freebsd-skq/usr.bin/locale/locale.c
vangyzen 27e4e37a51 locale: enable more warnings; fix them
Do not set WARNS, so it gets the current default of 6.
Fix the warnings by sprinkling static, const, or strdup.
Make some constant data tables const.  Fix whitespace.

MFC after:	1 week
Sponsored by:	Dell EMC
2016-12-01 17:36:31 +00:00

721 lines
19 KiB
C

/*-
* Copyright (c) 2002, 2003 Alexey Zelkin <phantom@FreeBSD.org>
* All rights reserved.
*
* Redistribution and use in source and binary forms, with or without
* modification, are permitted provided that the following conditions
* are met:
* 1. Redistributions of source code must retain the above copyright
* notice, this list of conditions and the following disclaimer.
* 2. Redistributions in binary form must reproduce the above copyright
* notice, this list of conditions and the following disclaimer in the
* documentation and/or other materials provided with the distribution.
*
* THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
* ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
* IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
* ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
* FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
* DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
* OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
* HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
* LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
* OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
* SUCH DAMAGE.
*
* $FreeBSD$
*/
/*
* XXX: implement missing era_* (LC_TIME) keywords (require libc &
* nl_langinfo(3) extensions)
*
* XXX: correctly handle reserved 'charmap' keyword and '-m' option (require
* localedef(1) implementation). Currently it's handled via
* nl_langinfo(CODESET).
*/
#include <sys/param.h>
#include <sys/types.h>
#include <dirent.h>
#include <err.h>
#include <limits.h>
#include <locale.h>
#include <langinfo.h>
#include <stdio.h>
#include <stdlib.h>
#include <string.h>
#include <stringlist.h>
#include <unistd.h>
#include "setlocale.h"
/* Local prototypes */
char *format_grouping(const char *);
void init_locales_list(void);
void list_charmaps(void);
void list_locales(void);
const char *lookup_localecat(int);
char *kwval_lconv(int);
int kwval_lookup(const char *, char **, int *, int *);
void showdetails(const char *);
void showkeywordslist(char *substring);
void showlocale(void);
void usage(void);
/* Global variables */
static StringList *locales = NULL;
static int all_locales = 0;
static int all_charmaps = 0;
static int prt_categories = 0;
static int prt_keywords = 0;
static const struct _lcinfo {
const char *name;
int id;
} lcinfo [] = {
{ "LC_CTYPE", LC_CTYPE },
{ "LC_COLLATE", LC_COLLATE },
{ "LC_TIME", LC_TIME },
{ "LC_NUMERIC", LC_NUMERIC },
{ "LC_MONETARY", LC_MONETARY },
{ "LC_MESSAGES", LC_MESSAGES }
};
#define NLCINFO nitems(lcinfo)
/* ids for values not referenced by nl_langinfo() */
#define KW_ZERO 10000
#define KW_GROUPING (KW_ZERO+1)
#define KW_INT_CURR_SYMBOL (KW_ZERO+2)
#define KW_CURRENCY_SYMBOL (KW_ZERO+3)
#define KW_MON_DECIMAL_POINT (KW_ZERO+4)
#define KW_MON_THOUSANDS_SEP (KW_ZERO+5)
#define KW_MON_GROUPING (KW_ZERO+6)
#define KW_POSITIVE_SIGN (KW_ZERO+7)
#define KW_NEGATIVE_SIGN (KW_ZERO+8)
#define KW_INT_FRAC_DIGITS (KW_ZERO+9)
#define KW_FRAC_DIGITS (KW_ZERO+10)
#define KW_P_CS_PRECEDES (KW_ZERO+11)
#define KW_P_SEP_BY_SPACE (KW_ZERO+12)
#define KW_N_CS_PRECEDES (KW_ZERO+13)
#define KW_N_SEP_BY_SPACE (KW_ZERO+14)
#define KW_P_SIGN_POSN (KW_ZERO+15)
#define KW_N_SIGN_POSN (KW_ZERO+16)
#define KW_INT_P_CS_PRECEDES (KW_ZERO+17)
#define KW_INT_P_SEP_BY_SPACE (KW_ZERO+18)
#define KW_INT_N_CS_PRECEDES (KW_ZERO+19)
#define KW_INT_N_SEP_BY_SPACE (KW_ZERO+20)
#define KW_INT_P_SIGN_POSN (KW_ZERO+21)
#define KW_INT_N_SIGN_POSN (KW_ZERO+22)
static const struct _kwinfo {
const char *name;
int isstr; /* true - string, false - number */
int catid; /* LC_* */
int value_ref;
const char *comment;
} kwinfo [] = {
{ "charmap", 1, LC_CTYPE, CODESET, "" }, /* hack */
{ "decimal_point", 1, LC_NUMERIC, RADIXCHAR, "" },
{ "thousands_sep", 1, LC_NUMERIC, THOUSEP, "" },
{ "grouping", 1, LC_NUMERIC, KW_GROUPING, "" },
{ "radixchar", 1, LC_NUMERIC, RADIXCHAR,
"Same as decimal_point (FreeBSD only)" }, /* compat */
{ "thousep", 1, LC_NUMERIC, THOUSEP,
"Same as thousands_sep (FreeBSD only)" }, /* compat */
{ "int_curr_symbol", 1, LC_MONETARY, KW_INT_CURR_SYMBOL, "" },
{ "currency_symbol", 1, LC_MONETARY, KW_CURRENCY_SYMBOL, "" },
{ "mon_decimal_point", 1, LC_MONETARY, KW_MON_DECIMAL_POINT, "" },
{ "mon_thousands_sep", 1, LC_MONETARY, KW_MON_THOUSANDS_SEP, "" },
{ "mon_grouping", 1, LC_MONETARY, KW_MON_GROUPING, "" },
{ "positive_sign", 1, LC_MONETARY, KW_POSITIVE_SIGN, "" },
{ "negative_sign", 1, LC_MONETARY, KW_NEGATIVE_SIGN, "" },
{ "int_frac_digits", 0, LC_MONETARY, KW_INT_FRAC_DIGITS, "" },
{ "frac_digits", 0, LC_MONETARY, KW_FRAC_DIGITS, "" },
{ "p_cs_precedes", 0, LC_MONETARY, KW_P_CS_PRECEDES, "" },
{ "p_sep_by_space", 0, LC_MONETARY, KW_P_SEP_BY_SPACE, "" },
{ "n_cs_precedes", 0, LC_MONETARY, KW_N_CS_PRECEDES, "" },
{ "n_sep_by_space", 0, LC_MONETARY, KW_N_SEP_BY_SPACE, "" },
{ "p_sign_posn", 0, LC_MONETARY, KW_P_SIGN_POSN, "" },
{ "n_sign_posn", 0, LC_MONETARY, KW_N_SIGN_POSN, "" },
{ "int_p_cs_precedes", 0, LC_MONETARY, KW_INT_P_CS_PRECEDES, "" },
{ "int_p_sep_by_space", 0, LC_MONETARY, KW_INT_P_SEP_BY_SPACE, "" },
{ "int_n_cs_precedes", 0, LC_MONETARY, KW_INT_N_CS_PRECEDES, "" },
{ "int_n_sep_by_space", 0, LC_MONETARY, KW_INT_N_SEP_BY_SPACE, "" },
{ "int_p_sign_posn", 0, LC_MONETARY, KW_INT_P_SIGN_POSN, "" },
{ "int_n_sign_posn", 0, LC_MONETARY, KW_INT_N_SIGN_POSN, "" },
{ "d_t_fmt", 1, LC_TIME, D_T_FMT, "" },
{ "d_fmt", 1, LC_TIME, D_FMT, "" },
{ "t_fmt", 1, LC_TIME, T_FMT, "" },
{ "am_str", 1, LC_TIME, AM_STR, "" },
{ "pm_str", 1, LC_TIME, PM_STR, "" },
{ "t_fmt_ampm", 1, LC_TIME, T_FMT_AMPM, "" },
{ "day_1", 1, LC_TIME, DAY_1, "" },
{ "day_2", 1, LC_TIME, DAY_2, "" },
{ "day_3", 1, LC_TIME, DAY_3, "" },
{ "day_4", 1, LC_TIME, DAY_4, "" },
{ "day_5", 1, LC_TIME, DAY_5, "" },
{ "day_6", 1, LC_TIME, DAY_6, "" },
{ "day_7", 1, LC_TIME, DAY_7, "" },
{ "abday_1", 1, LC_TIME, ABDAY_1, "" },
{ "abday_2", 1, LC_TIME, ABDAY_2, "" },
{ "abday_3", 1, LC_TIME, ABDAY_3, "" },
{ "abday_4", 1, LC_TIME, ABDAY_4, "" },
{ "abday_5", 1, LC_TIME, ABDAY_5, "" },
{ "abday_6", 1, LC_TIME, ABDAY_6, "" },
{ "abday_7", 1, LC_TIME, ABDAY_7, "" },
{ "mon_1", 1, LC_TIME, MON_1, "" },
{ "mon_2", 1, LC_TIME, MON_2, "" },
{ "mon_3", 1, LC_TIME, MON_3, "" },
{ "mon_4", 1, LC_TIME, MON_4, "" },
{ "mon_5", 1, LC_TIME, MON_5, "" },
{ "mon_6", 1, LC_TIME, MON_6, "" },
{ "mon_7", 1, LC_TIME, MON_7, "" },
{ "mon_8", 1, LC_TIME, MON_8, "" },
{ "mon_9", 1, LC_TIME, MON_9, "" },
{ "mon_10", 1, LC_TIME, MON_10, "" },
{ "mon_11", 1, LC_TIME, MON_11, "" },
{ "mon_12", 1, LC_TIME, MON_12, "" },
{ "abmon_1", 1, LC_TIME, ABMON_1, "" },
{ "abmon_2", 1, LC_TIME, ABMON_2, "" },
{ "abmon_3", 1, LC_TIME, ABMON_3, "" },
{ "abmon_4", 1, LC_TIME, ABMON_4, "" },
{ "abmon_5", 1, LC_TIME, ABMON_5, "" },
{ "abmon_6", 1, LC_TIME, ABMON_6, "" },
{ "abmon_7", 1, LC_TIME, ABMON_7, "" },
{ "abmon_8", 1, LC_TIME, ABMON_8, "" },
{ "abmon_9", 1, LC_TIME, ABMON_9, "" },
{ "abmon_10", 1, LC_TIME, ABMON_10, "" },
{ "abmon_11", 1, LC_TIME, ABMON_11, "" },
{ "abmon_12", 1, LC_TIME, ABMON_12, "" },
{ "altmon_1", 1, LC_TIME, ALTMON_1, "(FreeBSD only)" },
{ "altmon_2", 1, LC_TIME, ALTMON_2, "(FreeBSD only)" },
{ "altmon_3", 1, LC_TIME, ALTMON_3, "(FreeBSD only)" },
{ "altmon_4", 1, LC_TIME, ALTMON_4, "(FreeBSD only)" },
{ "altmon_5", 1, LC_TIME, ALTMON_5, "(FreeBSD only)" },
{ "altmon_6", 1, LC_TIME, ALTMON_6, "(FreeBSD only)" },
{ "altmon_7", 1, LC_TIME, ALTMON_7, "(FreeBSD only)" },
{ "altmon_8", 1, LC_TIME, ALTMON_8, "(FreeBSD only)" },
{ "altmon_9", 1, LC_TIME, ALTMON_9, "(FreeBSD only)" },
{ "altmon_10", 1, LC_TIME, ALTMON_10, "(FreeBSD only)" },
{ "altmon_11", 1, LC_TIME, ALTMON_11, "(FreeBSD only)" },
{ "altmon_12", 1, LC_TIME, ALTMON_12, "(FreeBSD only)" },
{ "era", 1, LC_TIME, ERA, "(unavailable)" },
{ "era_d_fmt", 1, LC_TIME, ERA_D_FMT, "(unavailable)" },
{ "era_d_t_fmt", 1, LC_TIME, ERA_D_T_FMT, "(unavailable)" },
{ "era_t_fmt", 1, LC_TIME, ERA_T_FMT, "(unavailable)" },
{ "alt_digits", 1, LC_TIME, ALT_DIGITS, "" },
{ "d_md_order", 1, LC_TIME, D_MD_ORDER,
"(FreeBSD only)" }, /* local */
{ "yesexpr", 1, LC_MESSAGES, YESEXPR, "" },
{ "noexpr", 1, LC_MESSAGES, NOEXPR, "" },
{ "yesstr", 1, LC_MESSAGES, YESSTR,
"(POSIX legacy)" }, /* compat */
{ "nostr", 1, LC_MESSAGES, NOSTR,
"(POSIX legacy)" } /* compat */
};
#define NKWINFO (nitems(kwinfo))
static const char *boguslocales[] = { "UTF-8" };
#define NBOGUS (nitems(boguslocales))
int
main(int argc, char *argv[])
{
int ch;
int tmp;
while ((ch = getopt(argc, argv, "ackms:")) != -1) {
switch (ch) {
case 'a':
all_locales = 1;
break;
case 'c':
prt_categories = 1;
break;
case 'k':
prt_keywords = 1;
break;
case 'm':
all_charmaps = 1;
break;
default:
usage();
}
}
argc -= optind;
argv += optind;
/* validate arguments */
if (all_locales && all_charmaps)
usage();
if ((all_locales || all_charmaps) && argc > 0)
usage();
if ((all_locales || all_charmaps) && (prt_categories || prt_keywords))
usage();
/* process '-a' */
if (all_locales) {
list_locales();
exit(0);
}
/* process '-m' */
if (all_charmaps) {
list_charmaps();
exit(0);
}
/* check for special case '-k list' */
tmp = 0;
if (prt_keywords && argc > 0)
while (tmp < argc)
if (strcasecmp(argv[tmp++], "list") == 0) {
showkeywordslist(argv[tmp]);
exit(0);
}
/* process '-c', '-k', or command line arguments. */
if (prt_categories || prt_keywords || argc > 0) {
if (prt_keywords || argc > 0)
setlocale(LC_ALL, "");
if (argc > 0) {
while (argc > 0) {
showdetails(*argv);
argv++;
argc--;
}
} else {
uint i;
for (i = 0; i < nitems(kwinfo); i++)
showdetails(kwinfo[i].name);
}
exit(0);
}
/* no arguments, show current locale state */
showlocale();
return (0);
}
void
usage(void)
{
printf("Usage: locale [ -a | -m ]\n"
" locale -k list [prefix]\n"
" locale [ -ck ] [keyword ...]\n");
exit(1);
}
/*
* Output information about all available locales
*
* XXX actually output of this function does not guarantee that locale
* is really available to application, since it can be broken or
* inconsistent thus setlocale() will fail. Maybe add '-V' function to
* also validate these locales?
*/
void
list_locales(void)
{
size_t i;
init_locales_list();
for (i = 0; i < locales->sl_cur; i++) {
printf("%s\n", locales->sl_str[i]);
}
}
/*
* qsort() helper function
*/
static int
scmp(const void *s1, const void *s2)
{
return strcmp(*(const char * const *)s1, *(const char * const *)s2);
}
/*
* Output information about all available charmaps
*
* XXX this function is doing a task in hackish way, i.e. by scaning
* list of locales, spliting their codeset part and building list of
* them.
*/
void
list_charmaps(void)
{
size_t i;
char *s, *cs;
StringList *charmaps;
/* initialize StringList */
charmaps = sl_init();
if (charmaps == NULL)
err(1, "could not allocate memory");
/* fetch locales list */
init_locales_list();
/* split codesets and build their list */
for (i = 0; i < locales->sl_cur; i++) {
s = locales->sl_str[i];
if ((cs = strchr(s, '.')) != NULL) {
cs++;
if (sl_find(charmaps, cs) == NULL)
sl_add(charmaps, cs);
}
}
/* add US-ASCII, if not yet added */
if (sl_find(charmaps, "US-ASCII") == NULL)
sl_add(charmaps, strdup("US-ASCII"));
/* sort the list */
qsort(charmaps->sl_str, charmaps->sl_cur, sizeof(char *), scmp);
/* print results */
for (i = 0; i < charmaps->sl_cur; i++) {
printf("%s\n", charmaps->sl_str[i]);
}
}
/*
* Retrieve sorted list of system locales (or user locales, if PATH_LOCALE
* environment variable is set)
*/
void
init_locales_list(void)
{
DIR *dirp;
struct dirent *dp;
size_t i;
int bogus;
/* why call this function twice ? */
if (locales != NULL)
return;
/* initialize StringList */
locales = sl_init();
if (locales == NULL)
err(1, "could not allocate memory");
/* get actual locales directory name */
if (__detect_path_locale() != 0)
err(1, "unable to find locales storage");
/* open locales directory */
dirp = opendir(_PathLocale);
if (dirp == NULL)
err(1, "could not open directory '%s'", _PathLocale);
/* scan directory and store its contents except "." and ".." */
while ((dp = readdir(dirp)) != NULL) {
if (*(dp->d_name) == '.')
continue; /* exclude "." and ".." */
for (bogus = i = 0; i < NBOGUS; i++)
if (strncmp(dp->d_name, boguslocales[i],
strlen(boguslocales[i])) == 0)
bogus = 1;
if (!bogus)
sl_add(locales, strdup(dp->d_name));
}
closedir(dirp);
/* make sure that 'POSIX' and 'C' locales are present in the list.
* POSIX 1003.1-2001 requires presence of 'POSIX' name only here, but
* we also list 'C' for constistency
*/
if (sl_find(locales, "POSIX") == NULL)
sl_add(locales, strdup("POSIX"));
if (sl_find(locales, "C") == NULL)
sl_add(locales, strdup("C"));
/* make output nicer, sort the list */
qsort(locales->sl_str, locales->sl_cur, sizeof(char *), scmp);
}
/*
* Show current locale status, depending on environment variables
*/
void
showlocale(void)
{
size_t i;
const char *lang, *vval, *eval;
setlocale(LC_ALL, "");
lang = getenv("LANG");
if (lang == NULL) {
lang = "";
}
printf("LANG=%s\n", lang);
/* XXX: if LANG is null, then set it to "C" to get implied values? */
for (i = 0; i < NLCINFO; i++) {
vval = setlocale(lcinfo[i].id, NULL);
eval = getenv(lcinfo[i].name);
if (eval != NULL && !strcmp(eval, vval)
&& strcmp(lang, vval)) {
/*
* Appropriate environment variable set, its value
* is valid and not overridden by LC_ALL
*
* XXX: possible side effect: if both LANG and
* overridden environment variable are set into same
* value, then it'll be assumed as 'implied'
*/
printf("%s=%s\n", lcinfo[i].name, vval);
} else {
printf("%s=\"%s\"\n", lcinfo[i].name, vval);
}
}
vval = getenv("LC_ALL");
if (vval == NULL) {
vval = "";
}
printf("LC_ALL=%s\n", vval);
}
char *
format_grouping(const char *binary)
{
static char rval[64];
const char *cp;
size_t roff;
int len;
rval[0] = '\0';
roff = 0;
for (cp = binary; *cp != '\0'; ++cp) {
if (*cp < 0)
break; /* garbage input */
len = snprintf(&rval[roff], sizeof(rval) - roff, "%u;", *cp);
if (len < 0 || (unsigned)len >= sizeof(rval) - roff)
break; /* insufficient space for output */
roff += len;
if (*cp == CHAR_MAX)
break; /* special termination */
}
/* Truncate at the last successfully snprintf()ed semicolon. */
if (roff != 0)
rval[roff - 1] = '\0';
return (&rval[0]);
}
/*
* keyword value lookup helper (via localeconv())
*/
char *
kwval_lconv(int id)
{
struct lconv *lc;
char *rval;
rval = NULL;
lc = localeconv();
switch (id) {
case KW_GROUPING:
rval = format_grouping(lc->grouping);
break;
case KW_INT_CURR_SYMBOL:
rval = lc->int_curr_symbol;
break;
case KW_CURRENCY_SYMBOL:
rval = lc->currency_symbol;
break;
case KW_MON_DECIMAL_POINT:
rval = lc->mon_decimal_point;
break;
case KW_MON_THOUSANDS_SEP:
rval = lc->mon_thousands_sep;
break;
case KW_MON_GROUPING:
rval = format_grouping(lc->mon_grouping);
break;
case KW_POSITIVE_SIGN:
rval = lc->positive_sign;
break;
case KW_NEGATIVE_SIGN:
rval = lc->negative_sign;
break;
case KW_INT_FRAC_DIGITS:
rval = &(lc->int_frac_digits);
break;
case KW_FRAC_DIGITS:
rval = &(lc->frac_digits);
break;
case KW_P_CS_PRECEDES:
rval = &(lc->p_cs_precedes);
break;
case KW_P_SEP_BY_SPACE:
rval = &(lc->p_sep_by_space);
break;
case KW_N_CS_PRECEDES:
rval = &(lc->n_cs_precedes);
break;
case KW_N_SEP_BY_SPACE:
rval = &(lc->n_sep_by_space);
break;
case KW_P_SIGN_POSN:
rval = &(lc->p_sign_posn);
break;
case KW_N_SIGN_POSN:
rval = &(lc->n_sign_posn);
break;
case KW_INT_P_CS_PRECEDES:
rval = &(lc->int_p_cs_precedes);
break;
case KW_INT_P_SEP_BY_SPACE:
rval = &(lc->int_p_sep_by_space);
break;
case KW_INT_N_CS_PRECEDES:
rval = &(lc->int_n_cs_precedes);
break;
case KW_INT_N_SEP_BY_SPACE:
rval = &(lc->int_n_sep_by_space);
break;
case KW_INT_P_SIGN_POSN:
rval = &(lc->int_p_sign_posn);
break;
case KW_INT_N_SIGN_POSN:
rval = &(lc->int_n_sign_posn);
break;
default:
break;
}
return (rval);
}
/*
* keyword value and properties lookup
*/
int
kwval_lookup(const char *kwname, char **kwval, int *cat, int *isstr)
{
int rval;
size_t i;
rval = 0;
for (i = 0; i < NKWINFO; i++) {
if (strcasecmp(kwname, kwinfo[i].name) == 0) {
rval = 1;
*cat = kwinfo[i].catid;
*isstr = kwinfo[i].isstr;
if (kwinfo[i].value_ref < KW_ZERO) {
*kwval = nl_langinfo(kwinfo[i].value_ref);
} else {
*kwval = kwval_lconv(kwinfo[i].value_ref);
}
break;
}
}
return (rval);
}
/*
* Show details about requested keyword according to '-k' and/or '-c'
* command line options specified.
*/
void
showdetails(const char *kw)
{
int isstr, cat, tmpval;
char *kwval;
if (kwval_lookup(kw, &kwval, &cat, &isstr) == 0) {
/*
* invalid keyword specified.
* XXX: any actions?
*/
fprintf(stderr, "Unknown keyword: `%s'\n", kw);
return;
}
if (prt_categories) {
if (prt_keywords)
printf("%-20s ", lookup_localecat(cat));
else
printf("%-20s\t%s\n", kw, lookup_localecat(cat));
}
if (prt_keywords) {
if (isstr) {
printf("%s=\"%s\"\n", kw, kwval);
} else {
tmpval = (char) *kwval;
printf("%s=%d\n", kw, tmpval);
}
}
if (!prt_categories && !prt_keywords) {
if (isstr) {
printf("%s\n", kwval);
} else {
tmpval = (char) *kwval;
printf("%d\n", tmpval);
}
}
}
/*
* Convert locale category id into string
*/
const char *
lookup_localecat(int cat)
{
size_t i;
for (i = 0; i < NLCINFO; i++)
if (lcinfo[i].id == cat) {
return (lcinfo[i].name);
}
return ("UNKNOWN");
}
/*
* Show list of keywords
*/
void
showkeywordslist(char *substring)
{
size_t i;
#define FMT "%-20s %-12s %-7s %-20s\n"
if (substring == NULL)
printf("List of available keywords\n\n");
else
printf("List of available keywords starting with '%s'\n\n",
substring);
printf(FMT, "Keyword", "Category", "Type", "Comment");
printf("-------------------- ------------ ------- --------------------\n");
for (i = 0; i < NKWINFO; i++) {
if (substring != NULL) {
if (strncmp(kwinfo[i].name, substring,
strlen(substring)) != 0)
continue;
}
printf(FMT,
kwinfo[i].name,
lookup_localecat(kwinfo[i].catid),
(kwinfo[i].isstr == 0) ? "number" : "string",
kwinfo[i].comment);
}
}