+ }
+ /* move the newline over the '\' before it (effectively eats the '\') */
+ memmove(&ptr[idx], &ptr[idx+1], strlen(&ptr[idx+1]));
+ ptr[strlen(ptr)-1] = 0;
+ slashes_eaten++;
+ /* substitue \r for \n if needed */
+ if (ptr[idx] == '\r')
+ ptr[idx] = '\n';
+ }
+
+out:
+ /* this accounts for discrepancies between the modified string and the
+ * original string passed in to this function */
+ idx += slashes_eaten;
+
+ /* figure out if we need to add a newline */
+ if (ptr[idx-1] != '\n') {
+ ptr[idx] = '\n';
+ idx++;
+ }
+
+ /* terminate string */
+ ptr[idx]= 0;
+ /* adjust for opening 2 chars [aic]\ */
+ idx += 2;
+
+ return idx;
+}
+
+
+static int parse_file_cmd(struct sed_cmd *sed_cmd, const char *filecmdstr)
+{
+ int idx = 0;
+ int filenamelen = 0;
+
+ /*
+ * the string that gets passed to this function should look like this:
+ * '[ ]filename'
+ * | |
+ * | a filename
+ * |
+ * optional whitespace
+
+ * re: the file to be read, the GNU manual says the following: "Note that
+ * if filename cannot be read, it is treated as if it were an empty file,
+ * without any error indication." Thus, all of the following commands are
+ * perfectly leagal:
+ *
+ * sed -e '1r noexist'
+ * sed -e '1r ;'
+ * sed -e '1r'
+ */
+
+ /* the file command may be followed by whitespace; move past it. */
+ while (isspace(filecmdstr[++idx]))
+ { ; }
+
+ /* the first non-whitespace we get is a filename. the filename ends when we
+ * hit a normal sed command terminator or end of string */
+ filenamelen = strcspn(&filecmdstr[idx], "; \n\r\t\v\0");
+ sed_cmd->filename = xmalloc(filenamelen + 1);
+ safe_strncpy(sed_cmd->filename, &filecmdstr[idx], filenamelen + 1);
+
+ return idx + filenamelen;
+}
+
+
+static char *parse_cmd_str(struct sed_cmd *sed_cmd, const char *cmdstr)
+{
+ int idx = 0;
+
+ /* parse the command
+ * format is: [addr][,addr]cmd
+ * |----||-----||-|
+ * part1 part2 part3
+ */
+
+ /* first part (if present) is an address: either a number or a /regex/ */
+ if (isdigit(cmdstr[idx]) || cmdstr[idx] == '/')
+ idx = get_address(sed_cmd, cmdstr, &sed_cmd->beg_line, &sed_cmd->beg_match);
+
+ /* second part (if present) will begin with a comma */
+ if (cmdstr[idx] == ',')
+ idx += get_address(sed_cmd, &cmdstr[++idx], &sed_cmd->end_line, &sed_cmd->end_match);
+
+ /* last part (mandatory) will be a command */
+ if (cmdstr[idx] == '\0')
+ error_msg_and_die("missing command");
+ sed_cmd->cmd = cmdstr[idx];
+
+ /* if it was a single-letter command that takes no arguments (such as 'p'
+ * or 'd') all we need to do is increment the index past that command */
+ if (strchr("pd", cmdstr[idx])) {
+ idx++;
+ }
+ /* handle (s)ubstitution command */
+ else if (sed_cmd->cmd == 's') {
+ idx += parse_subst_cmd(sed_cmd, &cmdstr[idx]);
+ }
+ /* handle edit cmds: (a)ppend, (i)nsert, and (c)hange */
+ else if (strchr("aic", sed_cmd->cmd)) {
+ if ((sed_cmd->end_line || sed_cmd->end_match) && sed_cmd->cmd != 'c')
+ error_msg_and_die("only a beginning address can be specified for edit commands");
+ idx += parse_edit_cmd(sed_cmd, &cmdstr[idx]);
+ }
+ /* handle file cmds: (r)ead */
+ else if (sed_cmd->cmd == 'r') {
+ if (sed_cmd->end_line || sed_cmd->end_match)
+ error_msg_and_die("Command only uses one address");
+ idx += parse_file_cmd(sed_cmd, &cmdstr[idx]);
+ }
+ else {
+ error_msg_and_die("invalid command");
+ }
+
+ /* give back whatever's left over */
+ return (char *)&cmdstr[idx];
+}
+
+static void add_cmd_str(const char *cmdstr)
+{
+ char *mystr = (char *)cmdstr;
+
+ do {
+
+ /* trim leading whitespace and semicolons */
+ memmove(mystr, &mystr[strspn(mystr, "; \n\r\t\v")], strlen(mystr));
+ /* if we ate the whole thing, that means there was just trailing
+ * whitespace or a final / no-op semicolon. either way, get out */
+ if (strlen(mystr) == 0)
+ return;
+ /* if this is a comment, jump past it and keep going */
+ if (mystr[0] == '#') {
+ mystr = strpbrk(mystr, ";\n\r");
+ continue;
+ }
+ /* grow the array */
+ sed_cmds = xrealloc(sed_cmds, sizeof(struct sed_cmd) * (++ncmds));
+ /* zero new element */
+ memset(&sed_cmds[ncmds-1], 0, sizeof(struct sed_cmd));
+ /* load command string into new array element, get remainder */
+ mystr = parse_cmd_str(&sed_cmds[ncmds-1], mystr);
+
+ } while (mystr && strlen(mystr));
+}
+
+
+static void load_cmd_file(char *filename)
+{
+ FILE *cmdfile;
+ char *line;
+ char *nextline;
+
+ cmdfile = xfopen(filename, "r");
+
+ while ((line = get_line_from_file(cmdfile)) != NULL) {
+ /* if a line ends with '\' it needs the next line appended to it */
+ while (line[strlen(line)-2] == '\\' &&
+ (nextline = get_line_from_file(cmdfile)) != NULL) {
+ line = xrealloc(line, strlen(line) + strlen(nextline) + 1);
+ strcat(line, nextline);
+ free(nextline);
+ }
+ /* eat trailing newline (if any) --if I don't do this, edit commands
+ * (aic) will print an extra newline */
+ chomp(line);
+ add_cmd_str(line);
+ free(line);
+ }
+}
+
+static void print_subst_w_backrefs(const char *line, const char *replace, regmatch_t *regmatch)
+{
+ int i;
+
+ /* go through the replacement string */
+ for (i = 0; replace[i]; i++) {
+ /* if we find a backreference (\1, \2, etc.) print the backref'ed * text */
+ if (replace[i] == '\\' && isdigit(replace[i+1])) {
+ int j;
+ char tmpstr[2];
+ int backref;
+ ++i; /* i now indexes the backref number, instead of the leading slash */
+ tmpstr[0] = replace[i];
+ tmpstr[1] = 0;
+ backref = atoi(tmpstr);
+ /* print out the text held in regmatch[backref] */
+ for (j = regmatch[backref].rm_so; j < regmatch[backref].rm_eo; j++)
+ fputc(line[j], stdout);
+ }
+
+ /* if we find a backslash escaped character, print the character */
+ else if (replace[i] == '\\') {
+ ++i;
+ fputc(replace[i], stdout);
+ }
+
+ /* if we find an unescaped '&' print out the whole matched text.
+ * fortunately, regmatch[0] contains the indicies to the whole matched
+ * expression (kinda seems like it was designed for just such a
+ * purpose...) */
+ else if (replace[i] == '&' && replace[i-1] != '\\') {
+ int j;
+ for (j = regmatch[0].rm_so; j < regmatch[0].rm_eo; j++)
+ fputc(line[j], stdout);
+ }
+ /* nothing special, just print this char of the replacement string to stdout */
+ else
+ fputc(replace[i], stdout);
+ }
+}
+
+static int do_subst_command(const struct sed_cmd *sed_cmd, const char *line)
+{
+ char *hackline = (char *)line;
+ int altered = 0;
+ regmatch_t *regmatch = NULL;
+
+ /* we only proceed if the substitution 'search' expression matches */
+ if (regexec(sed_cmd->sub_match, line, 0, NULL, 0) == REG_NOMATCH)
+ return 0;
+
+ /* whaddaya know, it matched. get the number of back references */
+ regmatch = xmalloc(sizeof(regmatch_t) * (sed_cmd->num_backrefs+1));
+
+ /* and now, as long as we've got a line to try matching and if we can match
+ * the search string, we make substitutions */
+ while (*hackline && (regexec(sed_cmd->sub_match, hackline,
+ sed_cmd->num_backrefs+1, regmatch, 0) == 0) ) {
+ int i;
+
+ /* print everything before the match */
+ for (i = 0; i < regmatch[0].rm_so; i++)
+ fputc(hackline[i], stdout);
+
+ /* then print the substitution string */
+ print_subst_w_backrefs(hackline, sed_cmd->replace, regmatch);
+
+ /* advance past the match */
+ hackline += regmatch[0].rm_eo;
+ /* flag that something has changed */
+ altered++;
+
+ /* if we're not doing this globally, get out now */
+ if (!sed_cmd->sub_g)
+ break;
+ }
+
+ puts(hackline);
+
+ /* cleanup */
+ free(regmatch);
+
+ return altered;
+}
+
+
+static void process_file(FILE *file)
+{
+ char *line = NULL;
+ static int linenum = 0; /* GNU sed does not restart counting lines at EOF */
+ unsigned int still_in_range = 0;
+ int altered;
+ int i;
+
+ /* go through every line in the file */
+ while ((line = get_line_from_file(file)) != NULL) {
+
+ chomp(line);
+ linenum++;
+ altered = 0;
+
+ /* for every line, go through all the commands */
+ for (i = 0; i < ncmds; i++) {
+
+
+ /*
+ * entry point into sedding...
+ */
+ if (
+ /* no range necessary */
+ (sed_cmds[i].beg_line == 0 && sed_cmds[i].end_line == 0 &&
+ sed_cmds[i].beg_match == NULL &&
+ sed_cmds[i].end_match == NULL) ||
+ /* this line number is the first address we're looking for */
+ (sed_cmds[i].beg_line && (sed_cmds[i].beg_line == linenum)) ||
+ /* this line matches our first address regex */
+ (sed_cmds[i].beg_match && (regexec(sed_cmds[i].beg_match, line, 0, NULL, 0) == 0)) ||
+ /* we are currently within the beginning & ending address range */
+ still_in_range
+ ) {
+
+ /*
+ * actual sedding
+ */
+ switch (sed_cmds[i].cmd) {
+
+ case 'p':
+ puts(line);
+ break;
+
+ case 'd':
+ altered++;
+ break;
+
+ case 's':
+
+ /*
+ * Some special cases for 's' printing to make it compliant with
+ * GNU sed printing behavior (aka "The -n | s///p Matrix"):
+ *
+ * -n ONLY = never print anything regardless of any successful
+ * substitution
+ *
+ * s///p ONLY = always print successful substitutions, even if
+ * the line is going to be printed anyway (line will be printed
+ * twice).
+ *
+ * -n AND s///p = print ONLY a successful substitution ONE TIME;
+ * no other lines are printed - this is the reason why the 'p'
+ * flag exists in the first place.
+ */
+
+ /* if the user specified that they didn't want anything printed (i.e., a -n
+ * flag and no 'p' flag after the s///), then there's really no point doing
+ * anything here. */
+ if (be_quiet && !sed_cmds[i].sub_p)
+ break;
+
+ /* we print the line once, unless we were told to be quiet */
+ if (!be_quiet)
+ altered = do_subst_command(&sed_cmds[i], line);
+
+ /* we also print the line if we were given the 'p' flag
+ * (this is quite possibly the second printing) */
+ if (sed_cmds[i].sub_p)
+ altered = do_subst_command(&sed_cmds[i], line);
+
+ break;
+
+ case 'a':
+ puts(line);
+ fputs(sed_cmds[i].editline, stdout);
+ altered++;
+ break;
+
+ case 'i':
+ fputs(sed_cmds[i].editline, stdout);
+ break;
+
+ case 'c':
+ /* single-address case */
+ if (sed_cmds[i].end_match == NULL && sed_cmds[i].end_line == 0) {
+ fputs(sed_cmds[i].editline, stdout);
+ }
+ /* multi-address case */
+ else {
+ /* matching text */
+ if (sed_cmds[i].end_match && (regexec(sed_cmds[i].end_match, line, 0, NULL, 0) == 0))
+ fputs(sed_cmds[i].editline, stdout);
+ /* matching line numbers */
+ if (sed_cmds[i].end_line > 0 && sed_cmds[i].end_line == linenum)
+ fputs(sed_cmds[i].editline, stdout);
+ }
+ altered++;
+
+ break;
+
+ case 'r': {
+ FILE *outfile;
+ puts(line);
+ outfile = fopen(sed_cmds[i].filename, "r");
+ if (outfile)
+ print_file(outfile);
+ /* else if we couldn't open the output file,
+ * no biggie, just don't print anything */
+ altered++;
+ }
+ break;
+ }
+
+ /*
+ * exit point from sedding...
+ */
+ if (
+ /* this is a single-address command or... */
+ (sed_cmds[i].end_line == 0 && sed_cmds[i].end_match == NULL) || (
+ /* we were in the middle of our address range (this
+ * isn't the first time through) and.. */
+ (still_in_range == 1) && (
+ /* this line number is the last address we're looking for or... */
+ (sed_cmds[i].end_line && (sed_cmds[i].end_line == linenum)) ||
+ /* this line matches our last address regex */
+ (sed_cmds[i].end_match && (regexec(sed_cmds[i].end_match, line, 0, NULL, 0) == 0))
+ )
+ )
+ ) {
+ /* we're out of our address range */
+ still_in_range = 0;
+ }
+
+ /* didn't hit the exit? then we're still in the middle of an address range */
+ else {
+ still_in_range = 1;
+ }