summaryrefslogtreecommitdiffstats
path: root/contrib/ncurses
diff options
context:
space:
mode:
Diffstat (limited to 'contrib/ncurses')
-rw-r--r--contrib/ncurses/ncurses/tinfo/comp_scan.c908
-rw-r--r--contrib/ncurses/ncurses/tinfo/lib_termcap.c148
2 files changed, 545 insertions, 511 deletions
diff --git a/contrib/ncurses/ncurses/tinfo/comp_scan.c b/contrib/ncurses/ncurses/tinfo/comp_scan.c
index 39cf2cd..617ecc1 100644
--- a/contrib/ncurses/ncurses/tinfo/comp_scan.c
+++ b/contrib/ncurses/ncurses/tinfo/comp_scan.c
@@ -1,5 +1,5 @@
/****************************************************************************
- * Copyright (c) 1998 Free Software Foundation, Inc. *
+ * Copyright (c) 1998,1999,2000 Free Software Foundation, Inc. *
* *
* Permission is hereby granted, free of charge, to any person obtaining a *
* copy of this software and associated documentation files (the *
@@ -31,6 +31,8 @@
* and: Eric S. Raymond <esr@snark.thyrsus.com> *
****************************************************************************/
+/* $FreeBSD$ */
+
/*
* comp_scan.c --- Lexical scanner for terminfo compiler.
*
@@ -47,9 +49,10 @@
#include <curses.priv.h>
#include <ctype.h>
+#include <term_entry.h>
#include <tic.h>
-MODULE_ID("$Id: comp_scan.c,v 1.34 1998/11/01 00:56:39 tom Exp $")
+MODULE_ID("$Id: comp_scan.c,v 1.44 2000/06/10 21:59:21 tom Exp $")
/*
* Maximum length of string capability we'll accept before raising an error.
@@ -59,11 +62,14 @@ MODULE_ID("$Id: comp_scan.c,v 1.34 1998/11/01 00:56:39 tom Exp $")
#define iswhite(ch) (ch == ' ' || ch == '\t')
-int _nc_syntax; /* termcap or terminfo? */
-long _nc_curr_file_pos; /* file offset of current line */
-long _nc_comment_start; /* start of comment range before name */
-long _nc_comment_end; /* end of comment range before name */
-long _nc_start_line; /* start line of current entry */
+int _nc_syntax = 0; /* termcap or terminfo? */
+long _nc_curr_file_pos = 0; /* file offset of current line */
+long _nc_comment_start = 0; /* start of comment range before name */
+long _nc_comment_end = 0; /* end of comment range before name */
+long _nc_start_line = 0; /* start line of current entry */
+
+struct token _nc_curr_token =
+{0, 0, 0};
/*****************************************************************************
*
@@ -74,21 +80,22 @@ long _nc_start_line; /* start line of current entry */
static bool first_column; /* See 'next_char()' below */
static char separator; /* capability separator */
static int pushtype; /* type of pushback token */
-static char pushname[MAX_NAME_SIZE+1];
+static char pushname[MAX_NAME_SIZE + 1];
-static int last_char(void);
-static int next_char(void);
+static int last_char(void);
+static int next_char(void);
static long stream_pos(void);
static bool end_of_stream(void);
static void push_back(char c);
/* Assume we may be looking at a termcap-style continuation */
-static inline int eat_escaped_newline(int ch)
+static inline int
+eat_escaped_newline(int ch)
{
- if (ch == '\\')
- while ((ch = next_char()) == '\n' || iswhite(ch))
- continue;
- return ch;
+ if (ch == '\\')
+ while ((ch = next_char()) == '\n' || iswhite(ch))
+ continue;
+ return ch;
}
/*
@@ -124,318 +131,308 @@ static inline int eat_escaped_newline(int ch)
*
*/
-int _nc_get_token(void)
+int
+_nc_get_token(void)
{
-static const char terminfo_punct[] = "@%&*!#";
-long number;
-int type;
-int ch;
-char * numchk;
-char numbuf[80];
-unsigned found;
-static char buffer[MAX_ENTRY_SIZE];
-char *ptr;
-int dot_flag = FALSE;
-long token_start;
-
- if (pushtype != NO_PUSHBACK)
- {
- int retval = pushtype;
-
- _nc_set_type(pushname);
- DEBUG(3, ("pushed-back token: `%s', class %d",
- _nc_curr_token.tk_name, pushtype));
-
- pushtype = NO_PUSHBACK;
- pushname[0] = '\0';
-
- /* currtok wasn't altered by _nc_push_token() */
- return(retval);
- }
+ static const char terminfo_punct[] = "@%&*!#";
+ long number;
+ int type;
+ int ch;
+ char *numchk;
+ char numbuf[80];
+ unsigned found;
+ static char buffer[MAX_ENTRY_SIZE];
+ char *ptr;
+ int dot_flag = FALSE;
+ long token_start;
+
+ if (pushtype != NO_PUSHBACK) {
+ int retval = pushtype;
+
+ _nc_set_type(pushname);
+ DEBUG(3, ("pushed-back token: `%s', class %d",
+ _nc_curr_token.tk_name, pushtype));
- if (end_of_stream())
- return(EOF);
+ pushtype = NO_PUSHBACK;
+ pushname[0] = '\0';
-start_token:
- token_start = stream_pos();
- while ((ch = next_char()) == '\n' || iswhite(ch))
- continue;
+ /* currtok wasn't altered by _nc_push_token() */
+ return (retval);
+ }
- ch = eat_escaped_newline(ch);
+ if (end_of_stream())
+ return (EOF);
- if (ch == EOF)
- type = EOF;
- else {
- /* if this is a termcap entry, skip a leading separator */
- if (separator == ':' && ch == ':')
- ch = next_char();
+ start_token:
+ token_start = stream_pos();
+ while ((ch = next_char()) == '\n' || iswhite(ch))
+ continue;
- if (ch == '.') {
- dot_flag = TRUE;
- DEBUG(8, ("dot-flag set"));
+ ch = eat_escaped_newline(ch);
- while ((ch = next_char())=='.' || iswhite(ch))
- continue;
- }
+ if (ch == EOF)
+ type = EOF;
+ else {
+ /* if this is a termcap entry, skip a leading separator */
+ if (separator == ':' && ch == ':')
+ ch = next_char();
- if (ch == EOF) {
- type = EOF;
- goto end_of_token;
- }
+ if (ch == '.'
+#ifdef NCURSES_EXT_FUNCS
+ && !_nc_disable_period
+#endif
+ ) {
+ dot_flag = TRUE;
+ DEBUG(8, ("dot-flag set"));
- /* have to make some punctuation chars legal for terminfo */
- if (!isalnum(ch) && !strchr(terminfo_punct, (char)ch)) {
- _nc_warning("Illegal character (expected alphanumeric or %s) - %s",
- terminfo_punct, _tracechar((chtype)ch));
- _nc_panic_mode(separator);
- goto start_token;
- }
+ while ((ch = next_char()) == '.' || iswhite(ch))
+ continue;
+ }
- ptr = buffer;
- *(ptr++) = ch;
-
- if (first_column) {
- char *desc;
-
- _nc_comment_start = token_start;
- _nc_comment_end = _nc_curr_file_pos;
- _nc_start_line = _nc_curr_line;
-
- _nc_syntax = ERR;
- while ((ch = next_char()) != '\n')
- {
- if (ch == EOF)
- _nc_err_abort("premature EOF");
- else if (ch == ':' && last_char() != ',')
- {
- _nc_syntax = SYN_TERMCAP;
- separator = ':';
- break;
- }
- else if (ch == ',')
- {
- _nc_syntax = SYN_TERMINFO;
- separator = ',';
- /*
- * Fall-through here is not an accident.
- * The idea is that if we see a comma, we
- * figure this is terminfo unless we
- * subsequently run into a colon -- but
- * we don't stop looking for that colon until
- * hitting a newline. This allows commas to
- * be embedded in description fields of
- * either syntax.
- */
- /* FALLTHRU */
- }
- else
- ch = eat_escaped_newline(ch);
-
- *ptr++ = ch;
- }
- ptr[0] = '\0';
- if (_nc_syntax == ERR)
- {
- /*
- * Grrr...what we ought to do here is barf,
- * complaining that the entry is malformed.
- * But because a couple of name fields in the
- * 8.2 termcap file end with |\, we just have
- * to assume it's termcap syntax.
- */
- _nc_syntax = SYN_TERMCAP;
- separator = ':';
- }
- else if (_nc_syntax == SYN_TERMINFO)
- {
- /* throw away trailing /, *$/ */
- for (--ptr; iswhite(*ptr) || *ptr == ','; ptr--)
- continue;
- ptr[1] = '\0';
- }
+ if (ch == EOF) {
+ type = EOF;
+ goto end_of_token;
+ }
- /*
- * This is the soonest we have the terminal name
- * fetched. Set up for following warning messages.
- */
- ptr = strchr(buffer, '|');
- if (ptr == (char *)NULL)
- ptr = buffer + strlen(buffer);
- ch = *ptr;
- *ptr = '\0';
- _nc_set_type(buffer);
- *ptr = ch;
-
- /*
- * Compute the boundary between the aliases and the
- * description field for syntax-checking purposes.
- */
- desc = strrchr(buffer, '|');
- if (desc) {
- if (*desc == '\0')
- _nc_warning("empty longname field");
-#ifndef FREEBSD_NATIVE
- else if (strchr(desc, ' ') == (char *)NULL)
- _nc_warning("older tic versions may treat the description field as an alias");
+ /* have to make some punctuation chars legal for terminfo */
+ if (!isalnum(ch)
+#ifdef NCURSES_EXT_FUNCS
+ && !(ch == '.' && _nc_disable_period)
#endif
- }
- if (!desc)
- desc = buffer + strlen(buffer);
-
- /*
- * Whitespace in a name field other than the long name
- * can confuse rdist and some termcap tools. Slashes
- * are a no-no. Other special characters can be
- * dangerous due to shell expansion.
- */
- for (ptr = buffer; ptr < desc; ptr++)
- {
- if (isspace(*ptr))
- {
- _nc_warning("whitespace in name or alias field");
- break;
- }
- else if (*ptr == '/')
- {
- _nc_warning("slashes aren't allowed in names or aliases");
- break;
- }
- else if (strchr("$[]!*?", *ptr))
- {
- _nc_warning("dubious character `%c' in name or alias field", *ptr);
- break;
- }
- }
-
- ptr = buffer;
+ && !strchr(terminfo_punct, (char) ch)) {
+ _nc_warning("Illegal character (expected alphanumeric or %s) - %s",
+ terminfo_punct, unctrl(ch));
+ _nc_panic_mode(separator);
+ goto start_token;
+ }
- _nc_curr_token.tk_name = buffer;
- type = NAMES;
- } else {
- while ((ch = next_char()) != EOF) {
- if (!isalnum(ch)) {
- if (_nc_syntax == SYN_TERMINFO) {
- if (ch != '_')
- break;
- } else { /* allow ';' for "k;" */
- if (ch != ';')
- break;
- }
- }
- *(ptr++) = ch;
- }
+ ptr = buffer;
+ *(ptr++) = ch;
- *ptr++ = '\0';
- switch (ch) {
- case ',':
- case ':':
- if (ch != separator)
- _nc_err_abort("Separator inconsistent with syntax");
- _nc_curr_token.tk_name = buffer;
- type = BOOLEAN;
- break;
- case '@':
- if ((ch = next_char()) != separator)
- _nc_warning("Missing separator after `%s', have %s",
- buffer, _tracechar((chtype)ch));
- _nc_curr_token.tk_name = buffer;
- type = CANCEL;
- break;
-
- case '#':
- found = 0;
- while (isalnum(ch = next_char())) {
- numbuf[found++] = ch;
- if (found >= sizeof(numbuf)-1)
- break;
- }
- numbuf[found] = '\0';
- number = strtol(numbuf, &numchk, 0);
- if (numchk == numbuf)
- _nc_warning("no value given for `%s'", buffer);
- if ((*numchk != '\0') || (ch != separator))
- _nc_warning("Missing separator");
- _nc_curr_token.tk_name = buffer;
- _nc_curr_token.tk_valnumber = number;
- type = NUMBER;
- break;
-
- case '=':
- ch = _nc_trans_string(ptr);
- if (ch != separator)
- _nc_warning("Missing separator");
- _nc_curr_token.tk_name = buffer;
- _nc_curr_token.tk_valstring = ptr;
- type = STRING;
- break;
-
- case EOF:
- type = EOF;
- break;
- default:
- /* just to get rid of the compiler warning */
- type = UNDEF;
- _nc_warning("Illegal character - %s",
- _tracechar((chtype)ch));
- }
- } /* end else (first_column == FALSE) */
- } /* end else (ch != EOF) */
+ if (first_column) {
+ char *desc;
-end_of_token:
+ _nc_comment_start = token_start;
+ _nc_comment_end = _nc_curr_file_pos;
+ _nc_start_line = _nc_curr_line;
-#ifdef TRACE
- if (dot_flag == TRUE)
- DEBUG(8, ("Commented out "));
-
- if (_nc_tracing & TRACE_IEVENT)
- {
- fprintf(stderr, "Token: ");
- switch (type)
- {
- case BOOLEAN:
- fprintf(stderr, "Boolean; name='%s'\n",
- _nc_curr_token.tk_name);
+ _nc_syntax = ERR;
+ while ((ch = next_char()) != '\n') {
+ if (ch == EOF)
+ _nc_err_abort("premature EOF");
+ else if (ch == ':' && last_char() != ',') {
+ _nc_syntax = SYN_TERMCAP;
+ separator = ':';
break;
+ } else if (ch == ',') {
+ _nc_syntax = SYN_TERMINFO;
+ separator = ',';
+ /*
+ * Fall-through here is not an accident.
+ * The idea is that if we see a comma, we
+ * figure this is terminfo unless we
+ * subsequently run into a colon -- but
+ * we don't stop looking for that colon until
+ * hitting a newline. This allows commas to
+ * be embedded in description fields of
+ * either syntax.
+ */
+ /* FALLTHRU */
+ } else
+ ch = eat_escaped_newline(ch);
+
+ *ptr++ = ch;
+ }
+ ptr[0] = '\0';
+ if (_nc_syntax == ERR) {
+ /*
+ * Grrr...what we ought to do here is barf,
+ * complaining that the entry is malformed.
+ * But because a couple of name fields in the
+ * 8.2 termcap file end with |\, we just have
+ * to assume it's termcap syntax.
+ */
+ _nc_syntax = SYN_TERMCAP;
+ separator = ':';
+ } else if (_nc_syntax == SYN_TERMINFO) {
+ /* throw away trailing /, *$/ */
+ for (--ptr; iswhite(*ptr) || *ptr == ','; ptr--)
+ continue;
+ ptr[1] = '\0';
+ }
- case NUMBER:
- fprintf(stderr, "Number; name='%s', value=%d\n",
- _nc_curr_token.tk_name,
- _nc_curr_token.tk_valnumber);
+ /*
+ * This is the soonest we have the terminal name
+ * fetched. Set up for following warning messages.
+ */
+ ptr = strchr(buffer, '|');
+ if (ptr == (char *) NULL)
+ ptr = buffer + strlen(buffer);
+ ch = *ptr;
+ *ptr = '\0';
+ _nc_set_type(buffer);
+ *ptr = ch;
+
+ /*
+ * Compute the boundary between the aliases and the
+ * description field for syntax-checking purposes.
+ */
+ desc = strrchr(buffer, '|');
+ if (desc) {
+ if (*desc == '\0')
+ _nc_warning("empty longname field");
+#ifndef FREEBSD_NATIVE
+ else if (strchr(desc, ' ') == (char *) NULL)
+ _nc_warning("older tic versions may treat the description field as an alias");
+#endif
+ }
+ if (!desc)
+ desc = buffer + strlen(buffer);
+
+ /*
+ * Whitespace in a name field other than the long name
+ * can confuse rdist and some termcap tools. Slashes
+ * are a no-no. Other special characters can be
+ * dangerous due to shell expansion.
+ */
+ for (ptr = buffer; ptr < desc; ptr++) {
+ if (isspace(*ptr)) {
+ _nc_warning("whitespace in name or alias field");
break;
-
- case STRING:
- fprintf(stderr, "String; name='%s', value=%s\n",
- _nc_curr_token.tk_name,
- _nc_visbuf(_nc_curr_token.tk_valstring));
+ } else if (*ptr == '/') {
+ _nc_warning("slashes aren't allowed in names or aliases");
break;
-
- case CANCEL:
- fprintf(stderr, "Cancel; name='%s'\n",
- _nc_curr_token.tk_name);
+ } else if (strchr("$[]!*?", *ptr)) {
+ _nc_warning("dubious character `%c' in name or alias field", *ptr);
break;
+ }
+ }
- case NAMES:
+ ptr = buffer;
- fprintf(stderr, "Names; value='%s'\n",
- _nc_curr_token.tk_name);
- break;
+ _nc_curr_token.tk_name = buffer;
+ type = NAMES;
+ } else {
+ while ((ch = next_char()) != EOF) {
+ if (!isalnum(ch)) {
+ if (_nc_syntax == SYN_TERMINFO) {
+ if (ch != '_')
+ break;
+ } else { /* allow ';' for "k;" */
+ if (ch != ';')
+ break;
+ }
+ }
+ *(ptr++) = ch;
+ }
- case EOF:
- fprintf(stderr, "End of file\n");
- break;
+ *ptr++ = '\0';
+ switch (ch) {
+ case ',':
+ case ':':
+ if (ch != separator)
+ _nc_err_abort("Separator inconsistent with syntax");
+ _nc_curr_token.tk_name = buffer;
+ type = BOOLEAN;
+ break;
+ case '@':
+ if ((ch = next_char()) != separator)
+ _nc_warning("Missing separator after `%s', have %s",
+ buffer, unctrl(ch));
+ _nc_curr_token.tk_name = buffer;
+ type = CANCEL;
+ break;
- default:
- _nc_warning("Bad token type");
+ case '#':
+ found = 0;
+ while (isalnum(ch = next_char())) {
+ numbuf[found++] = ch;
+ if (found >= sizeof(numbuf) - 1)
+ break;
+ }
+ numbuf[found] = '\0';
+ number = strtol(numbuf, &numchk, 0);
+ if (numchk == numbuf)
+ _nc_warning("no value given for `%s'", buffer);
+ if ((*numchk != '\0') || (ch != separator))
+ _nc_warning("Missing separator");
+ _nc_curr_token.tk_name = buffer;
+ _nc_curr_token.tk_valnumber = number;
+ type = NUMBER;
+ break;
+
+ case '=':
+ ch = _nc_trans_string(ptr, buffer + sizeof(buffer));
+ if (ch != separator)
+ _nc_warning("Missing separator");
+ _nc_curr_token.tk_name = buffer;
+ _nc_curr_token.tk_valstring = ptr;
+ type = STRING;
+ break;
+
+ case EOF:
+ type = EOF;
+ break;
+ default:
+ /* just to get rid of the compiler warning */
+ type = UNDEF;
+ _nc_warning("Illegal character - %s", unctrl(ch));
}
+ } /* end else (first_column == FALSE) */
+ } /* end else (ch != EOF) */
+
+ end_of_token:
+
+#ifdef TRACE
+ if (dot_flag == TRUE)
+ DEBUG(8, ("Commented out "));
+
+ if (_nc_tracing >= DEBUG_LEVEL(7)) {
+ switch (type) {
+ case BOOLEAN:
+ _tracef("Token: Boolean; name='%s'",
+ _nc_curr_token.tk_name);
+ break;
+
+ case NUMBER:
+ _tracef("Token: Number; name='%s', value=%d",
+ _nc_curr_token.tk_name,
+ _nc_curr_token.tk_valnumber);
+ break;
+
+ case STRING:
+ _tracef("Token: String; name='%s', value=%s",
+ _nc_curr_token.tk_name,
+ _nc_visbuf(_nc_curr_token.tk_valstring));
+ break;
+
+ case CANCEL:
+ _tracef("Token: Cancel; name='%s'",
+ _nc_curr_token.tk_name);
+ break;
+
+ case NAMES:
+
+ _tracef("Token: Names; value='%s'",
+ _nc_curr_token.tk_name);
+ break;
+
+ case EOF:
+ _tracef("Token: End of file");
+ break;
+
+ default:
+ _nc_warning("Bad token type");
}
+ }
#endif
- if (dot_flag == TRUE) /* if commented out, use the next one */
- type = _nc_get_token();
+ if (dot_flag == TRUE) /* if commented out, use the next one */
+ type = _nc_get_token();
- DEBUG(3, ("token: `%s', class %d", _nc_curr_token.tk_name, type));
+ DEBUG(3, ("token: `%s', class %d", _nc_curr_token.tk_name, type));
- return(type);
+ return (type);
}
/*
@@ -458,120 +455,150 @@ end_of_token:
*/
char
-_nc_trans_string(char *ptr)
+_nc_trans_string(char *ptr, char *last)
{
-int count = 0;
-int number;
-int i, c;
-chtype ch, last_ch = '\0';
-bool ignored = FALSE;
-
- while ((ch = c = next_char()) != (chtype)separator && c != EOF) {
- if ((_nc_syntax == SYN_TERMCAP) && c == '\n')
- break;
- if (ch == '^' && last_ch != '%') {
- ch = c = next_char();
- if (c == EOF)
- _nc_err_abort("Premature EOF");
-
- if (! (is7bits(ch) && isprint(ch))) {
- _nc_warning("Illegal ^ character - %s",
- _tracechar((unsigned char)ch));
- }
- if (ch == '?') {
- *(ptr++) = '\177';
- } else {
- if ((ch &= 037) == 0)
- ch = 128;
- *(ptr++) = (char)(ch);
- }
+ int count = 0;
+ int number;
+ int i, c;
+ chtype ch, last_ch = '\0';
+ bool ignored = FALSE;
+ bool long_warning = FALSE;
+
+ while ((ch = c = next_char()) != (chtype) separator && c != EOF) {
+ if (ptr == (last - 1))
+ break;
+ if ((_nc_syntax == SYN_TERMCAP) && c == '\n')
+ break;
+ if (ch == '^' && last_ch != '%') {
+ ch = c = next_char();
+ if (c == EOF)
+ _nc_err_abort("Premature EOF");
+
+ if (!(is7bits(ch) && isprint(ch))) {
+ _nc_warning("Illegal ^ character - %s", unctrl(ch));
+ }
+ if (ch == '?') {
+ *(ptr++) = '\177';
+ if (_nc_tracing)
+ _nc_warning("Allow ^? as synonym for \\177");
+ } else {
+ if ((ch &= 037) == 0)
+ ch = 128;
+ *(ptr++) = (char) (ch);
}
- else if (ch == '\\') {
- ch = c = next_char();
- if (c == EOF)
- _nc_err_abort("Premature EOF");
-
- if (ch >= '0' && ch <= '7') {
- number = ch - '0';
- for (i=0; i < 2; i++) {
- ch = c = next_char();
- if (c == EOF)
- _nc_err_abort("Premature EOF");
-
- if (c < '0' || c > '7') {
- if (isdigit(c)) {
- _nc_warning("Non-octal digit `%c' in \\ sequence", c);
- /* allow the digit; it'll do less harm */
- } else {
- push_back((char)c);
- break;
- }
+ } else if (ch == '\\') {
+ ch = c = next_char();
+ if (c == EOF)
+ _nc_err_abort("Premature EOF");
+
+ if (ch >= '0' && ch <= '7') {
+ number = ch - '0';
+ for (i = 0; i < 2; i++) {
+ ch = c = next_char();
+ if (c == EOF)
+ _nc_err_abort("Premature EOF");
+
+ if (c < '0' || c > '7') {
+ if (isdigit(c)) {
+ _nc_warning("Non-octal digit `%c' in \\ sequence", c);
+ /* allow the digit; it'll do less harm */
+ } else {
+ push_back((char) c);
+ break;
}
-
- number = number * 8 + c - '0';
}
- if (number == 0)
- number = 0200;
- *(ptr++) = (char) number;
- } else {
- switch (c) {
- case 'E':
- case 'e': *(ptr++) = '\033'; break;
+ number = number * 8 + c - '0';
+ }
+
+ if (number == 0)
+ number = 0200;
+ *(ptr++) = (char) number;
+ } else {
+ switch (c) {
+ case 'E':
+ case 'e':
+ *(ptr++) = '\033';
+ break;
- case 'a': *(ptr++) = '\007'; break;
+ case 'a':
+ *(ptr++) = '\007';
+ break;
- case 'l':
- case 'n': *(ptr++) = '\n'; break;
+ case 'l':
+ case 'n':
+ *(ptr++) = '\n';
+ break;
- case 'r': *(ptr++) = '\r'; break;
+ case 'r':
+ *(ptr++) = '\r';
+ break;
- case 'b': *(ptr++) = '\010'; break;
+ case 'b':
+ *(ptr++) = '\010';
+ break;
- case 's': *(ptr++) = ' '; break;
+ case 's':
+ *(ptr++) = ' ';
+ break;
- case 'f': *(ptr++) = '\014'; break;
+ case 'f':
+ *(ptr++) = '\014';
+ break;
- case 't': *(ptr++) = '\t'; break;
+ case 't':
+ *(ptr++) = '\t';
+ break;
- case '\\': *(ptr++) = '\\'; break;
+ case '\\':
+ *(ptr++) = '\\';
+ break;
- case '^': *(ptr++) = '^'; break;
+ case '^':
+ *(ptr++) = '^';
+ break;
- case ',': *(ptr++) = ','; break;
+ case ',':
+ *(ptr++) = ',';
+ break;
- case ':': *(ptr++) = ':'; break;
+ case ':':
+ *(ptr++) = ':';
+ break;
- case '\n':
- continue;
+ case '\n':
+ continue;
- default:
- _nc_warning("Illegal character %s in \\ sequence",
- _tracechar((unsigned char)ch));
- *(ptr++) = (char)ch;
- } /* endswitch (ch) */
- } /* endelse (ch < '0' || ch > '7') */
- } /* end else if (ch == '\\') */
- else if (ch == '\n' && (_nc_syntax == SYN_TERMINFO)) {
- /* newlines embedded in a terminfo string are ignored */
- ignored = TRUE;
- } else {
- *(ptr++) = (char)ch;
- }
+ default:
+ _nc_warning("Illegal character %s in \\ sequence",
+ unctrl(ch));
+ *(ptr++) = (char) ch;
+ } /* endswitch (ch) */
+ } /* endelse (ch < '0' || ch > '7') */
+ }
+ /* end else if (ch == '\\') */
+ else if (ch == '\n' && (_nc_syntax == SYN_TERMINFO)) {
+ /* newlines embedded in a terminfo string are ignored */
+ ignored = TRUE;
+ } else {
+ *(ptr++) = (char) ch;
+ }
- if (!ignored) {
- last_ch = ch;
- count ++;
- }
- ignored = FALSE;
+ if (!ignored) {
+ last_ch = ch;
+ count++;
+ }
+ ignored = FALSE;
- if (count > MAXCAPLEN)
- _nc_warning("Very long string found. Missing separator?");
- } /* end while */
+ if (count > MAXCAPLEN && !long_warning) {
+ _nc_warning("Very long string found. Missing separator?");
+ long_warning = TRUE;
+ }
+ } /* end while */
- *ptr = '\0';
+ *ptr = '\0';
- return(ch);
+ return (ch);
}
/*
@@ -581,7 +608,8 @@ bool ignored = FALSE;
* get_token() call.
*/
-void _nc_push_token(int tokclass)
+void
+_nc_push_token(int tokclass)
{
/*
* This implementation is kind of bogus, it will fail if we ever do
@@ -593,23 +621,24 @@ void _nc_push_token(int tokclass)
_nc_get_type(pushname);
DEBUG(3, ("pushing token: `%s', class %d",
- _nc_curr_token.tk_name, pushtype));
+ _nc_curr_token.tk_name, pushtype));
}
/*
* Panic mode error recovery - skip everything until a "ch" is found.
*/
-void _nc_panic_mode(char ch)
+void
+_nc_panic_mode(char ch)
{
- int c;
-
- for (;;) {
- c = next_char();
- if (c == ch)
- return;
- if (c == EOF)
- return;
- }
+ int c;
+
+ for (;;) {
+ c = next_char();
+ if (c == ch)
+ return;
+ if (c == EOF)
+ return;
+ }
}
/*****************************************************************************
@@ -632,16 +661,17 @@ static FILE *yyin; /* scanner's input file descriptor */
* non-null.
*/
-void _nc_reset_input(FILE *fp, char *buf)
+void
+_nc_reset_input(FILE * fp, char *buf)
{
- pushtype = NO_PUSHBACK;
- pushname[0] = '\0';
- yyin = fp;
- bufstart = bufptr = buf;
- _nc_curr_file_pos = 0L;
- if (fp != 0)
- _nc_curr_line = 0;
- _nc_curr_col = 0;
+ pushtype = NO_PUSHBACK;
+ pushname[0] = '\0';
+ yyin = fp;
+ bufstart = bufptr = buf;
+ _nc_curr_file_pos = 0L;
+ if (fp != 0)
+ _nc_curr_line = 0;
+ _nc_curr_col = 0;
}
/*
@@ -652,12 +682,12 @@ void _nc_reset_input(FILE *fp, char *buf)
static int
last_char(void)
{
- size_t len = strlen(bufptr);
- while (len--) {
- if (!isspace(bufptr[len]))
- return bufptr[len];
- }
- return 0;
+ size_t len = strlen(bufptr);
+ while (len--) {
+ if (!isspace(bufptr[len]))
+ return bufptr[len];
+ }
+ return 0;
}
/*
@@ -677,17 +707,14 @@ last_char(void)
static int
next_char(void)
{
- if (!yyin)
- {
+ if (!yyin) {
if (*bufptr == '\0')
- return(EOF);
+ return (EOF);
if (*bufptr == '\n') {
_nc_curr_line++;
_nc_curr_col = 0;
}
- }
- else if (!bufptr || !*bufptr)
- {
+ } else if (!bufptr || !*bufptr) {
/*
* In theory this could be recoded to do its I/O one
* character at a time, saving the buffer space. In
@@ -699,15 +726,15 @@ next_char(void)
size_t len;
do {
- _nc_curr_file_pos = ftell(yyin);
+ _nc_curr_file_pos = ftell(yyin);
- if ((bufstart = fgets(line, LEXBUFSIZ, yyin)) != NULL) {
- _nc_curr_line++;
- _nc_curr_col = 0;
- }
- bufptr = bufstart;
- } while
- (bufstart != NULL && line[0] == '#');
+ if ((bufstart = fgets(line, LEXBUFSIZ, yyin)) != NULL) {
+ _nc_curr_line++;
+ _nc_curr_col = 0;
+ }
+ bufptr = bufstart;
+ } while
+ (bufstart != NULL && line[0] == '#');
if (bufstart == NULL)
return (EOF);
@@ -720,10 +747,10 @@ next_char(void)
* files on OS/2, etc.
*/
if ((len = strlen(bufptr)) > 1) {
- if (bufptr[len-1] == '\n'
- && bufptr[len-2] == '\r') {
- bufptr[len-2] = '\n';
- bufptr[len-1] = '\0';
+ if (bufptr[len - 1] == '\n'
+ && bufptr[len - 2] == '\r') {
+ bufptr[len - 2] = '\n';
+ bufptr[len - 1] = '\0';
}
}
}
@@ -731,28 +758,31 @@ next_char(void)
first_column = (bufptr == bufstart);
_nc_curr_col++;
- return(*bufptr++);
+ return (*bufptr++);
}
-static void push_back(char c)
+static void
+push_back(char c)
/* push a character back onto the input stream */
{
if (bufptr == bufstart)
- _nc_syserr_abort("Can't backspace off beginning of line");
+ _nc_syserr_abort("Can't backspace off beginning of line");
*--bufptr = c;
}
-static long stream_pos(void)
+static long
+stream_pos(void)
/* return our current character position in the input stream */
{
return (yyin ? ftell(yyin) : (bufptr ? bufptr - bufstart : 0));
}
-static bool end_of_stream(void)
+static bool
+end_of_stream(void)
/* are we at end of input? */
{
return ((yyin ? feof(yyin) : (bufptr && *bufptr == '\0'))
- ? TRUE : FALSE);
+ ? TRUE : FALSE);
}
/* comp_scan.c ends here */
diff --git a/contrib/ncurses/ncurses/tinfo/lib_termcap.c b/contrib/ncurses/ncurses/tinfo/lib_termcap.c
index b34ced5..fdfce05 100644
--- a/contrib/ncurses/ncurses/tinfo/lib_termcap.c
+++ b/contrib/ncurses/ncurses/tinfo/lib_termcap.c
@@ -1,5 +1,5 @@
/****************************************************************************
- * Copyright (c) 1998 Free Software Foundation, Inc. *
+ * Copyright (c) 1998,1999,2000 Free Software Foundation, Inc. *
* *
* Permission is hereby granted, free of charge, to any person obtaining a *
* copy of this software and associated documentation files (the *
@@ -41,15 +41,15 @@
#define __INTERNAL_CAPS_VISIBLE
#include <term_entry.h>
-MODULE_ID("$Id: lib_termcap.c,v 1.29 1999/09/05 01:06:43 tom Exp $")
+MODULE_ID("$Id: lib_termcap.c,v 1.36 2000/02/13 01:01:26 tom Exp $")
/*
some of the code in here was contributed by:
Magnus Bengtsson, d6mbeng@dtek.chalmers.se
*/
-char *UP;
-char *BC;
+char *UP = 0;
+char *BC = 0;
#ifdef FREEBSD_NATIVE
#undef GCC_UNUSED
@@ -72,29 +72,30 @@ extern char _nc_termcap[]; /* buffer to copy out */
*
***************************************************************************/
-int tgetent(char *bufp GCC_UNUSED, const char *name)
+int
+tgetent(char *bufp GCC_UNUSED, const char *name)
{
-int errcode;
+ int errcode;
- T((T_CALLED("tgetent()")));
+ T((T_CALLED("tgetent()")));
- setupterm((NCURSES_CONST char *)name, STDOUT_FILENO, &errcode);
+ setupterm((NCURSES_CONST char *) name, STDOUT_FILENO, &errcode);
- if (errcode == 1) {
+ if (errcode == 1) {
- if (cursor_left)
- if ((backspaces_with_bs = !strcmp(cursor_left, "\b")) == 0)
- backspace_if_not_bs = cursor_left;
+ if (cursor_left)
+ if ((backspaces_with_bs = !strcmp(cursor_left, "\b")) == 0)
+ backspace_if_not_bs = cursor_left;
- /* we're required to export these */
- if (pad_char != NULL)
- PC = pad_char[0];
- if (cursor_up != NULL)
- UP = cursor_up;
- if (backspace_if_not_bs != NULL)
- BC = backspace_if_not_bs;
+ /* we're required to export these */
+ if (pad_char != NULL)
+ PC = pad_char[0];
+ if (cursor_up != NULL)
+ UP = cursor_up;
+ if (backspace_if_not_bs != NULL)
+ BC = backspace_if_not_bs;
- (void) baudrate(); /* sets ospeed as a side-effect */
+ (void) baudrate(); /* sets ospeed as a side-effect */
/* LINT_PREPRO
#if 0*/
@@ -102,7 +103,6 @@ int errcode;
/* LINT_PREPRO
#endif*/
- }
#ifdef FREEBSD_NATIVE
/*
* This is a REALLY UGLY hack. Basically, if we originate with
@@ -112,7 +112,8 @@ int errcode;
strncpy(bufp, _nc_termcap, 1024);
#endif
- returnCode(errcode);
+ }
+ returnCode(errcode);
}
/***************************************************************************
@@ -124,22 +125,23 @@ int errcode;
*
***************************************************************************/
-int tgetflag(NCURSES_CONST char *id)
+int
+tgetflag(NCURSES_CONST char *id)
{
-int i;
-
- T((T_CALLED("tgetflag(%s)"), id));
- if (cur_term != 0) {
- TERMTYPE *tp = &(cur_term->type);
- for_each_boolean(i, tp) {
- const char *capname = ExtBoolname(tp, i, boolcodes);
- if (!strncmp(id, capname, 2)) {
- /* setupterm forces invalid booleans to false */
- returnCode(tp->Booleans[i]);
- }
+ int i;
+
+ T((T_CALLED("tgetflag(%s)"), id));
+ if (cur_term != 0) {
+ TERMTYPE *tp = &(cur_term->type);
+ for_each_boolean(i, tp) {
+ const char *capname = ExtBoolname(tp, i, boolcodes);
+ if (!strncmp(id, capname, 2)) {
+ /* setupterm forces invalid booleans to false */
+ returnCode(tp->Booleans[i]);
}
}
- returnCode(0); /* Solaris does this */
+ }
+ returnCode(0); /* Solaris does this */
}
/***************************************************************************
@@ -151,23 +153,24 @@ int i;
*
***************************************************************************/
-int tgetnum(NCURSES_CONST char *id)
+int
+tgetnum(NCURSES_CONST char *id)
{
-int i;
-
- T((T_CALLED("tgetnum(%s)"), id));
- if (cur_term != 0) {
- TERMTYPE *tp = &(cur_term->type);
- for_each_number(i, tp) {
- const char *capname = ExtNumname(tp, i, numcodes);
- if (!strncmp(id, capname, 2)) {
- if (!VALID_NUMERIC(tp->Numbers[i]))
- return -1;
- returnCode(tp->Numbers[i]);
- }
+ int i;
+
+ T((T_CALLED("tgetnum(%s)"), id));
+ if (cur_term != 0) {
+ TERMTYPE *tp = &(cur_term->type);
+ for_each_number(i, tp) {
+ const char *capname = ExtNumname(tp, i, numcodes);
+ if (!strncmp(id, capname, 2)) {
+ if (!VALID_NUMERIC(tp->Numbers[i]))
+ returnCode(ABSENT_NUMERIC);
+ returnCode(tp->Numbers[i]);
}
}
- returnCode(ERR);
+ }
+ returnCode(ABSENT_NUMERIC);
}
/***************************************************************************
@@ -179,30 +182,30 @@ int i;
*
***************************************************************************/
-char *tgetstr(NCURSES_CONST char *id, char **area)
+char *
+tgetstr(NCURSES_CONST char *id, char **area)
{
-int i;
-
- T((T_CALLED("tgetstr(%s,%p)"), id, area));
- if (cur_term != 0) {
- TERMTYPE *tp = &(cur_term->type);
- for_each_string(i, tp) {
- const char *capname = ExtStrname(tp, i, strcodes);
- T(("trying %s", capname));
- if (!strncmp(id, capname, 2)) {
- T(("found match : %s", _nc_visbuf(tp->Strings[i])));
- /* setupterm forces cancelled strings to null */
- if (area != 0
- && *area != 0
- && VALID_STRING(tp->Strings[i])) {
- (void) strcpy(*area, tp->Strings[i]);
- *area += strlen(*area) + 1;
- }
- returnPtr(tp->Strings[i]);
+ int i;
+
+ T((T_CALLED("tgetstr(%s,%p)"), id, area));
+ if (cur_term != 0) {
+ TERMTYPE *tp = &(cur_term->type);
+ for_each_string(i, tp) {
+ const char *capname = ExtStrname(tp, i, strcodes);
+ if (!strncmp(id, capname, 2)) {
+ TR(TRACE_DATABASE,("found match : %s", _nc_visbuf(tp->Strings[i])));
+ /* setupterm forces canceled strings to null */
+ if (area != 0
+ && *area != 0
+ && VALID_STRING(tp->Strings[i])) {
+ (void) strcpy(*area, tp->Strings[i]);
+ *area += strlen(*area) + 1;
}
+ returnPtr(tp->Strings[i]);
}
}
- returnPtr(NULL);
+ }
+ returnPtr(NULL);
}
/*
@@ -214,8 +217,9 @@ int i;
*
*/
-char *tgoto(const char *string, int x, int y)
+char *
+tgoto(const char *string, int x, int y)
{
- T((T_CALLED("tgoto(%s,%d,%d)"), string, x, y));
- returnPtr(tparm((NCURSES_CONST char *)string, y, x));
+ T((T_CALLED("tgoto(%s,%d,%d)"), string, x, y));
+ returnPtr(tparm((NCURSES_CONST char *) string, y, x));
}
OpenPOWER on IntegriCloud