FILE * fp)
-{
- struct stat stbuf;
- unicode wbuf[E_TABSZ];
- unsigned char buf[E_TABSZ];
- int parse_failed = 0;
- int is_unicode;
-
- if (fstat(fileno(fp), &stbuf))
- bb_perror_msg_and_die("Cannot stat map file");
-
- /* first try a UTF screen-map: either ASCII (no restriction) or binary (regular file) */
- if (!
- (parse_failed =
- (-1 == uni_screen_map_read_ascii(fp, wbuf, &is_unicode)))
-|| (S_ISREG(stbuf.st_mode) && (stbuf.st_size == (sizeof(unicode) * E_TABSZ)))) { /* test for binary UTF map by size */
- if (parse_failed) {
- if (-1 == fseek(fp, 0, SEEK_SET)) {
- if (errno == ESPIPE)
- bb_error_msg_and_die("16bit screen-map MUST be a regular file.");
- else
- bb_perror_msg_and_die("fseek failed reading binary 16bit screen-map");
- }
-
- if (fread(wbuf, sizeof(unicode) * E_TABSZ, 1, fp) != 1)
- bb_perror_msg_and_die("Cannot read [new] map from file");
-#if 0
- else
- bb_error_msg("Input screen-map is binary.");
-#endif
- }
-
- /* if it was effectively a 16-bit ASCII, OK, else try to read as 8-bit map */
- /* same if it was binary, ie. if parse_failed */
- if (parse_failed || is_unicode) {
- if (ioctl(fd, PIO_UNISCRNMAP, wbuf))
- bb_perror_msg_and_die("PIO_UNISCRNMAP ioctl");
- else
- return 0;
- }
- }
-
- /* rewind... */
- if (-1 == fseek(fp, 0, SEEK_SET)) {
- if (errno == ESPIPE)
- bb_error_msg("Assuming 8bit screen-map - MUST be a regular file."),
- exit(1);
- else
- bb_perror_msg_and_die("fseek failed assuming 8bit screen-map");
- }
-
- /* ... and try an old 8-bit screen-map */
- if (!(parse_failed = (-1 == old_screen_map_read_ascii(fp, buf))) ||
- (S_ISREG(stbuf.st_mode) && (stbuf.st_size == E_TABSZ))) { /* test for binary old 8-bit map by size */
- if (parse_failed) {
- if (-1 == fseek(fp, 0, SEEK_SET)) {
- if (errno == ESPIPE)
- /* should not - it succedeed above */
- bb_error_msg_and_die("fseek() returned ESPIPE !");
- else
- bb_perror_msg_and_die("fseek for binary 8bit screen-map");
- }
-
- if (fread(buf, E_TABSZ, 1, fp) != 1)
- bb_perror_msg_and_die("Cannot read [old] map from file");
-#if 0
- else
- bb_error_msg("Input screen-map is binary.");
-#endif
- }
-
- if (ioctl(fd, PIO_SCRNMAP, buf))
- bb_perror_msg_and_die("PIO_SCRNMAP ioctl");
- else
- return 0;
- }
- bb_error_msg("Error parsing symbolic map");
- return(1);
-}
-
-
-/*
- * - reads `fp' as a 16-bit ASCII SFM file.
- * - returns -1 on error.
- * - returns it in `unicode' in an E_TABSZ-elements array.
- * - sets `*is_unicode' flagiff there were any non-8-bit
- * (ie. real 16-bit) mapping.
- *
- * FIXME: ignores everything after second word
- */
-static int uni_screen_map_read_ascii(FILE * fp, unicode buf[], int *is_unicode)
-{
- char buffer[256]; /* line buffer reading file */
- char *p, *q; /* 1st + 2nd words in line */
- int in, on; /* the same, as numbers */
- int tmp_is_unicode; /* tmp for is_unicode calculation */
- int i; /* loop index - result holder */
- int ret_code = 0; /* return code */
- sigset_t acmsigset, old_sigset;
-
- assert(is_unicode);
-
- *is_unicode = 0;
-
- /* first 128 codes defaults to ASCII */
- for (i = 0; i < 128; i++)
- buf[i] = i;
- /* remaining defaults to replacement char (usually E_TABSZ = 256) */
- for (; i < E_TABSZ; i++)
- buf[i] = 0xfffd;
-
- /* block SIGCHLD */
- sigemptyset(&acmsigset);
- sigaddset(&acmsigset, SIGCHLD);
- sigprocmask(SIG_BLOCK, &acmsigset, &old_sigset);
-
- do {
- if (NULL == fgets(buffer, sizeof(buffer), fp)) {
- if (feof(fp))
- break;
- else
- bb_perror_msg_and_die("uni_screen_map_read_ascii() can't read line");
- }
-
- /* get "charset-relative charcode", stripping leading spaces */
- p = strtok(buffer, " \t\n");
-
- /* skip empty lines and comments */
- if (!p || *p == '#')
- continue;
-
- /* get unicode mapping */
- q = strtok(NULL, " \t\n");
- if (q) {
- in = ctoi(p, NULL);
- if (in < 0 || in > 255) {
- ret_code = -1;
- break;
- }
-
- on = ctoi(q, &tmp_is_unicode);
- if (in < 0 && on > 65535) {
- ret_code = -1;
- break;
- }
-
- *is_unicode |= tmp_is_unicode;
- buf[in] = on;
- } else {
- ret_code = -1;
- break;
- }
- }
- while (1); /* terminated by break on feof() */
-
- /* restore sig mask */
- sigprocmask(SIG_SETMASK, &old_sigset, NULL);
-
- return ret_code;
-}
-
-
-static int old_screen_map_read_ascii(FILE * fp, unsigned char buf[])
-{
- char buffer[256];
- int in, on;
- char *p, *q;
-
- for (in = 0; in < 256; in++)
- buf[in] = in;
-
- while (fgets(buffer, sizeof(buffer) - 1, fp)) {
- p = strtok(buffer, " \t\n");
-
- if (!p || *p == '#')
- continue;
-
- q = strtok(NULL, " \t\n#");
- if (q) {
- in = ctoi(p, NULL);
- if (in < 0 || in > 255)
- return -1;
-
- on = ctoi(q, NULL);
- if (in < 0 && on > 255)
- return -1;
-
- buf[in] = on;
- } else
- return -1;
- }
-
- return (0);
-}
-
-
-/*
- * - converts a string into an int.
- * - supports dec and hex bytes, hex UCS2, single-quoted byte and UTF8 chars.
- * - returns the converted value
- * - if `is_unicode != NULL', use it to tell whether it was unicode
- *
- * CAVEAT: will report valid UTF mappings using only 1 byte as 8-bit ones.
- */
-static long int ctoi(unsigned char *s, int *is_unicode)
-{
- int i;
- size_t ls;
-
- ls = strlen(s);
- if (is_unicode)
- *is_unicode = 0;
-
- /* hex-specified UCS2 */
- if ((strncmp(s, "U+", 2) == 0) &&
- (strspn(s + 2, "0123456789abcdefABCDEF") == ls - 2)) {
- sscanf(s + 2, "%x", &i);
- if (is_unicode)
- *is_unicode = 1;
- }
-
- /* hex-specified byte */
- else if ((ls <= 4) && (strncmp(s, "0x", 2) == 0) &&
- (strspn(s + 2, "0123456789abcdefABCDEF") == ls - 2))
- sscanf(s + 2, "%x", &i);
-
- /* oct-specified number (byte) */
- else if ((*s == '0') && (strspn(s, "01234567") == ls))
- sscanf(s, "%o", &i);
-
- /* dec-specified number (byte) */
- else if (strspn(s, "0123456789") == ls)
- sscanf(s, "%d", &i);
-
- /* single-byte quoted char */
- else if ((strlen(s) == 3) && (s[0] == '\'') && (s[2] == '\''))
- i = s[1];
-
- /* multi-byte UTF8 quoted char */
- else if ((s[0] == '\'') && (s[ls - 1] == '\'')) {
- s[ls - 1] = 0; /* ensure we'll not "parse UTF too far" */
- i = utf8_to_ucs2(s + 1);
- if (is_unicode)
- *is_unicode = 1;
- } else
- return (-1);
-
- return (i);
-}
-
-
-static unicode utf8_to_ucs2(char *buf)
-{
- int utf_count = 0;
- long utf_char = 0;
- unicode tc = 0;
- unsigned char c;
-
- do {
- c = *buf;
- buf++;
-
- /* if byte should be part of multi-byte sequence */
- if (c & 0x80) {
- /* if we have already started to parse a UTF8 sequence */
- if (utf_count > 0 && (c & 0xc0) == 0x80) {
- utf_char = (utf_char << 6) | (c & 0x3f);
- utf_count--;
- if (utf_count == 0)
- tc = utf_char;
- else
- continue;
- } else { /* Possibly 1st char of a UTF8 sequence */
-
- if ((c & 0xe0) == 0xc0) {
- utf_count = 1;
- utf_char = (c & 0x1f);
- } else if ((c & 0xf0) == 0xe0) {
- utf_count = 2;
- utf_char = (c & 0x0f);
- } else if ((c & 0xf8) == 0xf0) {
- utf_count = 3;
- utf_char = (c & 0x07);
- } else if ((c & 0xfc) == 0xf8) {
- utf_count = 4;
- utf_char = (c & 0x03);
- } else if ((c & 0xfe) == 0xfc) {
- utf_count = 5;
- utf_char = (c & 0x01);
- } else
- utf_count = 0;
- continue;
- }
- } else { /* not part of multi-byte sequence - treat as ASCII
- * this makes incomplete sequences to be ignored
- */
- tc = c;
- utf_count = 0;
- }
- }
- while (utf_count);
-
- return tc;
-}