mirror of https://github.com/asterisk/asterisk
You can not select more than 25 topics
Topics must start with a letter or number, can include dashes ('-') and can be up to 35 characters long.
6125 lines
182 KiB
6125 lines
182 KiB
/*
|
|
* Asterisk -- A telephony toolkit for Linux.
|
|
*
|
|
* Say numbers and dates (maybe words one day too)
|
|
*
|
|
* Copyright (C) 1999, Mark Spencer
|
|
*
|
|
* Mark Spencer <markster@linux-support.net>
|
|
*
|
|
* This program is free software, distributed under the terms of
|
|
* the GNU General Public License
|
|
*
|
|
* 12-16-2004 : Support for Greek added by InAccess Networks (work funded by HOL, www.hol.gr)
|
|
* George Konstantoulakis <gkon@inaccessnetworks.com>
|
|
*/
|
|
|
|
#include <sys/types.h>
|
|
#include <string.h>
|
|
#include <stdlib.h>
|
|
#include <netinet/in.h>
|
|
#include <time.h>
|
|
#include <ctype.h>
|
|
#include <math.h>
|
|
#include <stdio.h>
|
|
|
|
#ifdef SOLARIS
|
|
#include <iso/limits_iso.h>
|
|
#endif
|
|
|
|
#include "asterisk.h"
|
|
|
|
ASTERISK_FILE_VERSION(__FILE__, "$Revision$")
|
|
|
|
#include "asterisk/file.h"
|
|
#include "asterisk/channel.h"
|
|
#include "asterisk/logger.h"
|
|
#include "asterisk/options.h"
|
|
#include "asterisk/say.h"
|
|
#include "asterisk/lock.h"
|
|
#include "asterisk/localtime.h"
|
|
#include "asterisk/utils.h"
|
|
|
|
/* Forward declaration */
|
|
static int wait_file(struct ast_channel *chan, const char *ints, const char *file, const char *lang);
|
|
|
|
int ast_say_character_str_full(struct ast_channel *chan, const char *str, const char *ints, const char *lang, int audiofd, int ctrlfd)
|
|
{
|
|
const char *fn;
|
|
char fnbuf[256];
|
|
char ltr;
|
|
int num = 0;
|
|
int res = 0;
|
|
|
|
while (str[num]) {
|
|
fn = NULL;
|
|
switch (str[num]) {
|
|
case ('*'):
|
|
fn = "digits/star";
|
|
break;
|
|
case ('#'):
|
|
fn = "digits/pound";
|
|
break;
|
|
case ('!'):
|
|
fn = "letters/exclaimation-point";
|
|
break;
|
|
case ('@'):
|
|
fn = "letters/at";
|
|
break;
|
|
case ('$'):
|
|
fn = "letters/dollar";
|
|
break;
|
|
case ('-'):
|
|
fn = "letters/dash";
|
|
break;
|
|
case ('.'):
|
|
fn = "letters/dot";
|
|
break;
|
|
case ('='):
|
|
fn = "letters/equals";
|
|
break;
|
|
case ('+'):
|
|
fn = "letters/plus";
|
|
break;
|
|
case ('/'):
|
|
fn = "letters/slash";
|
|
break;
|
|
case (' '):
|
|
fn = "letters/space";
|
|
break;
|
|
case ('0'):
|
|
case ('1'):
|
|
case ('2'):
|
|
case ('3'):
|
|
case ('4'):
|
|
case ('5'):
|
|
case ('6'):
|
|
case ('7'):
|
|
case ('8'):
|
|
case ('9'):
|
|
strcpy(fnbuf, "digits/X");
|
|
fnbuf[7] = str[num];
|
|
fn = fnbuf;
|
|
break;
|
|
default:
|
|
ltr = str[num];
|
|
if ('A' <= ltr && ltr <= 'Z') ltr += 'a' - 'A'; /* file names are all lower-case */
|
|
strcpy(fnbuf, "letters/X");
|
|
fnbuf[8] = ltr;
|
|
fn = fnbuf;
|
|
}
|
|
res = ast_streamfile(chan, fn, lang);
|
|
if (!res)
|
|
res = ast_waitstream_full(chan, ints, audiofd, ctrlfd);
|
|
ast_stopstream(chan);
|
|
num++;
|
|
}
|
|
|
|
return res;
|
|
}
|
|
|
|
int ast_say_character_str(struct ast_channel *chan, const char *str, const char *ints, const char *lang)
|
|
{
|
|
return ast_say_character_str_full(chan, str, ints, lang, -1, -1);
|
|
}
|
|
|
|
int ast_say_phonetic_str_full(struct ast_channel *chan, const char *str, const char *ints, const char *lang, int audiofd, int ctrlfd)
|
|
{
|
|
const char *fn;
|
|
char fnbuf[256];
|
|
char ltr;
|
|
int num = 0;
|
|
int res = 0;
|
|
|
|
while (str[num]) {
|
|
fn = NULL;
|
|
switch (str[num]) {
|
|
case ('*'):
|
|
fn = "digits/star";
|
|
break;
|
|
case ('#'):
|
|
fn = "digits/pound";
|
|
break;
|
|
case ('!'):
|
|
fn = "letters/exclaimation-point";
|
|
break;
|
|
case ('@'):
|
|
fn = "letters/at";
|
|
break;
|
|
case ('$'):
|
|
fn = "letters/dollar";
|
|
break;
|
|
case ('-'):
|
|
fn = "letters/dash";
|
|
break;
|
|
case ('.'):
|
|
fn = "letters/dot";
|
|
break;
|
|
case ('='):
|
|
fn = "letters/equals";
|
|
break;
|
|
case ('+'):
|
|
fn = "letters/plus";
|
|
break;
|
|
case ('/'):
|
|
fn = "letters/slash";
|
|
break;
|
|
case (' '):
|
|
fn = "letters/space";
|
|
break;
|
|
case ('0'):
|
|
case ('1'):
|
|
case ('2'):
|
|
case ('3'):
|
|
case ('4'):
|
|
case ('5'):
|
|
case ('6'):
|
|
case ('7'):
|
|
case ('8'):
|
|
strcpy(fnbuf, "digits/X");
|
|
fnbuf[7] = str[num];
|
|
fn = fnbuf;
|
|
break;
|
|
default: /* '9' falls here... */
|
|
ltr = str[num];
|
|
if ('A' <= ltr && ltr <= 'Z') ltr += 'a' - 'A'; /* file names are all lower-case */
|
|
strcpy(fnbuf, "phonetic/X_p");
|
|
fnbuf[9] = ltr;
|
|
fn = fnbuf;
|
|
}
|
|
res = ast_streamfile(chan, fn, lang);
|
|
if (!res)
|
|
res = ast_waitstream_full(chan, ints, audiofd, ctrlfd);
|
|
ast_stopstream(chan);
|
|
num++;
|
|
}
|
|
|
|
return res;
|
|
}
|
|
|
|
int ast_say_phonetic_str(struct ast_channel *chan, const char *str, const char *ints, const char *lang)
|
|
{
|
|
return ast_say_phonetic_str_full(chan, str, ints, lang, -1, -1);
|
|
}
|
|
|
|
int ast_say_digit_str_full(struct ast_channel *chan, const char *str, const char *ints, const char *lang, int audiofd, int ctrlfd)
|
|
{
|
|
const char *fn;
|
|
char fnbuf[256];
|
|
int num = 0;
|
|
int res = 0;
|
|
|
|
while (str[num] && !res) {
|
|
fn = NULL;
|
|
switch (str[num]) {
|
|
case ('*'):
|
|
fn = "digits/star";
|
|
break;
|
|
case ('#'):
|
|
fn = "digits/pound";
|
|
break;
|
|
case ('-'):
|
|
fn = "digits/minus";
|
|
break;
|
|
case '0':
|
|
case '1':
|
|
case '2':
|
|
case '3':
|
|
case '4':
|
|
case '5':
|
|
case '6':
|
|
case '7':
|
|
case '8':
|
|
case '9':
|
|
strcpy(fnbuf, "digits/X");
|
|
fnbuf[7] = str[num];
|
|
fn = fnbuf;
|
|
break;
|
|
}
|
|
if (fn) {
|
|
res = ast_streamfile(chan, fn, lang);
|
|
if (!res)
|
|
res = ast_waitstream_full(chan, ints, audiofd, ctrlfd);
|
|
ast_stopstream(chan);
|
|
}
|
|
num++;
|
|
}
|
|
|
|
return res;
|
|
}
|
|
|
|
int ast_say_digit_str(struct ast_channel *chan, const char *str, const char *ints, const char *lang)
|
|
{
|
|
return ast_say_digit_str_full(chan, str, ints, lang, -1, -1);
|
|
}
|
|
|
|
int ast_say_digits_full(struct ast_channel *chan, int num, const char *ints, const char *lang, int audiofd, int ctrlfd)
|
|
{
|
|
char fn2[256];
|
|
|
|
snprintf(fn2, sizeof(fn2), "%d", num);
|
|
return ast_say_digit_str_full(chan, fn2, ints, lang, audiofd, ctrlfd);
|
|
}
|
|
|
|
int ast_say_digits(struct ast_channel *chan, int num, const char *ints, const char *lang)
|
|
{
|
|
return ast_say_digits_full(chan, num, ints, lang, -1, -1);
|
|
}
|
|
|
|
/* Forward declarations */
|
|
/* Syntaxes supported, not really language codes.
|
|
da - Danish
|
|
de - German
|
|
en - English (US)
|
|
en_GB - English (British)
|
|
es - Spanish, Mexican
|
|
fr - French
|
|
he - Hebrew
|
|
it - Italian
|
|
nl - Dutch
|
|
no - Norwegian
|
|
pl - Polish
|
|
pt - Portuguese
|
|
se - Swedish
|
|
tw - Taiwanese
|
|
ru - Russian
|
|
|
|
Gender:
|
|
For Some languages the numbers differ for gender and plural
|
|
Use the option argument 'f' for female, 'm' for male and 'n' for neuter in languages like Portuguese, French, Spanish and German.
|
|
use the option argument 'c' is for commune and 'n' for neuter gender in nordic languages like Danish, Swedish and Norwegian.
|
|
use the option argument 'p' for plural enumerations like in German
|
|
|
|
Date/Time functions currently have less languages supported than saynumber().
|
|
|
|
Note that in future, we need to move to a model where we can differentiate further - e.g. between en_US & en_UK
|
|
|
|
See contrib/i18n.testsuite.conf for some examples of the different syntaxes
|
|
|
|
Portuguese sound files needed for Time/Date functions:
|
|
pt-ah
|
|
pt-ao
|
|
pt-de
|
|
pt-e
|
|
pt-ora
|
|
pt-meianoite
|
|
pt-meiodia
|
|
pt-sss
|
|
|
|
Spanish sound files needed for Time/Date functions:
|
|
es-de
|
|
es-el
|
|
|
|
Italian sound files needed for Time/Date functions:
|
|
ore-una
|
|
ore-mezzanotte
|
|
|
|
*/
|
|
|
|
/* Forward declarations of language specific variants of ast_say_number_full */
|
|
static int ast_say_number_full_en(struct ast_channel *chan, int num, const char *ints, const char *language, int audiofd, int ctrlfd);
|
|
static int ast_say_number_full_cz(struct ast_channel *chan, int num, const char *ints, const char *language, const char *options, int audiofd, int ctrlfd);
|
|
static int ast_say_number_full_da(struct ast_channel *chan, int num, const char *ints, const char *language, const char *options, int audiofd, int ctrlfd);
|
|
static int ast_say_number_full_de(struct ast_channel *chan, int num, const char *ints, const char *language, const char *options, int audiofd, int ctrlfd);
|
|
static int ast_say_number_full_en_GB(struct ast_channel *chan, int num, const char *ints, const char *language, int audiofd, int ctrlfd);
|
|
static int ast_say_number_full_es(struct ast_channel *chan, int num, const char *ints, const char *language, const char *options, int audiofd, int ctrlfd);
|
|
static int ast_say_number_full_fr(struct ast_channel *chan, int num, const char *ints, const char *language, const char *options, int audiofd, int ctrlfd);
|
|
static int ast_say_number_full_he(struct ast_channel *chan, int num, const char *ints, const char *language, const char *options, int audiofd, int ctrlfd);
|
|
static int ast_say_number_full_it(struct ast_channel *chan, int num, const char *ints, const char *language, int audiofd, int ctrlfd);
|
|
static int ast_say_number_full_nl(struct ast_channel *chan, int num, const char *ints, const char *language, int audiofd, int ctrlfd);
|
|
static int ast_say_number_full_no(struct ast_channel *chan, int num, const char *ints, const char *language, const char *options, int audiofd, int ctrlfd);
|
|
static int ast_say_number_full_pl(struct ast_channel *chan, int num, const char *ints, const char *language, const char *options, int audiofd, int ctrlfd);
|
|
static int ast_say_number_full_pt(struct ast_channel *chan, int num, const char *ints, const char *language, const char *options, int audiofd, int ctrlfd);
|
|
static int ast_say_number_full_se(struct ast_channel *chan, int num, const char *ints, const char *language, const char *options, int audiofd, int ctrlfd);
|
|
static int ast_say_number_full_tw(struct ast_channel *chan, int num, const char *ints, const char *language, int audiofd, int ctrlfd);
|
|
static int ast_say_number_full_gr(struct ast_channel *chan, int num, const char *ints, const char *language, int audiofd, int ctrlfd);
|
|
static int ast_say_number_full_ru(struct ast_channel *chan, int num, const char *ints, const char *language, const char *options, int audiofd, int ctrlfd);
|
|
|
|
/* Forward declarations of language specific variants of ast_say_enumeration_full */
|
|
static int ast_say_enumeration_full_en(struct ast_channel *chan, int num, const char *ints, const char *language, int audiofd, int ctrlfd);
|
|
static int ast_say_enumeration_full_da(struct ast_channel *chan, int num, const char *ints, const char *language, const char *options, int audiofd, int ctrlfd);
|
|
static int ast_say_enumeration_full_de(struct ast_channel *chan, int num, const char *ints, const char *language, const char *options, int audiofd, int ctrlfd);
|
|
|
|
/* Forward declarations of ast_say_date, ast_say_datetime and ast_say_time functions */
|
|
static int ast_say_date_en(struct ast_channel *chan, time_t t, const char *ints, const char *lang);
|
|
static int ast_say_date_da(struct ast_channel *chan, time_t t, const char *ints, const char *lang);
|
|
static int ast_say_date_de(struct ast_channel *chan, time_t t, const char *ints, const char *lang);
|
|
static int ast_say_date_fr(struct ast_channel *chan, time_t t, const char *ints, const char *lang);
|
|
static int ast_say_date_nl(struct ast_channel *chan, time_t t, const char *ints, const char *lang);
|
|
static int ast_say_date_pt(struct ast_channel *chan, time_t t, const char *ints, const char *lang);
|
|
static int ast_say_date_gr(struct ast_channel *chan, time_t t, const char *ints, const char *lang);
|
|
|
|
static int ast_say_date_with_format_en(struct ast_channel *chan, time_t time, const char *ints, const char *lang, const char *format, const char *timezone);
|
|
static int ast_say_date_with_format_da(struct ast_channel *chan, time_t time, const char *ints, const char *lang, const char *format, const char *timezone);
|
|
static int ast_say_date_with_format_de(struct ast_channel *chan, time_t time, const char *ints, const char *lang, const char *format, const char *timezone);
|
|
static int ast_say_date_with_format_es(struct ast_channel *chan, time_t time, const char *ints, const char *lang, const char *format, const char *timezone);
|
|
static int ast_say_date_with_format_he(struct ast_channel *chan, time_t time, const char *ints, const char *lang, const char *format, const char *timezone);
|
|
static int ast_say_date_with_format_fr(struct ast_channel *chan, time_t time, const char *ints, const char *lang, const char *format, const char *timezone);
|
|
static int ast_say_date_with_format_it(struct ast_channel *chan, time_t time, const char *ints, const char *lang, const char *format, const char *timezone);
|
|
static int ast_say_date_with_format_nl(struct ast_channel *chan, time_t time, const char *ints, const char *lang, const char *format, const char *timezone);
|
|
static int ast_say_date_with_format_pt(struct ast_channel *chan, time_t time, const char *ints, const char *lang, const char *format, const char *timezone);
|
|
static int ast_say_date_with_format_tw(struct ast_channel *chan, time_t time, const char *ints, const char *lang, const char *format, const char *timezone);
|
|
static int ast_say_date_with_format_gr(struct ast_channel *chan, time_t time, const char *ints, const char *lang, const char *format, const char *timezone);
|
|
|
|
static int ast_say_time_en(struct ast_channel *chan, time_t t, const char *ints, const char *lang);
|
|
static int ast_say_time_de(struct ast_channel *chan, time_t t, const char *ints, const char *lang);
|
|
static int ast_say_time_fr(struct ast_channel *chan, time_t t, const char *ints, const char *lang);
|
|
static int ast_say_time_nl(struct ast_channel *chan, time_t t, const char *ints, const char *lang);
|
|
static int ast_say_time_pt(struct ast_channel *chan, time_t t, const char *ints, const char *lang);
|
|
static int ast_say_time_tw(struct ast_channel *chan, time_t t, const char *ints, const char *lang);
|
|
static int ast_say_time_gr(struct ast_channel *chan, time_t t, const char *ints, const char *lang);
|
|
|
|
static int ast_say_datetime_en(struct ast_channel *chan, time_t t, const char *ints, const char *lang);
|
|
static int ast_say_datetime_de(struct ast_channel *chan, time_t t, const char *ints, const char *lang);
|
|
static int ast_say_datetime_fr(struct ast_channel *chan, time_t t, const char *ints, const char *lang);
|
|
static int ast_say_datetime_nl(struct ast_channel *chan, time_t t, const char *ints, const char *lang);
|
|
static int ast_say_datetime_pt(struct ast_channel *chan, time_t t, const char *ints, const char *lang);
|
|
static int ast_say_datetime_tw(struct ast_channel *chan, time_t t, const char *ints, const char *lang);
|
|
static int ast_say_datetime_gr(struct ast_channel *chan, time_t t, const char *ints, const char *lang);
|
|
|
|
static int ast_say_datetime_from_now_en(struct ast_channel *chan, time_t t, const char *ints, const char *lang);
|
|
static int ast_say_datetime_from_now_fr(struct ast_channel *chan, time_t t, const char *ints, const char *lang);
|
|
static int ast_say_datetime_from_now_pt(struct ast_channel *chan, time_t t, const char *ints, const char *lang);
|
|
|
|
static int wait_file(struct ast_channel *chan, const char *ints, const char *file, const char *lang)
|
|
{
|
|
int res;
|
|
if ((res = ast_streamfile(chan, file, lang)))
|
|
ast_log(LOG_WARNING, "Unable to play message %s\n", file);
|
|
if (!res)
|
|
res = ast_waitstream(chan, ints);
|
|
return res;
|
|
}
|
|
|
|
/*--- ast_say_number_full: call language-specific functions */
|
|
/* Called from AGI */
|
|
int ast_say_number_full(struct ast_channel *chan, int num, const char *ints, const char *language, const char *options, int audiofd, int ctrlfd)
|
|
{
|
|
if (!strcasecmp(language,"en") ) { /* English syntax */
|
|
return(ast_say_number_full_en(chan, num, ints, language, audiofd, ctrlfd));
|
|
} else if (!strcasecmp(language, "cz") ) { /* Czech syntax */
|
|
return(ast_say_number_full_cz(chan, num, ints, language, options, audiofd, ctrlfd));
|
|
} else if (!strcasecmp(language, "da") ) { /* Danish syntax */
|
|
return(ast_say_number_full_da(chan, num, ints, language, options, audiofd, ctrlfd));
|
|
} else if (!strcasecmp(language, "de") ) { /* German syntax */
|
|
return(ast_say_number_full_de(chan, num, ints, language, options, audiofd, ctrlfd));
|
|
} else if (!strcasecmp(language, "en_GB") ) { /* British syntax */
|
|
return(ast_say_number_full_en_GB(chan, num, ints, language, audiofd, ctrlfd));
|
|
} else if (!strcasecmp(language, "no") ) { /* Norwegian syntax */
|
|
return(ast_say_number_full_no(chan, num, ints, language, options, audiofd, ctrlfd));
|
|
} else if (!strcasecmp(language, "es") || !strcasecmp(language, "mx")) { /* Spanish syntax */
|
|
return(ast_say_number_full_es(chan, num, ints, language, options, audiofd, ctrlfd));
|
|
} else if (!strcasecmp(language, "fr") ) { /* French syntax */
|
|
return(ast_say_number_full_fr(chan, num, ints, language, options, audiofd, ctrlfd));
|
|
} else if (!strcasecmp(language, "he") ) { /* Hebrew syntax */
|
|
return(ast_say_number_full_he(chan, num, ints, language, options, audiofd, ctrlfd));
|
|
} else if (!strcasecmp(language, "it") ) { /* Italian syntax */
|
|
return(ast_say_number_full_it(chan, num, ints, language, audiofd, ctrlfd));
|
|
} else if (!strcasecmp(language, "nl") ) { /* Dutch syntax */
|
|
return(ast_say_number_full_nl(chan, num, ints, language, audiofd, ctrlfd));
|
|
} else if (!strcasecmp(language, "pl") ) { /* Polish syntax */
|
|
return(ast_say_number_full_pl(chan, num, ints, language, options, audiofd, ctrlfd));
|
|
} else if (!strcasecmp(language, "pt") ) { /* Portuguese syntax */
|
|
return(ast_say_number_full_pt(chan, num, ints, language, options, audiofd, ctrlfd));
|
|
} else if (!strcasecmp(language, "se") ) { /* Swedish syntax */
|
|
return(ast_say_number_full_se(chan, num, ints, language, options, audiofd, ctrlfd));
|
|
} else if (!strcasecmp(language, "tw")) { /* Taiwanese syntax */
|
|
return(ast_say_number_full_tw(chan, num, ints, language, audiofd, ctrlfd));
|
|
} else if (!strcasecmp(language, "gr") ) { /* Greek syntax */
|
|
return(ast_say_number_full_gr(chan, num, ints, language, audiofd, ctrlfd));
|
|
} else if (!strcasecmp(language, "ru") ) { /* Russian syntax */
|
|
return(ast_say_number_full_ru(chan, num, ints, language, options, audiofd, ctrlfd));
|
|
}
|
|
|
|
/* Default to english */
|
|
return(ast_say_number_full_en(chan, num, ints, language, audiofd, ctrlfd));
|
|
}
|
|
|
|
/*--- ast_say_number: call language-specific functions without file descriptors */
|
|
int ast_say_number(struct ast_channel *chan, int num, const char *ints, const char *language, const char *options)
|
|
{
|
|
return(ast_say_number_full(chan, num, ints, language, options, -1, -1));
|
|
}
|
|
|
|
/*--- ast_say_number_full_en: English syntax */
|
|
/* This is the default syntax, if no other syntax defined in this file is used */
|
|
static int ast_say_number_full_en(struct ast_channel *chan, int num, const char *ints, const char *language, int audiofd, int ctrlfd)
|
|
{
|
|
int res = 0;
|
|
int playh = 0;
|
|
char fn[256] = "";
|
|
if (!num)
|
|
return ast_say_digits_full(chan, 0,ints, language, audiofd, ctrlfd);
|
|
|
|
while(!res && (num || playh)) {
|
|
if (num < 0) {
|
|
snprintf(fn, sizeof(fn), "digits/minus");
|
|
if ( num > INT_MIN ) {
|
|
num = -num;
|
|
} else {
|
|
num = 0;
|
|
}
|
|
} else if (playh) {
|
|
snprintf(fn, sizeof(fn), "digits/hundred");
|
|
playh = 0;
|
|
} else if (num < 20) {
|
|
snprintf(fn, sizeof(fn), "digits/%d", num);
|
|
num = 0;
|
|
} else if (num < 100) {
|
|
snprintf(fn, sizeof(fn), "digits/%d", (num /10) * 10);
|
|
num -= ((num / 10) * 10);
|
|
} else {
|
|
if (num < 1000){
|
|
snprintf(fn, sizeof(fn), "digits/%d", (num/100));
|
|
playh++;
|
|
num -= ((num / 100) * 100);
|
|
} else {
|
|
if (num < 1000000) { /* 1,000,000 */
|
|
res = ast_say_number_full_en(chan, num / 1000, ints, language, audiofd, ctrlfd);
|
|
if (res)
|
|
return res;
|
|
num = num % 1000;
|
|
snprintf(fn, sizeof(fn), "digits/thousand");
|
|
} else {
|
|
if (num < 1000000000) { /* 1,000,000,000 */
|
|
res = ast_say_number_full_en(chan, num / 1000000, ints, language, audiofd, ctrlfd);
|
|
if (res)
|
|
return res;
|
|
num = num % 1000000;
|
|
snprintf(fn, sizeof(fn), "digits/million");
|
|
} else {
|
|
ast_log(LOG_DEBUG, "Number '%d' is too big for me\n", num);
|
|
res = -1;
|
|
}
|
|
}
|
|
}
|
|
}
|
|
if (!res) {
|
|
if(!ast_streamfile(chan, fn, language)) {
|
|
if ((audiofd > -1) && (ctrlfd > -1))
|
|
res = ast_waitstream_full(chan, ints, audiofd, ctrlfd);
|
|
else
|
|
res = ast_waitstream(chan, ints);
|
|
}
|
|
ast_stopstream(chan);
|
|
}
|
|
}
|
|
return res;
|
|
}
|
|
|
|
static int exp10_int(int power)
|
|
{
|
|
int x, res= 1;
|
|
for (x=0;x<power;x++)
|
|
res *= 10;
|
|
return res;
|
|
}
|
|
|
|
/*--- ast_say_number_full_cz: Czech syntax */
|
|
/* files needed:
|
|
* 1m,2m - gender male
|
|
* 1w,2w - gender female
|
|
* 3,4,...,20
|
|
* 30,40,...,90
|
|
*
|
|
* hundereds - 100 - sto, 200 - 2ste, 300,400 3,4sta, 500,600,...,900 5,6,...9set
|
|
*
|
|
* for each number 10^(3n + 3) exist 3 files represented as:
|
|
* 1 tousand = jeden tisic = 1_E3
|
|
* 2,3,4 tousands = dva,tri,ctyri tisice = 2-3_E3
|
|
* 5,6,... tousands = pet,sest,... tisic = 5_E3
|
|
*
|
|
* million = _E6
|
|
* miliard = _E9
|
|
* etc...
|
|
*
|
|
* tousand, milion are gender male, so 1 and 2 is 1m 2m
|
|
* miliard is gender female, so 1 and 2 is 1w 2w
|
|
*/
|
|
static int ast_say_number_full_cz(struct ast_channel *chan, int num, const char *ints, const char *language, const char *options, int audiofd, int ctrlfd)
|
|
{
|
|
int res = 0;
|
|
int playh = 0;
|
|
char fn[256] = "";
|
|
|
|
int hundered = 0;
|
|
int left = 0;
|
|
int length = 0;
|
|
|
|
/* options - w = woman, m = man, n = neutral. Defaultl is woman */
|
|
if (!options)
|
|
options = "w";
|
|
|
|
if (!num)
|
|
return ast_say_digits_full(chan, 0,ints, language, audiofd, ctrlfd);
|
|
|
|
while(!res && (num || playh)) {
|
|
if (num < 0) {
|
|
snprintf(fn, sizeof(fn), "digits/minus");
|
|
if ( num > INT_MIN ) {
|
|
num = -num;
|
|
} else {
|
|
num = 0;
|
|
}
|
|
} else if (num < 3 ) {
|
|
snprintf(fn, sizeof(fn), "digits/%d%c",num,options[0]);
|
|
playh = 0;
|
|
num = 0;
|
|
} else if (num < 20) {
|
|
snprintf(fn, sizeof(fn), "digits/%d",num);
|
|
playh = 0;
|
|
num = 0;
|
|
} else if (num < 100) {
|
|
snprintf(fn, sizeof(fn), "digits/%d", (num /10) * 10);
|
|
num -= ((num / 10) * 10);
|
|
} else if (num < 1000) {
|
|
hundered = num / 100;
|
|
if ( hundered == 1 ) {
|
|
snprintf(fn, sizeof(fn), "digits/1sto");
|
|
} else if ( hundered == 2 ) {
|
|
snprintf(fn, sizeof(fn), "digits/2ste");
|
|
} else {
|
|
res = ast_say_number_full_cz(chan,hundered,ints,language,options,audiofd,ctrlfd);
|
|
if (res)
|
|
return res;
|
|
if (hundered == 3 || hundered == 4) {
|
|
snprintf(fn, sizeof(fn), "digits/sta");
|
|
} else if ( hundered > 4 ) {
|
|
snprintf(fn, sizeof(fn), "digits/set");
|
|
}
|
|
}
|
|
num -= (hundered * 100);
|
|
} else { /* num > 1000 */
|
|
length = (int)log10(num)+1;
|
|
while ( (length % 3 ) != 1 ) {
|
|
length--;
|
|
}
|
|
left = num / (exp10_int(length-1));
|
|
if ( left == 2 ) {
|
|
switch (length-1) {
|
|
case 9: options = "w"; /* 1,000,000,000 gender female */
|
|
break;
|
|
default : options = "m"; /* others are male */
|
|
}
|
|
}
|
|
if ( left > 1 ) { /* we dont say "one thousand" but only thousand */
|
|
res = ast_say_number_full_cz(chan,left,ints,language,options,audiofd,ctrlfd);
|
|
if (res)
|
|
return res;
|
|
}
|
|
if ( left >= 5 ) { /* >= 5 have the same declesion */
|
|
snprintf(fn, sizeof(fn), "digits/5_E%d",length-1);
|
|
} else if ( left >= 2 && left <= 4 ) {
|
|
snprintf(fn, sizeof(fn), "digits/2-4_E%d",length-1);
|
|
} else { /* left == 1 */
|
|
snprintf(fn, sizeof(fn), "digits/1_E%d",length-1);
|
|
}
|
|
num -= left * (exp10_int(length-1));
|
|
}
|
|
if (!res) {
|
|
if(!ast_streamfile(chan, fn, language)) {
|
|
if ((audiofd > -1) && (ctrlfd > -1)) {
|
|
res = ast_waitstream_full(chan, ints, audiofd, ctrlfd);
|
|
} else {
|
|
res = ast_waitstream(chan, ints);
|
|
}
|
|
}
|
|
ast_stopstream(chan);
|
|
}
|
|
}
|
|
return res;
|
|
}
|
|
|
|
/*--- ast_say_number_full_da: Danish syntax */
|
|
/* New files:
|
|
In addition to English, the following sounds are required: "1N", "millions", "and" and "1-and" through "9-and"
|
|
*/
|
|
static int ast_say_number_full_da(struct ast_channel *chan, int num, const char *ints, const char *language, const char *options, int audiofd, int ctrlfd)
|
|
{
|
|
int res = 0;
|
|
int playh = 0;
|
|
int playa = 0;
|
|
int cn = 1; /* +1 = commune; -1 = neuter */
|
|
char fn[256] = "";
|
|
if (!num)
|
|
return ast_say_digits_full(chan, 0,ints, language, audiofd, ctrlfd);
|
|
|
|
if (options && !strncasecmp(options, "n",1)) cn = -1;
|
|
|
|
while(!res && (num || playh || playa )) {
|
|
/* The grammar for Danish numbers is the same as for English except
|
|
* for the following:
|
|
* - 1 exists in both commune ("en", file "1N") and neuter ("et", file "1")
|
|
* - numbers 20 through 99 are said in reverse order, i.e. 21 is
|
|
* "one-and twenty" and 68 is "eight-and sixty".
|
|
* - "million" is different in singular and plural form
|
|
* - numbers > 1000 with zero as the third digit from last have an
|
|
* "and" before the last two digits, i.e. 2034 is "two thousand and
|
|
* four-and thirty" and 1000012 is "one million and twelve".
|
|
*/
|
|
if (num < 0) {
|
|
snprintf(fn, sizeof(fn), "digits/minus");
|
|
if ( num > INT_MIN ) {
|
|
num = -num;
|
|
} else {
|
|
num = 0;
|
|
}
|
|
} else if (playh) {
|
|
snprintf(fn, sizeof(fn), "digits/hundred");
|
|
playh = 0;
|
|
} else if (playa) {
|
|
snprintf(fn, sizeof(fn), "digits/and");
|
|
playa = 0;
|
|
} else if (num == 1 && cn == -1) {
|
|
snprintf(fn, sizeof(fn), "digits/1N");
|
|
num = 0;
|
|
} else if (num < 20) {
|
|
snprintf(fn, sizeof(fn), "digits/%d", num);
|
|
num = 0;
|
|
} else if (num < 100) {
|
|
int ones = num % 10;
|
|
if (ones) {
|
|
snprintf(fn, sizeof(fn), "digits/%d-and", ones);
|
|
num -= ones;
|
|
} else {
|
|
snprintf(fn, sizeof(fn), "digits/%d", num);
|
|
num = 0;
|
|
}
|
|
} else {
|
|
if (num < 1000) {
|
|
int hundreds = num / 100;
|
|
if (hundreds == 1)
|
|
snprintf(fn, sizeof(fn), "digits/1N");
|
|
else
|
|
snprintf(fn, sizeof(fn), "digits/%d", (num / 100));
|
|
|
|
playh++;
|
|
num -= 100 * hundreds;
|
|
if (num)
|
|
playa++;
|
|
|
|
} else {
|
|
if (num < 1000000) {
|
|
res = ast_say_number_full_da(chan, num / 1000, ints, language, "n", audiofd, ctrlfd);
|
|
if (res)
|
|
return res;
|
|
num = num % 1000;
|
|
snprintf(fn, sizeof(fn), "digits/thousand");
|
|
} else {
|
|
if (num < 1000000000) {
|
|
int millions = num / 1000000;
|
|
res = ast_say_number_full_da(chan, millions, ints, language, "c", audiofd, ctrlfd);
|
|
if (res)
|
|
return res;
|
|
if (millions == 1)
|
|
snprintf(fn, sizeof(fn), "digits/million");
|
|
else
|
|
snprintf(fn, sizeof(fn), "digits/millions");
|
|
num = num % 1000000;
|
|
} else {
|
|
ast_log(LOG_DEBUG, "Number '%d' is too big for me\n", num);
|
|
res = -1;
|
|
}
|
|
}
|
|
if (num && num < 100)
|
|
playa++;
|
|
}
|
|
}
|
|
if (!res) {
|
|
if(!ast_streamfile(chan, fn, language)) {
|
|
if ((audiofd > -1) && (ctrlfd > -1))
|
|
res = ast_waitstream_full(chan, ints, audiofd, ctrlfd);
|
|
else
|
|
res = ast_waitstream(chan, ints);
|
|
}
|
|
ast_stopstream(chan);
|
|
}
|
|
}
|
|
return res;
|
|
}
|
|
|
|
/*--- ast_say_number_full_de: German syntax */
|
|
/* New files:
|
|
In addition to English, the following sounds are required:
|
|
"millions"
|
|
"1-and" through "9-and"
|
|
"1F" (eine)
|
|
"1N" (ein)
|
|
NB "1" is recorded as 'eins'
|
|
*/
|
|
static int ast_say_number_full_de(struct ast_channel *chan, int num, const char *ints, const char *language, const char *options, int audiofd, int ctrlfd)
|
|
{
|
|
int res = 0, t = 0;
|
|
int mf = 1; /* +1 = male and neuter; -1 = female */
|
|
char fn[256] = "";
|
|
char fna[256] = "";
|
|
if (!num)
|
|
return ast_say_digits_full(chan, 0,ints, language, audiofd, ctrlfd);
|
|
|
|
if (options && (!strncasecmp(options, "f",1)))
|
|
mf = -1;
|
|
|
|
while(!res && num) {
|
|
/* The grammar for German numbers is the same as for English except
|
|
* for the following:
|
|
* - numbers 20 through 99 are said in reverse order, i.e. 21 is
|
|
* "one-and twenty" and 68 is "eight-and sixty".
|
|
* - "one" varies according to gender
|
|
* - 100 is 'hundert', however all other instances are 'ein hundert'
|
|
* - 1000 is 'tausend', however all other instances are 'ein tausend'
|
|
* - 1000000 is always 'eine million'
|
|
* - "million" is different in singular and plural form
|
|
*/
|
|
if (num < 0) {
|
|
snprintf(fn, sizeof(fn), "digits/minus");
|
|
if ( num > INT_MIN ) {
|
|
num = -num;
|
|
} else {
|
|
num = 0;
|
|
}
|
|
} else if (num < 100 && t) {
|
|
snprintf(fn, sizeof(fn), "digits/and");
|
|
t = 0;
|
|
} else if (num == 1 && mf == -1) {
|
|
snprintf(fn, sizeof(fn), "digits/%dF", num);
|
|
num = 0;
|
|
} else if (num < 20) {
|
|
snprintf(fn, sizeof(fn), "digits/%d", num);
|
|
num = 0;
|
|
} else if (num < 100) {
|
|
int ones = num % 10;
|
|
if (ones) {
|
|
snprintf(fn, sizeof(fn), "digits/%d-and", ones);
|
|
num -= ones;
|
|
} else {
|
|
snprintf(fn, sizeof(fn), "digits/%d", num);
|
|
num = 0;
|
|
}
|
|
} else if (num == 100 && t == 0) {
|
|
snprintf(fn, sizeof(fn), "digits/hundred");
|
|
num = 0;
|
|
} else if (num < 1000) {
|
|
int hundreds = num / 100;
|
|
num = num % 100;
|
|
if (hundreds == 1) {
|
|
snprintf(fn, sizeof(fn), "digits/1N");
|
|
} else {
|
|
snprintf(fn, sizeof(fn), "digits/%d", hundreds);
|
|
}
|
|
snprintf(fna, sizeof(fna), "digits/hundred");
|
|
t = 1;
|
|
} else if (num == 1000 && t == 0) {
|
|
snprintf(fn, sizeof(fn), "digits/thousand");
|
|
num = 0;
|
|
} else if (num < 1000000) {
|
|
int thousands = num / 1000;
|
|
num = num % 1000;
|
|
t = 1;
|
|
if (thousands == 1) {
|
|
snprintf(fn, sizeof(fn), "digits/1N");
|
|
snprintf(fna, sizeof(fna), "digits/thousand");
|
|
} else {
|
|
res = ast_say_number_full_de(chan, thousands, ints, language, options, audiofd, ctrlfd);
|
|
if (res)
|
|
return res;
|
|
snprintf(fn, sizeof(fn), "digits/thousand");
|
|
}
|
|
} else if (num < 1000000000) {
|
|
int millions = num / 1000000;
|
|
num = num % 1000000;
|
|
t = 1;
|
|
if (millions == 1) {
|
|
snprintf(fn, sizeof(fn), "digits/1F");
|
|
snprintf(fna, sizeof(fna), "digits/million");
|
|
} else {
|
|
res = ast_say_number_full_de(chan, millions, ints, language, options, audiofd, ctrlfd);
|
|
if (res)
|
|
return res;
|
|
snprintf(fn, sizeof(fn), "digits/millions");
|
|
}
|
|
} else if (num <= INT_MAX) {
|
|
int billions = num / 1000000000;
|
|
num = num % 1000000000;
|
|
t = 1;
|
|
if (billions == 1) {
|
|
snprintf(fn, sizeof(fn), "digits/1F");
|
|
snprintf(fna, sizeof(fna), "digits/milliard");
|
|
} else {
|
|
res = ast_say_number_full_de(chan, billions, ints, language, options, audiofd, ctrlfd);
|
|
if (res) {
|
|
return res;
|
|
}
|
|
snprintf(fn, sizeof(fn), "digits/milliards");
|
|
}
|
|
} else {
|
|
ast_log(LOG_DEBUG, "Number '%d' is too big for me\n", num);
|
|
res = -1;
|
|
}
|
|
if (!res) {
|
|
if(!ast_streamfile(chan, fn, language)) {
|
|
if ((audiofd > -1) && (ctrlfd > -1))
|
|
res = ast_waitstream_full(chan, ints, audiofd, ctrlfd);
|
|
else
|
|
res = ast_waitstream(chan, ints);
|
|
}
|
|
ast_stopstream(chan);
|
|
if (!res) {
|
|
if (strlen(fna) != 0 && !ast_streamfile(chan, fna, language)) {
|
|
if ((audiofd > -1) && (ctrlfd > -1))
|
|
res = ast_waitstream_full(chan, ints, audiofd, ctrlfd);
|
|
else
|
|
res = ast_waitstream(chan, ints);
|
|
}
|
|
ast_stopstream(chan);
|
|
strcpy(fna, "");
|
|
}
|
|
}
|
|
}
|
|
return res;
|
|
}
|
|
|
|
/*--- ast_say_number_full_en_GB: British and Norwegian syntax */
|
|
/* New files:
|
|
In addition to American English, the following sounds are required: "and"
|
|
*/
|
|
static int ast_say_number_full_en_GB(struct ast_channel *chan, int num, const char *ints, const char *language, int audiofd, int ctrlfd)
|
|
{
|
|
int res = 0;
|
|
int playh = 0;
|
|
int playa = 0;
|
|
char fn[256] = "";
|
|
if (!num)
|
|
return ast_say_digits_full(chan, 0,ints, language, audiofd, ctrlfd);
|
|
|
|
while(!res && (num || playh || playa )) {
|
|
if (num < 0) {
|
|
snprintf(fn, sizeof(fn), "digits/minus");
|
|
if ( num > INT_MIN ) {
|
|
num = -num;
|
|
} else {
|
|
num = 0;
|
|
}
|
|
} else if (playh) {
|
|
snprintf(fn, sizeof(fn), "digits/hundred");
|
|
playh = 0;
|
|
} else if (playa) {
|
|
snprintf(fn, sizeof(fn), "digits/and");
|
|
playa = 0;
|
|
} else if (num < 20) {
|
|
snprintf(fn, sizeof(fn), "digits/%d", num);
|
|
num = 0;
|
|
} else if (num < 100) {
|
|
snprintf(fn, sizeof(fn), "digits/%d", (num /10) * 10);
|
|
num -= ((num / 10) * 10);
|
|
} else if (num < 1000) {
|
|
int hundreds = num / 100;
|
|
snprintf(fn, sizeof(fn), "digits/%d", (num / 100));
|
|
|
|
playh++;
|
|
num -= 100 * hundreds;
|
|
if (num)
|
|
playa++;
|
|
} else if (num < 1000000) {
|
|
res = ast_say_number_full_en_GB(chan, num / 1000, ints, language, audiofd, ctrlfd);
|
|
if (res)
|
|
return res;
|
|
snprintf(fn, sizeof(fn), "digits/thousand");
|
|
num = num % 1000;
|
|
if (num && num < 100)
|
|
playa++;
|
|
} else if (num < 1000000000) {
|
|
int millions = num / 1000000;
|
|
res = ast_say_number_full_en_GB(chan, millions, ints, language, audiofd, ctrlfd);
|
|
if (res)
|
|
return res;
|
|
snprintf(fn, sizeof(fn), "digits/million");
|
|
num = num % 1000000;
|
|
if (num && num < 100)
|
|
playa++;
|
|
} else {
|
|
ast_log(LOG_DEBUG, "Number '%d' is too big for me\n", num);
|
|
res = -1;
|
|
}
|
|
|
|
if (!res) {
|
|
if(!ast_streamfile(chan, fn, language)) {
|
|
if ((audiofd > -1) && (ctrlfd > -1))
|
|
res = ast_waitstream_full(chan, ints, audiofd, ctrlfd);
|
|
else
|
|
res = ast_waitstream(chan, ints);
|
|
}
|
|
ast_stopstream(chan);
|
|
}
|
|
}
|
|
return res;
|
|
}
|
|
|
|
/*--- ast_say_number_full_es: Spanish syntax */
|
|
/* New files:
|
|
Requires a few new audios:
|
|
1F.gsm: feminine 'una'
|
|
21.gsm thru 29.gsm, cien.gsm, mil.gsm, millon.gsm, millones.gsm, 100.gsm, 200.gsm, 300.gsm, 400.gsm, 500.gsm, 600.gsm, 700.gsm, 800.gsm, 900.gsm, y.gsm
|
|
*/
|
|
static int ast_say_number_full_es(struct ast_channel *chan, int num, const char *ints, const char *language, const char *options, int audiofd, int ctrlfd)
|
|
{
|
|
int res = 0;
|
|
int playa = 0;
|
|
int mf = 0; /* +1 = male; -1 = female */
|
|
char fn[256] = "";
|
|
if (!num)
|
|
return ast_say_digits_full(chan, 0,ints, language, audiofd, ctrlfd);
|
|
|
|
if (options) {
|
|
if (!strncasecmp(options, "f",1))
|
|
mf = -1;
|
|
else if (!strncasecmp(options, "m", 1))
|
|
mf = 1;
|
|
}
|
|
|
|
while (!res && num) {
|
|
if (num < 0) {
|
|
snprintf(fn, sizeof(fn), "digits/minus");
|
|
if ( num > INT_MIN ) {
|
|
num = -num;
|
|
} else {
|
|
num = 0;
|
|
}
|
|
} else if (playa) {
|
|
snprintf(fn, sizeof(fn), "digits/and");
|
|
playa = 0;
|
|
} else if (num == 1) {
|
|
if (mf < 0)
|
|
snprintf(fn, sizeof(fn), "digits/%dF", num);
|
|
else if (mf > 0)
|
|
snprintf(fn, sizeof(fn), "digits/%dM", num);
|
|
else
|
|
snprintf(fn, sizeof(fn), "digits/%d", num);
|
|
num = 0;
|
|
} else if (num < 31) {
|
|
snprintf(fn, sizeof(fn), "digits/%d", num);
|
|
num = 0;
|
|
} else if (num < 100) {
|
|
snprintf(fn, sizeof(fn), "digits/%d", (num/10)*10);
|
|
num -= ((num/10)*10);
|
|
if (num)
|
|
playa++;
|
|
} else if (num == 100) {
|
|
snprintf(fn, sizeof(fn), "digits/100");
|
|
num = 0;
|
|
} else if (num < 200) {
|
|
snprintf(fn, sizeof(fn), "digits/100-and");
|
|
num -= 100;
|
|
} else {
|
|
if (num < 1000) {
|
|
snprintf(fn, sizeof(fn), "digits/%d", (num/100)*100);
|
|
num -= ((num/100)*100);
|
|
} else if (num < 2000) {
|
|
num = num % 1000;
|
|
snprintf(fn, sizeof(fn), "digits/thousand");
|
|
} else {
|
|
if (num < 1000000) {
|
|
res = ast_say_number_full_es(chan, num / 1000, ints, language, options, audiofd, ctrlfd);
|
|
if (res)
|
|
return res;
|
|
num = num % 1000;
|
|
snprintf(fn, sizeof(fn), "digits/thousand");
|
|
} else {
|
|
if (num < 2147483640) {
|
|
if ((num/1000000) == 1) {
|
|
res = ast_say_number_full_es(chan, num / 1000000, ints, language, "M", audiofd, ctrlfd);
|
|
if (res)
|
|
return res;
|
|
snprintf(fn, sizeof(fn), "digits/million");
|
|
} else {
|
|
res = ast_say_number_full_es(chan, num / 1000000, ints, language, options, audiofd, ctrlfd);
|
|
if (res)
|
|
return res;
|
|
snprintf(fn, sizeof(fn), "digits/millions");
|
|
}
|
|
num = num % 1000000;
|
|
} else {
|
|
ast_log(LOG_DEBUG, "Number '%d' is too big for me\n", num);
|
|
res = -1;
|
|
}
|
|
}
|
|
}
|
|
}
|
|
|
|
if (!res) {
|
|
if(!ast_streamfile(chan, fn, language)) {
|
|
if ((audiofd > -1) && (ctrlfd > -1))
|
|
res = ast_waitstream_full(chan, ints, audiofd, ctrlfd);
|
|
else
|
|
res = ast_waitstream(chan, ints);
|
|
}
|
|
ast_stopstream(chan);
|
|
|
|
}
|
|
|
|
}
|
|
return res;
|
|
}
|
|
|
|
/*--- ast_say_number_full_fr: French syntax */
|
|
/* Extra sounds needed:
|
|
1F: feminin 'une'
|
|
et: 'and' */
|
|
static int ast_say_number_full_fr(struct ast_channel *chan, int num, const char *ints, const char *language, const char *options, int audiofd, int ctrlfd)
|
|
{
|
|
int res = 0;
|
|
int playh = 0;
|
|
int playa = 0;
|
|
int mf = 1; /* +1 = male; -1 = female */
|
|
char fn[256] = "";
|
|
if (!num)
|
|
return ast_say_digits_full(chan, 0,ints, language, audiofd, ctrlfd);
|
|
|
|
if (options && !strncasecmp(options, "f",1))
|
|
mf = -1;
|
|
|
|
while(!res && (num || playh || playa)) {
|
|
if (num < 0) {
|
|
snprintf(fn, sizeof(fn), "digits/minus");
|
|
if ( num > INT_MIN ) {
|
|
num = -num;
|
|
} else {
|
|
num = 0;
|
|
}
|
|
} else if (playh) {
|
|
snprintf(fn, sizeof(fn), "digits/hundred");
|
|
playh = 0;
|
|
} else if (playa) {
|
|
snprintf(fn, sizeof(fn), "digits/et");
|
|
playa = 0;
|
|
} else if (num == 1) {
|
|
if (mf < 0)
|
|
snprintf(fn, sizeof(fn), "digits/%dF", num);
|
|
else
|
|
snprintf(fn, sizeof(fn), "digits/%d", num);
|
|
num = 0;
|
|
} else if (num < 21) {
|
|
snprintf(fn, sizeof(fn), "digits/%d", num);
|
|
num = 0;
|
|
} else if (num < 70) {
|
|
snprintf(fn, sizeof(fn), "digits/%d", (num/10)*10);
|
|
if ((num % 10) == 1) playa++;
|
|
num = num % 10;
|
|
} else if (num < 80) {
|
|
snprintf(fn, sizeof(fn), "digits/60");
|
|
if ((num % 10) == 1) playa++;
|
|
num = num - 60;
|
|
} else if (num < 100) {
|
|
snprintf(fn, sizeof(fn), "digits/80");
|
|
num = num - 80;
|
|
} else if (num < 200) {
|
|
snprintf(fn, sizeof(fn), "digits/hundred");
|
|
num = num - 100;
|
|
} else if (num < 1000) {
|
|
snprintf(fn, sizeof(fn), "digits/%d", (num/100));
|
|
playh++;
|
|
num = num % 100;
|
|
} else if (num < 2000) {
|
|
snprintf(fn, sizeof(fn), "digits/thousand");
|
|
num = num - 1000;
|
|
} else if (num < 1000000) {
|
|
res = ast_say_number_full_fr(chan, num / 1000, ints, language, options, audiofd, ctrlfd);
|
|
if (res)
|
|
return res;
|
|
snprintf(fn, sizeof(fn), "digits/thousand");
|
|
num = num % 1000;
|
|
} else if (num < 1000000000) {
|
|
res = ast_say_number_full_fr(chan, num / 1000000, ints, language, options, audiofd, ctrlfd);
|
|
if (res)
|
|
return res;
|
|
snprintf(fn, sizeof(fn), "digits/million");
|
|
num = num % 1000000;
|
|
} else {
|
|
ast_log(LOG_DEBUG, "Number '%d' is too big for me\n", num);
|
|
res = -1;
|
|
}
|
|
if (!res) {
|
|
if(!ast_streamfile(chan, fn, language)) {
|
|
if ((audiofd > -1) && (ctrlfd > -1))
|
|
res = ast_waitstream_full(chan, ints, audiofd, ctrlfd);
|
|
else
|
|
res = ast_waitstream(chan, ints);
|
|
}
|
|
ast_stopstream(chan);
|
|
}
|
|
}
|
|
return res;
|
|
}
|
|
|
|
|
|
|
|
/*--- ast_say_number_full_he: Hebrew syntax */
|
|
/* Extra sounds needed:
|
|
1F: feminin 'one'
|
|
ve: 'and'
|
|
2hundred: 2 hundred
|
|
2thousands: 2 thousand
|
|
thousands: plural of 'thousand'
|
|
3sF 'Smichut forms (female)
|
|
4sF
|
|
5sF
|
|
6sF
|
|
7sF
|
|
8sF
|
|
9sF
|
|
3s 'Smichut' forms (male)
|
|
4s
|
|
5s
|
|
6s
|
|
7s
|
|
9s
|
|
10s
|
|
11s
|
|
12s
|
|
13s
|
|
14s
|
|
15s
|
|
16s
|
|
17s
|
|
18s
|
|
19s
|
|
|
|
TODO: 've' should sometimed be 'hu':
|
|
* before 'shtaym' (2, F)
|
|
* before 'shnaym' (2, M)
|
|
* before 'shlosha' (3, M)
|
|
* before 'shmone' (8, M)
|
|
* before 'shlosim' (30)
|
|
* before 'shmonim' (80)
|
|
|
|
What about:
|
|
'sheva' (7, F)?
|
|
'tesha' (9, F)?
|
|
*/
|
|
#define SAY_NUM_BUF_SIZE 256
|
|
static int ast_say_number_full_he(struct ast_channel *chan, int num,
|
|
const char *ints, const char *language, const char *options,
|
|
int audiofd, int ctrlfd)
|
|
{
|
|
int res = 0;
|
|
int state = 0; /* no need to save anything */
|
|
int mf = 1; /* +1 = Masculin; -1 = Feminin */
|
|
char fn[SAY_NUM_BUF_SIZE] = "";
|
|
ast_verbose(VERBOSE_PREFIX_3 "ast_say_digits_full: started. "
|
|
"num: %d, options=\"%s\"\n",
|
|
num, options
|
|
);
|
|
if (!num)
|
|
return ast_say_digits_full(chan, 0,ints, language, audiofd, ctrlfd);
|
|
|
|
if (options && !strncasecmp(options, "f",1))
|
|
mf = -1;
|
|
|
|
/* Do we have work to do? */
|
|
while(!res && (num || (state>0) )) {
|
|
/* first type of work: play a second sound. In this loop
|
|
* we can only play one sound file at a time. Thus playing
|
|
* a second one requires repeating the loop just for the
|
|
* second file. The variable 'state' remembers where we were.
|
|
* state==0 is the normal mode and it means that we continue
|
|
* to check if the number num has yet anything left.
|
|
*/
|
|
ast_verbose(VERBOSE_PREFIX_3 "ast_say_digits_full: num: %d, "
|
|
"state=%d, options=\"%s\", mf=%d\n",
|
|
num, state, options, mf
|
|
);
|
|
if (state==1) {
|
|
snprintf(fn, sizeof(fn), "digits/hundred");
|
|
state = 0;
|
|
} else if (state==2) {
|
|
snprintf(fn, sizeof(fn), "digits/ve");
|
|
state = 0;
|
|
} else if (state==3) {
|
|
snprintf(fn, sizeof(fn), "digits/thousands");
|
|
state=0;
|
|
} else if (num <21) {
|
|
if (mf < 0)
|
|
snprintf(fn, sizeof(fn), "digits/%dF", num);
|
|
else
|
|
snprintf(fn, sizeof(fn), "digits/%d", num);
|
|
num = 0;
|
|
} else if (num < 100) {
|
|
snprintf(fn, sizeof(fn), "digits/%d", (num/10)*10);
|
|
num = num % 10;
|
|
if (num>0) state=2;
|
|
} else if (num < 200) {
|
|
snprintf(fn, sizeof(fn), "digits/hundred");
|
|
num = num - 100;
|
|
} else if (num < 300) {
|
|
snprintf(fn, sizeof(fn), "digits/hundred");
|
|
num = num - 100;
|
|
} else if (num < 1000) {
|
|
snprintf(fn, sizeof(fn), "digits/%d", (num/100));
|
|
state=1;
|
|
num = num % 100;
|
|
} else if (num < 2000) {
|
|
snprintf(fn, sizeof(fn), "digits/thousand");
|
|
num = num - 1000;
|
|
} else if (num < 3000) {
|
|
snprintf(fn, sizeof(fn), "digits/2thousand");
|
|
num = num - 2000;
|
|
if (num>0) state=2;
|
|
} else if (num < 20000) {
|
|
snprintf(fn, sizeof(fn), "digits/%ds",(num/1000));
|
|
num = num % 1000;
|
|
state=3;
|
|
} else if (num < 1000000) {
|
|
res = ast_say_number_full_he(chan, num / 1000, ints, language, options, audiofd, ctrlfd);
|
|
if (res)
|
|
return res;
|
|
snprintf(fn, sizeof(fn), "digits/thousand");
|
|
num = num % 1000;
|
|
} else if (num < 1000000000) {
|
|
res = ast_say_number_full_he(chan, num / 1000000, ints, language, options, audiofd, ctrlfd);
|
|
if (res)
|
|
return res;
|
|
snprintf(fn, sizeof(fn), "digits/million");
|
|
num = num % 1000000;
|
|
} else {
|
|
ast_log(LOG_DEBUG, "Number '%d' is too big for me\n", num);
|
|
res = -1;
|
|
}
|
|
if (!res) {
|
|
if(!ast_streamfile(chan, fn, language)) {
|
|
if ((audiofd > -1) && (ctrlfd > -1))
|
|
res = ast_waitstream_full(chan, ints, audiofd, ctrlfd);
|
|
else
|
|
res = ast_waitstream(chan, ints);
|
|
}
|
|
ast_stopstream(chan);
|
|
}
|
|
}
|
|
return res;
|
|
}
|
|
|
|
/*--- ast_say_number_full_it: Italian */
|
|
static int ast_say_number_full_it(struct ast_channel *chan, int num, const char *ints, const char *language, int audiofd, int ctrlfd)
|
|
{
|
|
int res = 0;
|
|
int playh = 0;
|
|
int tempnum = 0;
|
|
char fn[256] = "";
|
|
|
|
if (!num)
|
|
return ast_say_digits_full(chan, 0,ints, language, audiofd, ctrlfd);
|
|
|
|
/*
|
|
Italian support
|
|
|
|
Like english, numbers up to 20 are a single 'word', and others
|
|
compound, but with exceptions.
|
|
For example 21 is not twenty-one, but there is a single word in 'it'.
|
|
Idem for 28 (ie when a the 2nd part of a compund number
|
|
starts with a vowel)
|
|
|
|
There are exceptions also for hundred, thousand and million.
|
|
In english 100 = one hundred, 200 is two hundred.
|
|
In italian 100 = cento , like to say hundred (without one),
|
|
200 and more are like english.
|
|
|
|
Same applies for thousand:
|
|
1000 is one thousand in en, 2000 is two thousand.
|
|
In it we have 1000 = mille , 2000 = 2 mila
|
|
|
|
For million(s) we use the plural, if more than one
|
|
Also, one million is abbreviated in it, like on-million,
|
|
or 'un milione', not 'uno milione'.
|
|
So the right file is provided.
|
|
*/
|
|
|
|
while(!res && (num || playh)) {
|
|
if (num < 0) {
|
|
snprintf(fn, sizeof(fn), "digits/minus");
|
|
if ( num > INT_MIN ) {
|
|
num = -num;
|
|
} else {
|
|
num = 0;
|
|
}
|
|
} else if (playh) {
|
|
snprintf(fn, sizeof(fn), "digits/hundred");
|
|
playh = 0;
|
|
} else if (num < 20) {
|
|
snprintf(fn, sizeof(fn), "digits/%d", num);
|
|
num = 0;
|
|
} else if (num == 21) {
|
|
snprintf(fn, sizeof(fn), "digits/%d", num);
|
|
num = 0;
|
|
} else if (num == 28) {
|
|
snprintf(fn, sizeof(fn), "digits/%d", num);
|
|
num = 0;
|
|
} else if (num == 31) {
|
|
snprintf(fn, sizeof(fn), "digits/%d", num);
|
|
num = 0;
|
|
} else if (num == 38) {
|
|
snprintf(fn, sizeof(fn), "digits/%d", num);
|
|
num = 0;
|
|
} else if (num == 41) {
|
|
snprintf(fn, sizeof(fn), "digits/%d", num);
|
|
num = 0;
|
|
} else if (num == 48) {
|
|
snprintf(fn, sizeof(fn), "digits/%d", num);
|
|
num = 0;
|
|
} else if (num == 51) {
|
|
snprintf(fn, sizeof(fn), "digits/%d", num);
|
|
num = 0;
|
|
} else if (num == 58) {
|
|
snprintf(fn, sizeof(fn), "digits/%d", num);
|
|
num = 0;
|
|
} else if (num == 61) {
|
|
snprintf(fn, sizeof(fn), "digits/%d", num);
|
|
num = 0;
|
|
} else if (num == 68) {
|
|
snprintf(fn, sizeof(fn), "digits/%d", num);
|
|
num = 0;
|
|
} else if (num == 71) {
|
|
snprintf(fn, sizeof(fn), "digits/%d", num);
|
|
num = 0;
|
|
} else if (num == 78) {
|
|
snprintf(fn, sizeof(fn), "digits/%d", num);
|
|
num = 0;
|
|
} else if (num == 81) {
|
|
snprintf(fn, sizeof(fn), "digits/%d", num);
|
|
num = 0;
|
|
} else if (num == 88) {
|
|
snprintf(fn, sizeof(fn), "digits/%d", num);
|
|
num = 0;
|
|
} else if (num == 91) {
|
|
snprintf(fn, sizeof(fn), "digits/%d", num);
|
|
num = 0;
|
|
} else if (num == 98) {
|
|
snprintf(fn, sizeof(fn), "digits/%d", num);
|
|
num = 0;
|
|
} else if (num < 100) {
|
|
snprintf(fn, sizeof(fn), "digits/%d", (num /10) * 10);
|
|
num -= ((num / 10) * 10);
|
|
} else {
|
|
if (num < 1000) {
|
|
if ((num / 100) > 1) {
|
|
snprintf(fn, sizeof(fn), "digits/%d", (num/100));
|
|
playh++;
|
|
} else {
|
|
snprintf(fn, sizeof(fn), "digits/hundred");
|
|
}
|
|
num -= ((num / 100) * 100);
|
|
} else {
|
|
if (num < 1000000) { /* 1,000,000 */
|
|
if ((num/1000) > 1)
|
|
res = ast_say_number_full_it(chan, num / 1000, ints, language, audiofd, ctrlfd);
|
|
if (res)
|
|
return res;
|
|
tempnum = num;
|
|
num = num % 1000;
|
|
if ((tempnum / 1000) < 2)
|
|
snprintf(fn, sizeof(fn), "digits/thousand");
|
|
else /* for 1000 it says mille, for >1000 (eg 2000) says mila */
|
|
snprintf(fn, sizeof(fn), "digits/thousands");
|
|
} else {
|
|
if (num < 1000000000) { /* 1,000,000,000 */
|
|
if ((num / 1000000) > 1)
|
|
res = ast_say_number_full_it(chan, num / 1000000, ints, language, audiofd, ctrlfd);
|
|
if (res)
|
|
return res;
|
|
tempnum = num;
|
|
num = num % 1000000;
|
|
if ((tempnum / 1000000) < 2)
|
|
snprintf(fn, sizeof(fn), "digits/million");
|
|
else
|
|
snprintf(fn, sizeof(fn), "digits/millions");
|
|
} else {
|
|
ast_log(LOG_DEBUG, "Number '%d' is too big for me\n", num);
|
|
res = -1;
|
|
}
|
|
}
|
|
}
|
|
}
|
|
if (!res) {
|
|
if(!ast_streamfile(chan, fn, language)) {
|
|
if ((audiofd > -1) && (ctrlfd > -1))
|
|
res = ast_waitstream_full(chan, ints, audiofd, ctrlfd);
|
|
else
|
|
res = ast_waitstream(chan, ints);
|
|
}
|
|
ast_stopstream(chan);
|
|
}
|
|
}
|
|
return res;
|
|
}
|
|
|
|
/*--- ast_say_number_full_nl: dutch syntax */
|
|
/* New files: digits/nl-en
|
|
*/
|
|
static int ast_say_number_full_nl(struct ast_channel *chan, int num, const char *ints, const char *language, int audiofd, int ctrlfd)
|
|
{
|
|
int res = 0;
|
|
int playh = 0;
|
|
int units = 0;
|
|
char fn[256] = "";
|
|
if (!num)
|
|
return ast_say_digits_full(chan, 0,ints, language, audiofd, ctrlfd);
|
|
while (!res && (num || playh )) {
|
|
if (num < 0) {
|
|
snprintf(fn, sizeof(fn), "digits/minus");
|
|
if ( num > INT_MIN ) {
|
|
num = -num;
|
|
} else {
|
|
num = 0;
|
|
}
|
|
} else if (playh) {
|
|
snprintf(fn, sizeof(fn), "digits/hundred");
|
|
playh = 0;
|
|
} else if (num < 20) {
|
|
snprintf(fn, sizeof(fn), "digits/%d", num);
|
|
num = 0;
|
|
} else if (num < 100) {
|
|
units = num % 10;
|
|
if (units > 0) {
|
|
res = ast_say_number_full_nl(chan, units, ints, language, audiofd, ctrlfd);
|
|
if (res)
|
|
return res;
|
|
num = num - units;
|
|
snprintf(fn, sizeof(fn), "digits/nl-en");
|
|
} else {
|
|
snprintf(fn, sizeof(fn), "digits/%d", num - units);
|
|
num = 0;
|
|
}
|
|
} else {
|
|
if (num < 1000) {
|
|
snprintf(fn, sizeof(fn), "digits/%d", (num/100));
|
|
playh++;
|
|
num -= ((num / 100) * 100);
|
|
} else {
|
|
if (num < 1000000) { /* 1,000,000 */
|
|
res = ast_say_number_full_en(chan, num / 1000, ints, language, audiofd, ctrlfd);
|
|
if (res)
|
|
return res;
|
|
num = num % 1000;
|
|
snprintf(fn, sizeof(fn), "digits/thousand");
|
|
} else {
|
|
if (num < 1000000000) { /* 1,000,000,000 */
|
|
res = ast_say_number_full_en(chan, num / 1000000, ints, language, audiofd, ctrlfd);
|
|
if (res)
|
|
return res;
|
|
num = num % 1000000;
|
|
snprintf(fn, sizeof(fn), "digits/million");
|
|
} else {
|
|
ast_log(LOG_DEBUG, "Number '%d' is too big for me\n", num);
|
|
res = -1;
|
|
}
|
|
}
|
|
}
|
|
}
|
|
|
|
if (!res) {
|
|
if(!ast_streamfile(chan, fn, language)) {
|
|
if ((audiofd > -1) && (ctrlfd > -1))
|
|
res = ast_waitstream_full(chan, ints, audiofd, ctrlfd);
|
|
else
|
|
res = ast_waitstream(chan, ints);
|
|
}
|
|
ast_stopstream(chan);
|
|
}
|
|
}
|
|
return res;
|
|
}
|
|
|
|
/*--- ast_say_number_full_no: Norwegian syntax */
|
|
/* New files:
|
|
In addition to American English, the following sounds are required: "and", "1N"
|
|
*/
|
|
static int ast_say_number_full_no(struct ast_channel *chan, int num, const char *ints, const char *language, const char *options, int audiofd, int ctrlfd)
|
|
{
|
|
int res = 0;
|
|
int playh = 0;
|
|
int playa = 0;
|
|
int cn = 1; /* +1 = commune; -1 = neuter */
|
|
char fn[256] = "";
|
|
|
|
if (!num)
|
|
return ast_say_digits_full(chan, 0,ints, language, audiofd, ctrlfd);
|
|
|
|
if (options && !strncasecmp(options, "n",1)) cn = -1;
|
|
|
|
while(!res && (num || playh || playa )) {
|
|
/* The grammar for Norwegian numbers is the same as for English except
|
|
* for the following:
|
|
* - 1 exists in both commune ("en", file "1") and neuter ("ett", file "1N")
|
|
* "and" before the last two digits, i.e. 2034 is "two thousand and
|
|
* thirty-four" and 1000012 is "one million and twelve".
|
|
*/
|
|
if (num < 0) {
|
|
snprintf(fn, sizeof(fn), "digits/minus");
|
|
if ( num > INT_MIN ) {
|
|
num = -num;
|
|
} else {
|
|
num = 0;
|
|
}
|
|
} else if (playh) {
|
|
snprintf(fn, sizeof(fn), "digits/hundred");
|
|
playh = 0;
|
|
} else if (playa) {
|
|
snprintf(fn, sizeof(fn), "digits/and");
|
|
playa = 0;
|
|
} else if (num == 1 && cn == -1) {
|
|
snprintf(fn, sizeof(fn), "digits/1N");
|
|
num = 0;
|
|
} else if (num < 20) {
|
|
snprintf(fn, sizeof(fn), "digits/%d", num);
|
|
num = 0;
|
|
} else if (num < 100) {
|
|
snprintf(fn, sizeof(fn), "digits/%d", (num /10) * 10);
|
|
num -= ((num / 10) * 10);
|
|
} else if (num < 1000) {
|
|
int hundreds = num / 100;
|
|
if (hundreds == 1)
|
|
snprintf(fn, sizeof(fn), "digits/1N");
|
|
else
|
|
snprintf(fn, sizeof(fn), "digits/%d", (num / 100));
|
|
|
|
playh++;
|
|
num -= 100 * hundreds;
|
|
if (num)
|
|
playa++;
|
|
} else if (num < 1000000) {
|
|
res = ast_say_number_full_no(chan, num / 1000, ints, language, "n", audiofd, ctrlfd);
|
|
if (res)
|
|
return res;
|
|
snprintf(fn, sizeof(fn), "digits/thousand");
|
|
num = num % 1000;
|
|
if (num && num < 100)
|
|
playa++;
|
|
} else if (num < 1000000000) {
|
|
int millions = num / 1000000;
|
|
res = ast_say_number_full_no(chan, millions, ints, language, "c", audiofd, ctrlfd);
|
|
if (res)
|
|
return res;
|
|
snprintf(fn, sizeof(fn), "digits/million");
|
|
num = num % 1000000;
|
|
if (num && num < 100)
|
|
playa++;
|
|
} else {
|
|
ast_log(LOG_DEBUG, "Number '%d' is too big for me\n", num);
|
|
res = -1;
|
|
}
|
|
|
|
if (!res) {
|
|
if(!ast_streamfile(chan, fn, language)) {
|
|
if ((audiofd > -1) && (ctrlfd > -1))
|
|
res = ast_waitstream_full(chan, ints, audiofd, ctrlfd);
|
|
else
|
|
res = ast_waitstream(chan, ints);
|
|
}
|
|
ast_stopstream(chan);
|
|
}
|
|
}
|
|
return res;
|
|
}
|
|
|
|
typedef struct {
|
|
char *separator_dziesiatek;
|
|
char *cyfry[10];
|
|
char *cyfry2[10];
|
|
char *setki[10];
|
|
char *dziesiatki[10];
|
|
char *nastki[10];
|
|
char *rzedy[3][3];
|
|
} odmiana;
|
|
|
|
static char *pl_rzad_na_tekst(odmiana *odm, int i, int rzad)
|
|
{
|
|
if (rzad==0)
|
|
return "";
|
|
|
|
if (i==1)
|
|
return odm->rzedy[rzad - 1][0];
|
|
if ((i > 21 || i < 11) && i%10 > 1 && i%10 < 5)
|
|
return odm->rzedy[rzad - 1][1];
|
|
else
|
|
return odm->rzedy[rzad - 1][2];
|
|
}
|
|
|
|
static char* pl_append(char* buffer, char* str)
|
|
{
|
|
strcpy(buffer, str);
|
|
buffer += strlen(str);
|
|
return buffer;
|
|
}
|
|
|
|
static void pl_odtworz_plik(struct ast_channel *chan, const char *language, int audiofd, int ctrlfd, const char *ints, char *fn)
|
|
{
|
|
char file_name[255] = "digits/";
|
|
strcat(file_name, fn);
|
|
ast_log(LOG_DEBUG, "Trying to play: %s\n", file_name);
|
|
if (!ast_streamfile(chan, file_name, language)) {
|
|
if ((audiofd > -1) && (ctrlfd > -1))
|
|
ast_waitstream_full(chan, ints, audiofd, ctrlfd);
|
|
else
|
|
ast_waitstream(chan, ints);
|
|
}
|
|
ast_stopstream(chan);
|
|
}
|
|
|
|
static void powiedz(struct ast_channel *chan, const char *language, int audiofd, int ctrlfd, const char *ints, odmiana *odm, int rzad, int i)
|
|
{
|
|
/* Initialise variables to allow compilation on Debian-stable, etc */
|
|
int m1000E6 = 0;
|
|
int i1000E6 = 0;
|
|
int m1000E3 = 0;
|
|
int i1000E3 = 0;
|
|
int m1000 = 0;
|
|
int i1000 = 0;
|
|
int m100 = 0;
|
|
int i100 = 0;
|
|
|
|
if (i == 0 && rzad > 0) {
|
|
return;
|
|
}
|
|
if (i == 0) {
|
|
pl_odtworz_plik(chan, language, audiofd, ctrlfd, ints, odm->cyfry[0]);
|
|
}
|
|
|
|
m1000E6 = i % 1000000000;
|
|
i1000E6 = i / 1000000000;
|
|
|
|
powiedz(chan, language, audiofd, ctrlfd, ints, odm, rzad+3, i1000E6);
|
|
|
|
m1000E3 = m1000E6 % 1000000;
|
|
i1000E3 = m1000E6 / 1000000;
|
|
|
|
powiedz(chan, language, audiofd, ctrlfd, ints, odm, rzad+2, i1000E3);
|
|
|
|
m1000 = m1000E3 % 1000;
|
|
i1000 = m1000E3 / 1000;
|
|
|
|
powiedz(chan, language, audiofd, ctrlfd, ints, odm, rzad+1, i1000);
|
|
|
|
m100 = m1000 % 100;
|
|
i100 = m1000 / 100;
|
|
|
|
if (i100>0)
|
|
pl_odtworz_plik(chan, language, audiofd, ctrlfd, ints, odm->setki[i100]);
|
|
|
|
if ( m100 > 0 && m100 <=9 ) {
|
|
if (m1000>0)
|
|
pl_odtworz_plik(chan, language, audiofd, ctrlfd, ints, odm->cyfry2[m100]);
|
|
else
|
|
pl_odtworz_plik(chan, language, audiofd, ctrlfd, ints, odm->cyfry[m100]);
|
|
} else if (m100 % 10 == 0) {
|
|
pl_odtworz_plik(chan, language, audiofd, ctrlfd, ints, odm->dziesiatki[m100 / 10]);
|
|
} else if (m100 <= 19 ) {
|
|
pl_odtworz_plik(chan, language, audiofd, ctrlfd, ints, odm->nastki[m100 % 10]);
|
|
} else if (m100 != 0) {
|
|
if (odm->separator_dziesiatek[0]==' ') {
|
|
pl_odtworz_plik(chan, language, audiofd, ctrlfd, ints, odm->dziesiatki[m100 / 10]);
|
|
pl_odtworz_plik(chan, language, audiofd, ctrlfd, ints, odm->cyfry2[m100 % 10]);
|
|
} else {
|
|
char buf[10];
|
|
char *b = buf;
|
|
b = pl_append(b, odm->dziesiatki[m100 / 10]);
|
|
b = pl_append(b, odm->separator_dziesiatek);
|
|
b = pl_append(b, odm->cyfry2[m100 % 10]);
|
|
pl_odtworz_plik(chan, language, audiofd, ctrlfd, ints, buf);
|
|
}
|
|
}
|
|
|
|
if (rzad > 0) {
|
|
pl_odtworz_plik(chan, language, audiofd, ctrlfd, ints, pl_rzad_na_tekst(odm, i, rzad));
|
|
}
|
|
}
|
|
|
|
/* ast_say_number_full_pl: Polish syntax */
|
|
static int ast_say_number_full_pl(struct ast_channel *chan, int num, const char *ints, const char *language, const char *options, int audiofd, int ctrlfd)
|
|
/*
|
|
Sounds needed:
|
|
0 zero
|
|
1 jeden
|
|
10 dziesiec
|
|
100 sto
|
|
1000 tysiac
|
|
1000000 milion
|
|
1000000000 miliard
|
|
1000000000.2 miliardy
|
|
1000000000.5 miliardow
|
|
1000000.2 miliony
|
|
1000000.5 milionow
|
|
1000.2 tysiace
|
|
1000.5 tysiecy
|
|
100m stu
|
|
10m dziesieciu
|
|
11 jedenascie
|
|
11m jedenastu
|
|
12 dwanascie
|
|
12m dwunastu
|
|
13 trzynascie
|
|
13m trzynastu
|
|
14 czternascie
|
|
14m czternastu
|
|
15 pietnascie
|
|
15m pietnastu
|
|
16 szesnascie
|
|
16m szesnastu
|
|
17 siedemnascie
|
|
17m siedemnastu
|
|
18 osiemnascie
|
|
18m osiemnastu
|
|
19 dziewietnascie
|
|
19m dziewietnastu
|
|
1z jedna
|
|
2 dwie
|
|
20 dwadziescia
|
|
200 dwiescie
|
|
200m dwustu
|
|
20m dwudziestu
|
|
2-1m dwaj
|
|
2-2m dwoch
|
|
2z dwie
|
|
3 trzy
|
|
30 trzydziesci
|
|
300 trzysta
|
|
300m trzystu
|
|
30m trzydziestu
|
|
3-1m trzej
|
|
3-2m trzech
|
|
4 cztery
|
|
40 czterdziesci
|
|
400 czterysta
|
|
400m czterystu
|
|
40m czterdziestu
|
|
4-1m czterej
|
|
4-2m czterech
|
|
5 piec
|
|
50 piecdziesiat
|
|
500 piecset
|
|
500m pieciuset
|
|
50m piedziesieciu
|
|
5m pieciu
|
|
6 szesc
|
|
60 szescdziesiat
|
|
600 szescset
|
|
600m szesciuset
|
|
60m szescdziesieciu
|
|
6m szesciu
|
|
7 siedem
|
|
70 siedemdziesiat
|
|
700 siedemset
|
|
700m siedmiuset
|
|
70m siedemdziesieciu
|
|
7m siedmiu
|
|
8 osiem
|
|
80 osiemdziesiat
|
|
800 osiemset
|
|
800m osmiuset
|
|
80m osiemdziesieciu
|
|
8m osmiu
|
|
9 dziewiec
|
|
90 dziewiecdziesiat
|
|
900 dziewiecset
|
|
900m dziewieciuset
|
|
90m dziewiedziesieciu
|
|
9m dziewieciu
|
|
and combinations of eg.: 20_1, 30m_3m, etc...
|
|
|
|
*/
|
|
{
|
|
char *zenski_cyfry[] = {"0","1z", "2z", "3", "4", "5", "6", "7", "8", "9"};
|
|
|
|
char *zenski_cyfry2[] = {"0","1", "2z", "3", "4", "5", "6", "7", "8", "9"};
|
|
|
|
char *meski_cyfry[] = {"0","1", "2-1m", "3-1m", "4-1m", "5m", /*"2-1mdwaj"*/ "6m", "7m", "8m", "9m"};
|
|
|
|
char *meski_cyfry2[] = {"0","1", "2-2m", "3-2m", "4-2m", "5m", "6m", "7m", "8m", "9m"};
|
|
|
|
char *meski_setki[] = {"", "100m", "200m", "300m", "400m", "500m", "600m", "700m", "800m", "900m"};
|
|
|
|
char *meski_dziesiatki[] = {"", "10m", "20m", "30m", "40m", "50m", "60m", "70m", "80m", "90m"};
|
|
|
|
char *meski_nastki[] = {"", "11m", "12m", "13m", "14m", "15m", "16m", "17m", "18m", "19m"};
|
|
|
|
char *nijaki_cyfry[] = {"0","1", "2", "3", "4", "5", "6", "7", "8", "9"};
|
|
|
|
char *nijaki_cyfry2[] = {"0","1", "2", "3", "4", "5", "6", "7", "8", "9"};
|
|
|
|
char *nijaki_setki[] = {"", "100", "200", "300", "400", "500", "600", "700", "800", "900"};
|
|
|
|
char *nijaki_dziesiatki[] = {"", "10", "20", "30", "40", "50", "60", "70", "80", "90"};
|
|
|
|
char *nijaki_nastki[] = {"", "11", "12", "13", "14", "15", "16", "17", "18", "19"};
|
|
|
|
char *rzedy[][3] = { {"1000", "1000.2", "1000.5"}, {"1000000", "1000000.2", "1000000.5"}, {"1000000000", "1000000000.2", "1000000000.5"}};
|
|
|
|
/* Initialise variables to allow compilation on Debian-stable, etc */
|
|
odmiana *o;
|
|
|
|
static odmiana *odmiana_nieosobowa = NULL;
|
|
static odmiana *odmiana_meska = NULL;
|
|
static odmiana *odmiana_zenska = NULL;
|
|
|
|
if (odmiana_nieosobowa == NULL) {
|
|
odmiana_nieosobowa = (odmiana *) malloc(sizeof(odmiana));
|
|
|
|
odmiana_nieosobowa->separator_dziesiatek = "_";
|
|
|
|
memcpy(odmiana_nieosobowa->cyfry, nijaki_cyfry, sizeof(odmiana_nieosobowa->cyfry));
|
|
memcpy(odmiana_nieosobowa->cyfry2, nijaki_cyfry2, sizeof(odmiana_nieosobowa->cyfry));
|
|
memcpy(odmiana_nieosobowa->setki, nijaki_setki, sizeof(odmiana_nieosobowa->setki));
|
|
memcpy(odmiana_nieosobowa->dziesiatki, nijaki_dziesiatki, sizeof(odmiana_nieosobowa->dziesiatki));
|
|
memcpy(odmiana_nieosobowa->nastki, nijaki_nastki, sizeof(odmiana_nieosobowa->nastki));
|
|
memcpy(odmiana_nieosobowa->rzedy, rzedy, sizeof(odmiana_nieosobowa->rzedy));
|
|
}
|
|
|
|
if (odmiana_zenska == NULL) {
|
|
odmiana_zenska = (odmiana *) malloc(sizeof(odmiana));
|
|
|
|
odmiana_zenska->separator_dziesiatek = " ";
|
|
|
|
memcpy(odmiana_zenska->cyfry, zenski_cyfry, sizeof(odmiana_zenska->cyfry));
|
|
memcpy(odmiana_zenska->cyfry2, zenski_cyfry2, sizeof(odmiana_zenska->cyfry));
|
|
memcpy(odmiana_zenska->setki, nijaki_setki, sizeof(odmiana_zenska->setki));
|
|
memcpy(odmiana_zenska->dziesiatki, nijaki_dziesiatki, sizeof(odmiana_zenska->dziesiatki));
|
|
memcpy(odmiana_zenska->nastki, nijaki_nastki, sizeof(odmiana_zenska->nastki));
|
|
memcpy(odmiana_zenska->rzedy, rzedy, sizeof(odmiana_zenska->rzedy));
|
|
}
|
|
|
|
if (odmiana_meska == NULL) {
|
|
odmiana_meska = (odmiana *) malloc(sizeof(odmiana));
|
|
|
|
odmiana_meska->separator_dziesiatek = " ";
|
|
|
|
memcpy(odmiana_meska->cyfry, meski_cyfry, sizeof(odmiana_meska->cyfry));
|
|
memcpy(odmiana_meska->cyfry2, meski_cyfry2, sizeof(odmiana_meska->cyfry));
|
|
memcpy(odmiana_meska->setki, meski_setki, sizeof(odmiana_meska->setki));
|
|
memcpy(odmiana_meska->dziesiatki, meski_dziesiatki, sizeof(odmiana_meska->dziesiatki));
|
|
memcpy(odmiana_meska->nastki, meski_nastki, sizeof(odmiana_meska->nastki));
|
|
memcpy(odmiana_meska->rzedy, rzedy, sizeof(odmiana_meska->rzedy));
|
|
}
|
|
|
|
if (options) {
|
|
if (strncasecmp(options, "f", 1) == 0)
|
|
o = odmiana_zenska;
|
|
else if (strncasecmp(options, "m", 1) == 0)
|
|
o = odmiana_meska;
|
|
else
|
|
o = odmiana_nieosobowa;
|
|
} else
|
|
o = odmiana_nieosobowa;
|
|
|
|
powiedz(chan, language, audiofd, ctrlfd, ints, o, 0, num);
|
|
return 0;
|
|
}
|
|
|
|
/* ast_say_number_full_pt: Portuguese syntax */
|
|
/* Extra sounds needed: */
|
|
/* For feminin all sound files end with F */
|
|
/* 100E for 100+ something */
|
|
/* 1000000S for plural */
|
|
/* pt-e for 'and' */
|
|
static int ast_say_number_full_pt(struct ast_channel *chan, int num, const char *ints, const char *language, const char *options, int audiofd, int ctrlfd)
|
|
{
|
|
int res = 0;
|
|
int playh = 0;
|
|
int mf = 1; /* +1 = male; -1 = female */
|
|
char fn[256] = "";
|
|
|
|
if (!num)
|
|
return ast_say_digits_full(chan, 0,ints, language, audiofd, ctrlfd);
|
|
|
|
if (options && !strncasecmp(options, "f",1))
|
|
mf = -1;
|
|
|
|
while(!res && num ) {
|
|
if (num < 0) {
|
|
snprintf(fn, sizeof(fn), "digits/minus");
|
|
if ( num > INT_MIN ) {
|
|
num = -num;
|
|
} else {
|
|
num = 0;
|
|
}
|
|
} else if (num < 20) {
|
|
if ((num == 1 || num == 2) && (mf < 0))
|
|
snprintf(fn, sizeof(fn), "digits/%dF", num);
|
|
else
|
|
snprintf(fn, sizeof(fn), "digits/%d", num);
|
|
num = 0;
|
|
} else if (num < 100) {
|
|
snprintf(fn, sizeof(fn), "digits/%d", (num / 10) * 10);
|
|
if (num % 10)
|
|
playh = 1;
|
|
num = num % 10;
|
|
} else if (num < 1000) {
|
|
if (num == 100)
|
|
snprintf(fn, sizeof(fn), "digits/100");
|
|
else if (num < 200)
|
|
snprintf(fn, sizeof(fn), "digits/100E");
|
|
else {
|
|
if (mf < 0 && num > 199)
|
|
snprintf(fn, sizeof(fn), "digits/%dF", (num / 100) * 100);
|
|
else
|
|
snprintf(fn, sizeof(fn), "digits/%d", (num / 100) * 100);
|
|
if (num % 100)
|
|
playh = 1;
|
|
}
|
|
num = num % 100;
|
|
} else if (num < 1000000) {
|
|
if (num > 1999) {
|
|
res = ast_say_number_full_pt(chan, (num / 1000) * mf, ints, language, options, audiofd, ctrlfd);
|
|
if (res)
|
|
return res;
|
|
}
|
|
snprintf(fn, sizeof(fn), "digits/1000");
|
|
if ((num % 1000) && ((num % 1000) < 100 || !(num % 100)))
|
|
playh = 1;
|
|
num = num % 1000;
|
|
} else if (num < 1000000000) {
|
|
res = ast_say_number_full_pt(chan, (num / 1000000), ints, language, options, audiofd, ctrlfd );
|
|
if (res)
|
|
return res;
|
|
if (num < 2000000)
|
|
snprintf(fn, sizeof(fn), "digits/1000000");
|
|
else
|
|
snprintf(fn, sizeof(fn), "digits/1000000S");
|
|
|
|
if ((num % 1000000) &&
|
|
/* no thousands */
|
|
((!((num / 1000) % 1000) && ((num % 1000) < 100 || !(num % 100))) ||
|
|
/* no hundreds and below */
|
|
(!(num % 1000) && (((num / 1000) % 1000) < 100 || !((num / 1000) % 100))) ) )
|
|
playh = 1;
|
|
num = num % 1000000;
|
|
}
|
|
if (!res) {
|
|
if (!ast_streamfile(chan, fn, language)) {
|
|
if ((audiofd > -1) && (ctrlfd > -1))
|
|
res = ast_waitstream_full(chan, ints, audiofd, ctrlfd);
|
|
else
|
|
res = ast_waitstream(chan, ints);
|
|
}
|
|
ast_stopstream(chan);
|
|
}
|
|
if (!res && playh) {
|
|
res = wait_file(chan, ints, "digits/pt-e", language);
|
|
ast_stopstream(chan);
|
|
playh = 0;
|
|
}
|
|
}
|
|
return res;
|
|
}
|
|
|
|
/*--- ast_say_number_full_se: Swedish syntax */
|
|
static int ast_say_number_full_se(struct ast_channel *chan, int num, const char *ints, const char *language, const char *options, int audiofd, int ctrlfd)
|
|
{
|
|
int res = 0;
|
|
int playh = 0;
|
|
char fn[256] = "";
|
|
int cn = 1; /* +1 = commune; -1 = neuter */
|
|
if (!num)
|
|
return ast_say_digits_full(chan, 0,ints, language, audiofd, ctrlfd);
|
|
if (options && !strncasecmp(options, "n",1)) cn = -1;
|
|
|
|
while(!res && (num || playh)) {
|
|
if (num < 0) {
|
|
snprintf(fn, sizeof(fn), "digits/minus");
|
|
if ( num > INT_MIN ) {
|
|
num = -num;
|
|
} else {
|
|
num = 0;
|
|
}
|
|
} else if (playh) {
|
|
snprintf(fn, sizeof(fn), "digits/hundred");
|
|
playh = 0;
|
|
} else if (num < 20) {
|
|
snprintf(fn, sizeof(fn), "digits/%d", num);
|
|
num = 0;
|
|
} else if (num < 100) {
|
|
snprintf(fn, sizeof(fn), "digits/%d", (num /10) * 10);
|
|
num -= ((num / 10) * 10);
|
|
} else if (num == 1 && cn == -1) { /* En eller ett? */
|
|
snprintf(fn, sizeof(fn), "digits/1N");
|
|
num = 0;
|
|
} else {
|
|
if (num < 1000){
|
|
snprintf(fn, sizeof(fn), "digits/%d", (num/100));
|
|
playh++;
|
|
num -= ((num / 100) * 100);
|
|
} else {
|
|
if (num < 1000000) { /* 1,000,000 */
|
|
res = ast_say_number_full_se(chan, num / 1000, ints, language, options, audiofd, ctrlfd);
|
|
if (res) {
|
|
return res;
|
|
}
|
|
num = num % 1000;
|
|
snprintf(fn, sizeof(fn), "digits/thousand");
|
|
} else {
|
|
if (num < 1000000000) { /* 1,000,000,000 */
|
|
res = ast_say_number_full_se(chan, num / 1000000, ints, language, options, audiofd, ctrlfd);
|
|
if (res) {
|
|
return res;
|
|
}
|
|
num = num % 1000000;
|
|
snprintf(fn, sizeof(fn), "digits/million");
|
|
} else {
|
|
ast_log(LOG_DEBUG, "Number '%d' is too big for me\n", num);
|
|
res = -1;
|
|
}
|
|
}
|
|
}
|
|
}
|
|
if (!res) {
|
|
if(!ast_streamfile(chan, fn, language)) {
|
|
if ((audiofd > -1) && (ctrlfd > -1))
|
|
res = ast_waitstream_full(chan, ints, audiofd, ctrlfd);
|
|
else
|
|
res = ast_waitstream(chan, ints);
|
|
ast_stopstream(chan);
|
|
}
|
|
}
|
|
}
|
|
return res;
|
|
}
|
|
|
|
/*--- ast_say_number_full_tw: Taiwanese syntax */
|
|
static int ast_say_number_full_tw(struct ast_channel *chan, int num, const char *ints, const char *language, int audiofd, int ctrlfd)
|
|
{
|
|
int res = 0;
|
|
int playh = 0;
|
|
char fn[256] = "";
|
|
if (!num)
|
|
return ast_say_digits_full(chan, 0,ints, language, audiofd, ctrlfd);
|
|
|
|
while(!res && (num || playh)) {
|
|
if (num < 0) {
|
|
snprintf(fn, sizeof(fn), "digits/minus");
|
|
if ( num > INT_MIN ) {
|
|
num = -num;
|
|
} else {
|
|
num = 0;
|
|
}
|
|
} else if (playh) {
|
|
snprintf(fn, sizeof(fn), "digits/hundred");
|
|
playh = 0;
|
|
} else if (num < 10) {
|
|
snprintf(fn, sizeof(fn), "digits/%d", num);
|
|
num = 0;
|
|
} else if (num < 100) {
|
|
snprintf(fn, sizeof(fn), "digits/%d", (num /10) * 10);
|
|
num -= ((num / 10) * 10);
|
|
} else {
|
|
if (num < 1000){
|
|
snprintf(fn, sizeof(fn), "digits/%d", (num/100));
|
|
playh++;
|
|
num -= ((num / 100) * 100);
|
|
} else {
|
|
if (num < 1000000) { /* 1,000,000 */
|
|
res = ast_say_number_full_tw(chan, num / 1000, ints, language, audiofd, ctrlfd);
|
|
if (res)
|
|
return res;
|
|
num = num % 1000;
|
|
snprintf(fn, sizeof(fn), "digits/thousand");
|
|
} else {
|
|
if (num < 1000000000) { /* 1,000,000,000 */
|
|
res = ast_say_number_full_tw(chan, num / 1000000, ints, language, audiofd, ctrlfd);
|
|
if (res)
|
|
return res;
|
|
num = num % 1000000;
|
|
snprintf(fn, sizeof(fn), "digits/million");
|
|
} else {
|
|
ast_log(LOG_DEBUG, "Number '%d' is too big for me\n", num);
|
|
res = -1;
|
|
}
|
|
}
|
|
}
|
|
}
|
|
if (!res) {
|
|
if(!ast_streamfile(chan, fn, language)) {
|
|
if ((audiofd > -1) && (ctrlfd > -1))
|
|
res = ast_waitstream_full(chan, ints, audiofd, ctrlfd);
|
|
else
|
|
res = ast_waitstream(chan, ints);
|
|
}
|
|
ast_stopstream(chan);
|
|
}
|
|
}
|
|
return res;
|
|
}
|
|
|
|
|
|
/*--- determine last digits for thousands/millions (ru) */
|
|
static int get_lastdigits_ru(int num) {
|
|
if (num < 20) {
|
|
return num;
|
|
} else if (num < 100) {
|
|
return get_lastdigits_ru(num % 10);
|
|
} else if (num < 1000) {
|
|
return get_lastdigits_ru(num % 100);
|
|
}
|
|
return 0; /* number too big */
|
|
}
|
|
|
|
|
|
/*--- ast_say_number_full_ru: Russian syntax */
|
|
/*--- additional files:
|
|
n00.gsm (one hundred, two hundred, ...)
|
|
thousand.gsm
|
|
million.gsm
|
|
thousands-i.gsm (tisyachi)
|
|
million-a.gsm (milliona)
|
|
thousands.gsm
|
|
millions.gsm
|
|
1f.gsm (odna)
|
|
2f.gsm (dve)
|
|
|
|
where 'n' from 1 to 9
|
|
*/
|
|
static int ast_say_number_full_ru(struct ast_channel *chan, int num, const char *ints, const char *language, const char *options, int audiofd, int ctrlfd)
|
|
{
|
|
int res = 0;
|
|
int lastdigits = 0;
|
|
char fn[256] = "";
|
|
if (!num)
|
|
return ast_say_digits_full(chan, 0,ints, language, audiofd, ctrlfd);
|
|
|
|
while(!res && (num)) {
|
|
if (num < 0) {
|
|
snprintf(fn, sizeof(fn), "digits/minus");
|
|
if ( num > INT_MIN ) {
|
|
num = -num;
|
|
} else {
|
|
num = 0;
|
|
}
|
|
} else if (num < 20) {
|
|
if(options && strlen(options) == 1 && num < 3) {
|
|
snprintf(fn, sizeof(fn), "digits/%d%s", num, options);
|
|
} else {
|
|
snprintf(fn, sizeof(fn), "digits/%d", num);
|
|
}
|
|
num = 0;
|
|
} else if (num < 100) {
|
|
snprintf(fn, sizeof(fn), "digits/%d", num - (num % 10));
|
|
num %= 10;
|
|
} else if (num < 1000){
|
|
snprintf(fn, sizeof(fn), "digits/%d", num - (num % 100));
|
|
num %= 100;
|
|
} else if (num < 1000000) { /* 1,000,000 */
|
|
lastdigits = get_lastdigits_ru(num / 1000);
|
|
/* say thousands */
|
|
if (lastdigits < 3) {
|
|
res = ast_say_number_full_ru(chan, num / 1000, ints, language, "f", audiofd, ctrlfd);
|
|
} else {
|
|
res = ast_say_number_full_ru(chan, num / 1000, ints, language, NULL, audiofd, ctrlfd);
|
|
}
|
|
if (res)
|
|
return res;
|
|
if (lastdigits == 1) {
|
|
snprintf(fn, sizeof(fn), "digits/thousand");
|
|
} else if (lastdigits > 1 && lastdigits < 5) {
|
|
snprintf(fn, sizeof(fn), "digits/thousands-i");
|
|
} else {
|
|
snprintf(fn, sizeof(fn), "digits/thousands");
|
|
}
|
|
num %= 1000;
|
|
} else if (num < 1000000000) { /* 1,000,000,000 */
|
|
lastdigits = get_lastdigits_ru(num / 1000000);
|
|
/* say millions */
|
|
res = ast_say_number_full_ru(chan, num / 1000000, ints, language, NULL, audiofd, ctrlfd);
|
|
if (res)
|
|
return res;
|
|
if (lastdigits == 1) {
|
|
snprintf(fn, sizeof(fn), "digits/million");
|
|
} else if (lastdigits > 1 && lastdigits < 5) {
|
|
snprintf(fn, sizeof(fn), "digits/million-a");
|
|
} else {
|
|
snprintf(fn, sizeof(fn), "digits/millions");
|
|
}
|
|
num %= 1000000;
|
|
} else {
|
|
ast_log(LOG_DEBUG, "Number '%d' is too big for me\n", num);
|
|
res = -1;
|
|
}
|
|
if (!res) {
|
|
if (!ast_streamfile(chan, fn, language)) {
|
|
if ((audiofd > -1) && (ctrlfd > -1))
|
|
res = ast_waitstream_full(chan, ints, audiofd, ctrlfd);
|
|
else
|
|
res = ast_waitstream(chan, ints);
|
|
}
|
|
ast_stopstream(chan);
|
|
}
|
|
}
|
|
return res;
|
|
}
|
|
|
|
|
|
/*--- ast_say_enumeration_full: call language-specific functions */
|
|
/* Called from AGI */
|
|
int ast_say_enumeration_full(struct ast_channel *chan, int num, const char *ints, const char *language, const char *options, int audiofd, int ctrlfd)
|
|
{
|
|
if (!strcasecmp(language,"en") ) { /* English syntax */
|
|
return(ast_say_enumeration_full_en(chan, num, ints, language, audiofd, ctrlfd));
|
|
} else if (!strcasecmp(language, "da") ) { /* Danish syntax */
|
|
return(ast_say_enumeration_full_da(chan, num, ints, language, options, audiofd, ctrlfd));
|
|
} else if (!strcasecmp(language, "de") ) { /* German syntax */
|
|
return(ast_say_enumeration_full_de(chan, num, ints, language, options, audiofd, ctrlfd));
|
|
}
|
|
|
|
/* Default to english */
|
|
return(ast_say_enumeration_full_en(chan, num, ints, language, audiofd, ctrlfd));
|
|
}
|
|
|
|
/*--- ast_say_enumeration: call language-specific functions without file descriptors */
|
|
int ast_say_enumeration(struct ast_channel *chan, int num, const char *ints, const char *language, const char *options)
|
|
{
|
|
return(ast_say_enumeration_full(chan, num, ints, language, options, -1, -1));
|
|
}
|
|
|
|
/*--- ast_say_enumeration_full_en: English syntax */
|
|
/* This is the default syntax, if no other syntax defined in this file is used */
|
|
static int ast_say_enumeration_full_en(struct ast_channel *chan, int num, const char *ints, const char *language, int audiofd, int ctrlfd)
|
|
{
|
|
int res = 0, t = 0;
|
|
char fn[256] = "";
|
|
|
|
while(!res && num) {
|
|
if (num < 0) {
|
|
snprintf(fn, sizeof(fn), "digits/minus"); /* kind of senseless for enumerations, but our best effort for error checking */
|
|
if ( num > INT_MIN ) {
|
|
num = -num;
|
|
} else {
|
|
num = 0;
|
|
}
|
|
} else if (num < 20) {
|
|
snprintf(fn, sizeof(fn), "digits/h-%d", num);
|
|
num = 0;
|
|
} else if (num < 100) {
|
|
int tens = num / 10;
|
|
num = num % 10;
|
|
if (num == 0) {
|
|
snprintf(fn, sizeof(fn), "digits/h-%d", (tens * 10));
|
|
} else {
|
|
snprintf(fn, sizeof(fn), "digits/%d", (tens * 10));
|
|
}
|
|
} else if (num < 1000) {
|
|
int hundreds = num / 100;
|
|
num = num % 100;
|
|
if (hundreds > 1 || t == 1) {
|
|
res = ast_say_number_full_en(chan, hundreds, ints, language, audiofd, ctrlfd);
|
|
}
|
|
if (res)
|
|
return res;
|
|
if (num) {
|
|
snprintf(fn, sizeof(fn), "digits/hundred");
|
|
} else {
|
|
snprintf(fn, sizeof(fn), "digits/h-hundred");
|
|
}
|
|
} else if (num < 1000000) {
|
|
int thousands = num / 1000;
|
|
num = num % 1000;
|
|
if (thousands > 1 || t == 1) {
|
|
res = ast_say_number_full_en(chan, thousands, ints, language, audiofd, ctrlfd);
|
|
}
|
|
if (res)
|
|
return res;
|
|
if (num) {
|
|
snprintf(fn, sizeof(fn), "digits/thousand");
|
|
} else {
|
|
snprintf(fn, sizeof(fn), "digits/h-thousand");
|
|
}
|
|
t = 1;
|
|
} else if (num < 1000000000) {
|
|
int millions = num / 1000000;
|
|
num = num % 1000000;
|
|
t = 1;
|
|
res = ast_say_number_full_en(chan, millions, ints, language, audiofd, ctrlfd);
|
|
if (res)
|
|
return res;
|
|
if (num) {
|
|
snprintf(fn, sizeof(fn), "digits/million");
|
|
} else {
|
|
snprintf(fn, sizeof(fn), "digits/h-million");
|
|
}
|
|
} else if (num < INT_MAX) {
|
|
int billions = num / 1000000000;
|
|
num = num % 1000000000;
|
|
t = 1;
|
|
res = ast_say_number_full_en(chan, billions, ints, language, audiofd, ctrlfd);
|
|
if (res)
|
|
return res;
|
|
if (num) {
|
|
snprintf(fn, sizeof(fn), "digits/billion");
|
|
} else {
|
|
snprintf(fn, sizeof(fn), "digits/h-billion");
|
|
}
|
|
} else if (num == INT_MAX) {
|
|
snprintf(fn, sizeof(fn), "digits/h-last");
|
|
num = 0;
|
|
} else {
|
|
ast_log(LOG_DEBUG, "Number '%d' is too big for me\n", num);
|
|
res = -1;
|
|
}
|
|
|
|
if (!res) {
|
|
if (!ast_streamfile(chan, fn, language)) {
|
|
if ((audiofd > -1) && (ctrlfd > -1)) {
|
|
res = ast_waitstream_full(chan, ints, audiofd, ctrlfd);
|
|
} else {
|
|
res = ast_waitstream(chan, ints);
|
|
}
|
|
}
|
|
ast_stopstream(chan);
|
|
}
|
|
}
|
|
return res;
|
|
}
|
|
|
|
/*--- ast_say_enumeration_full_da: Danish syntax */
|
|
static int ast_say_enumeration_full_da(struct ast_channel *chan, int num, const char *ints, const char *language, const char *options, int audiofd, int ctrlfd)
|
|
{
|
|
/* options can be: '' or 'm' male gender; 'f' female gender; 'n' neuter gender; 'p' plural */
|
|
int res = 0, t = 0;
|
|
char fn[256] = "", fna[256] = "";
|
|
char *gender;
|
|
|
|
if (options && !strncasecmp(options, "f",1)) {
|
|
gender = "F";
|
|
} else if (options && !strncasecmp(options, "n",1)) {
|
|
gender = "N";
|
|
} else {
|
|
gender = "";
|
|
}
|
|
|
|
if (!num)
|
|
return ast_say_digits_full(chan, 0,ints, language, audiofd, ctrlfd);
|
|
|
|
while(!res && num) {
|
|
if (num < 0) {
|
|
snprintf(fn, sizeof(fn), "digits/minus"); /* kind of senseless for enumerations, but our best effort for error checking */
|
|
if ( num > INT_MIN ) {
|
|
num = -num;
|
|
} else {
|
|
num = 0;
|
|
}
|
|
} else if (num < 100 && t) {
|
|
snprintf(fn, sizeof(fn), "digits/and");
|
|
t = 0;
|
|
} else if (num < 20) {
|
|
snprintf(fn, sizeof(fn), "digits/h-%d%s", num, gender);
|
|
num = 0;
|
|
} else if (num < 100) {
|
|
int ones = num % 10;
|
|
if (ones) {
|
|
snprintf(fn, sizeof(fn), "digits/%d-and", ones);
|
|
num -= ones;
|
|
} else {
|
|
snprintf(fn, sizeof(fn), "digits/h-%d%s", num, gender);
|
|
num = 0;
|
|
}
|
|
} else if (num == 100 && t == 0) {
|
|
snprintf(fn, sizeof(fn), "digits/h-hundred%s", gender);
|
|
num = 0;
|
|
} else if (num < 1000) {
|
|
int hundreds = num / 100;
|
|
num = num % 100;
|
|
if (hundreds == 1) {
|
|
snprintf(fn, sizeof(fn), "digits/1N");
|
|
} else {
|
|
snprintf(fn, sizeof(fn), "digits/%d", hundreds);
|
|
}
|
|
if (num) {
|
|
snprintf(fna, sizeof(fna), "digits/hundred");
|
|
} else {
|
|
snprintf(fna, sizeof(fna), "digits/h-hundred%s", gender);
|
|
}
|
|
t = 1;
|
|
} else if (num < 1000000) {
|
|
int thousands = num / 1000;
|
|
num = num % 1000;
|
|
if (thousands == 1) {
|
|
if (num) {
|
|
snprintf(fn, sizeof(fn), "digits/1N");
|
|
snprintf(fna, sizeof(fna), "digits/thousand");
|
|
} else {
|
|
if (t) {
|
|
snprintf(fn, sizeof(fn), "digits/1N");
|
|
snprintf(fna, sizeof(fna), "digits/h-thousand%s", gender);
|
|
} else {
|
|
snprintf(fn, sizeof(fn), "digits/h-thousand%s", gender);
|
|
}
|
|
}
|
|
} else {
|
|
res = ast_say_number_full_de(chan, thousands, ints, language, options, audiofd, ctrlfd);
|
|
if (res) {
|
|
return res;
|
|
}
|
|
if (num) {
|
|
snprintf(fn, sizeof(fn), "digits/thousand");
|
|
} else {
|
|
snprintf(fn, sizeof(fn), "digits/h-thousand%s", gender);
|
|
}
|
|
}
|
|
t = 1;
|
|
} else if (num < 1000000000) {
|
|
int millions = num / 1000000;
|
|
num = num % 1000000;
|
|
if (millions == 1) {
|
|
if (num) {
|
|
snprintf(fn, sizeof(fn), "digits/1F");
|
|
snprintf(fna, sizeof(fna), "digits/million");
|
|
} else {
|
|
snprintf(fn, sizeof(fn), "digits/1N");
|
|
snprintf(fna, sizeof(fna), "digits/h-million%s", gender);
|
|
}
|
|
} else {
|
|
res = ast_say_number_full_de(chan, millions, ints, language, options, audiofd, ctrlfd);
|
|
if (res) {
|
|
return res;
|
|
}
|
|
if (num) {
|
|
snprintf(fn, sizeof(fn), "digits/millions");
|
|
} else {
|
|
snprintf(fn, sizeof(fn), "digits/h-million%s", gender);
|
|
}
|
|
}
|
|
t = 1;
|
|
} else if (num < INT_MAX) {
|
|
int billions = num / 1000000000;
|
|
num = num % 1000000000;
|
|
if (billions == 1) {
|
|
if (num) {
|
|
snprintf(fn, sizeof(fn), "digits/1F");
|
|
snprintf(fna, sizeof(fna), "digits/milliard");
|
|
} else {
|
|
snprintf(fn, sizeof(fn), "digits/1N");
|
|
snprintf(fna, sizeof(fna), "digits/h-milliard%s", gender);
|
|
}
|
|
} else {
|
|
res = ast_say_number_full_de(chan, billions, ints, language, options, audiofd, ctrlfd);
|
|
if (res)
|
|
return res;
|
|
if (num) {
|
|
snprintf(fn, sizeof(fna), "digits/milliards");
|
|
} else {
|
|
snprintf(fn, sizeof(fna), "digits/h-milliard%s", gender);
|
|
}
|
|
}
|
|
t = 1;
|
|
} else if (num == INT_MAX) {
|
|
snprintf(fn, sizeof(fn), "digits/h-last%s", gender);
|
|
num = 0;
|
|
} else {
|
|
ast_log(LOG_DEBUG, "Number '%d' is too big for me\n", num);
|
|
res = -1;
|
|
}
|
|
|
|
if (!res) {
|
|
if (!ast_streamfile(chan, fn, language)) {
|
|
if ((audiofd > -1) && (ctrlfd > -1))
|
|
res = ast_waitstream_full(chan, ints, audiofd, ctrlfd);
|
|
else
|
|
res = ast_waitstream(chan, ints);
|
|
}
|
|
ast_stopstream(chan);
|
|
if (!res) {
|
|
if (strlen(fna) != 0 && !ast_streamfile(chan, fna, language)) {
|
|
if ((audiofd > -1) && (ctrlfd > -1)) {
|
|
res = ast_waitstream_full(chan, ints, audiofd, ctrlfd);
|
|
} else {
|
|
res = ast_waitstream(chan, ints);
|
|
}
|
|
}
|
|
ast_stopstream(chan);
|
|
strcpy(fna, "");
|
|
}
|
|
}
|
|
}
|
|
return res;
|
|
}
|
|
|
|
/*--- ast_say_enumeration_full_de: German syntax */
|
|
static int ast_say_enumeration_full_de(struct ast_channel *chan, int num, const char *ints, const char *language, const char *options, int audiofd, int ctrlfd)
|
|
{
|
|
/* options can be: '' or 'm' male gender; 'f' female gender; 'n' neuter gender; 'p' plural */
|
|
int res = 0, t = 0;
|
|
char fn[256] = "", fna[256] = "";
|
|
char *gender;
|
|
|
|
if (options && !strncasecmp(options, "f",1)) {
|
|
gender = "F";
|
|
} else if (options && !strncasecmp(options, "n",1)) {
|
|
gender = "N";
|
|
} else {
|
|
gender = "";
|
|
}
|
|
|
|
if (!num)
|
|
return ast_say_digits_full(chan, 0,ints, language, audiofd, ctrlfd);
|
|
|
|
while(!res && num) {
|
|
if (num < 0) {
|
|
snprintf(fn, sizeof(fn), "digits/minus"); /* kind of senseless for enumerations, but our best effort for error checking */
|
|
if ( num > INT_MIN ) {
|
|
num = -num;
|
|
} else {
|
|
num = 0;
|
|
}
|
|
} else if (num < 100 && t) {
|
|
snprintf(fn, sizeof(fn), "digits/and");
|
|
t = 0;
|
|
} else if (num < 20) {
|
|
snprintf(fn, sizeof(fn), "digits/h-%d%s", num, gender);
|
|
num = 0;
|
|
} else if (num < 100) {
|
|
int ones = num % 10;
|
|
if (ones) {
|
|
snprintf(fn, sizeof(fn), "digits/%d-and", ones);
|
|
num -= ones;
|
|
} else {
|
|
snprintf(fn, sizeof(fn), "digits/h-%d%s", num, gender);
|
|
num = 0;
|
|
}
|
|
} else if (num == 100 && t == 0) {
|
|
snprintf(fn, sizeof(fn), "digits/h-hundred%s", gender);
|
|
num = 0;
|
|
} else if (num < 1000) {
|
|
int hundreds = num / 100;
|
|
num = num % 100;
|
|
if (hundreds == 1) {
|
|
snprintf(fn, sizeof(fn), "digits/1N");
|
|
} else {
|
|
snprintf(fn, sizeof(fn), "digits/%d", hundreds);
|
|
}
|
|
if (num) {
|
|
snprintf(fna, sizeof(fna), "digits/hundred");
|
|
} else {
|
|
snprintf(fna, sizeof(fna), "digits/h-hundred%s", gender);
|
|
}
|
|
t = 1;
|
|
} else if (num < 1000000) {
|
|
int thousands = num / 1000;
|
|
num = num % 1000;
|
|
if (thousands == 1) {
|
|
if (num) {
|
|
snprintf(fn, sizeof(fn), "digits/1N");
|
|
snprintf(fna, sizeof(fna), "digits/thousand");
|
|
} else {
|
|
if (t) {
|
|
snprintf(fn, sizeof(fn), "digits/1N");
|
|
snprintf(fna, sizeof(fna), "digits/h-thousand%s", gender);
|
|
} else {
|
|
snprintf(fn, sizeof(fn), "digits/h-thousand%s", gender);
|
|
}
|
|
}
|
|
} else {
|
|
res = ast_say_number_full_de(chan, thousands, ints, language, options, audiofd, ctrlfd);
|
|
if (res) {
|
|
return res;
|
|
}
|
|
if (num) {
|
|
snprintf(fn, sizeof(fn), "digits/thousand");
|
|
} else {
|
|
snprintf(fn, sizeof(fn), "digits/h-thousand%s", gender);
|
|
}
|
|
}
|
|
t = 1;
|
|
} else if (num < 1000000000) {
|
|
int millions = num / 1000000;
|
|
num = num % 1000000;
|
|
if (millions == 1) {
|
|
if (num) {
|
|
snprintf(fn, sizeof(fn), "digits/1F");
|
|
snprintf(fna, sizeof(fna), "digits/million");
|
|
} else {
|
|
snprintf(fn, sizeof(fn), "digits/1N");
|
|
snprintf(fna, sizeof(fna), "digits/h-million%s", gender);
|
|
}
|
|
} else {
|
|
res = ast_say_number_full_de(chan, millions, ints, language, options, audiofd, ctrlfd);
|
|
if (res) {
|
|
return res;
|
|
}
|
|
if (num) {
|
|
snprintf(fn, sizeof(fn), "digits/millions");
|
|
} else {
|
|
snprintf(fn, sizeof(fn), "digits/h-million%s", gender);
|
|
}
|
|
}
|
|
t = 1;
|
|
} else if (num < INT_MAX) {
|
|
int billions = num / 1000000000;
|
|
num = num % 1000000000;
|
|
if (billions == 1) {
|
|
if (num) {
|
|
snprintf(fn, sizeof(fn), "digits/1F");
|
|
snprintf(fna, sizeof(fna), "digits/milliard");
|
|
} else {
|
|
snprintf(fn, sizeof(fn), "digits/1N");
|
|
snprintf(fna, sizeof(fna), "digits/h-milliard%s", gender);
|
|
}
|
|
} else {
|
|
res = ast_say_number_full_de(chan, billions, ints, language, options, audiofd, ctrlfd);
|
|
if (res)
|
|
return res;
|
|
if (num) {
|
|
snprintf(fn, sizeof(fna), "digits/milliards");
|
|
} else {
|
|
snprintf(fn, sizeof(fna), "digits/h-milliard%s", gender);
|
|
}
|
|
}
|
|
t = 1;
|
|
} else if (num == INT_MAX) {
|
|
snprintf(fn, sizeof(fn), "digits/h-last%s", gender);
|
|
num = 0;
|
|
} else {
|
|
ast_log(LOG_DEBUG, "Number '%d' is too big for me\n", num);
|
|
res = -1;
|
|
}
|
|
|
|
if (!res) {
|
|
if (!ast_streamfile(chan, fn, language)) {
|
|
if ((audiofd > -1) && (ctrlfd > -1))
|
|
res = ast_waitstream_full(chan, ints, audiofd, ctrlfd);
|
|
else
|
|
res = ast_waitstream(chan, ints);
|
|
}
|
|
ast_stopstream(chan);
|
|
if (!res) {
|
|
if (strlen(fna) != 0 && !ast_streamfile(chan, fna, language)) {
|
|
if ((audiofd > -1) && (ctrlfd > -1)) {
|
|
res = ast_waitstream_full(chan, ints, audiofd, ctrlfd);
|
|
} else {
|
|
res = ast_waitstream(chan, ints);
|
|
}
|
|
}
|
|
ast_stopstream(chan);
|
|
strcpy(fna, "");
|
|
}
|
|
}
|
|
}
|
|
return res;
|
|
}
|
|
|
|
int ast_say_date(struct ast_channel *chan, time_t t, const char *ints, const char *lang)
|
|
{
|
|
if (!strcasecmp(lang, "en") ) { /* English syntax */
|
|
return(ast_say_date_en(chan, t, ints, lang));
|
|
} else if (!strcasecmp(lang, "da") ) { /* Danish syntax */
|
|
return(ast_say_date_da(chan, t, ints, lang));
|
|
} else if (!strcasecmp(lang, "de") ) { /* German syntax */
|
|
return(ast_say_date_de(chan, t, ints, lang));
|
|
} else if (!strcasecmp(lang, "fr") ) { /* French syntax */
|
|
return(ast_say_date_fr(chan, t, ints, lang));
|
|
} else if (!strcasecmp(lang, "nl") ) { /* Dutch syntax */
|
|
return(ast_say_date_nl(chan, t, ints, lang));
|
|
} else if (!strcasecmp(lang, "pt") ) { /* Portuguese syntax */
|
|
return(ast_say_date_pt(chan, t, ints, lang));
|
|
} else if (!strcasecmp(lang, "gr") ) { /* Greek syntax */
|
|
return(ast_say_date_gr(chan, t, ints, lang));
|
|
}
|
|
|
|
/* Default to English */
|
|
return(ast_say_date_en(chan, t, ints, lang));
|
|
}
|
|
|
|
/* English syntax */
|
|
int ast_say_date_en(struct ast_channel *chan, time_t t, const char *ints, const char *lang)
|
|
{
|
|
struct tm tm;
|
|
char fn[256];
|
|
int res = 0;
|
|
ast_localtime(&t,&tm,NULL);
|
|
if (!res) {
|
|
snprintf(fn, sizeof(fn), "digits/day-%d", tm.tm_wday);
|
|
res = ast_streamfile(chan, fn, lang);
|
|
if (!res)
|
|
res = ast_waitstream(chan, ints);
|
|
}
|
|
if (!res) {
|
|
snprintf(fn, sizeof(fn), "digits/mon-%d", tm.tm_mon);
|
|
res = ast_streamfile(chan, fn, lang);
|
|
if (!res)
|
|
res = ast_waitstream(chan, ints);
|
|
}
|
|
if (!res)
|
|
res = ast_say_number(chan, tm.tm_mday, ints, lang, (char * ) NULL);
|
|
if (!res)
|
|
res = ast_waitstream(chan, ints);
|
|
if (!res)
|
|
res = ast_say_number(chan, tm.tm_year + 1900, ints, lang, (char *) NULL);
|
|
return res;
|
|
}
|
|
|
|
/* Danish syntax */
|
|
int ast_say_date_da(struct ast_channel *chan, time_t t, const char *ints, const char *lang)
|
|
{
|
|
struct tm tm;
|
|
char fn[256];
|
|
int res = 0;
|
|
ast_localtime(&t,&tm,NULL);
|
|
if (!res) {
|
|
snprintf(fn, sizeof(fn), "digits/day-%d", tm.tm_wday);
|
|
res = ast_streamfile(chan, fn, lang);
|
|
if (!res)
|
|
res = ast_waitstream(chan, ints);
|
|
}
|
|
if (!res)
|
|
res = ast_say_enumeration(chan, tm.tm_mday, ints, lang, (char * ) NULL);
|
|
if (!res)
|
|
res = ast_waitstream(chan, ints);
|
|
if (!res) {
|
|
snprintf(fn, sizeof(fn), "digits/mon-%d", tm.tm_mon);
|
|
res = ast_streamfile(chan, fn, lang);
|
|
if (!res)
|
|
res = ast_waitstream(chan, ints);
|
|
}
|
|
if (!res) {
|
|
/* Year */
|
|
int year = tm.tm_year + 1900;
|
|
if (year > 1999) { /* year 2000 and later */
|
|
res = ast_say_number(chan, year, ints, lang, (char *) NULL);
|
|
} else {
|
|
if (year < 1100) {
|
|
/* I'm not going to handle 1100 and prior */
|
|
/* We'll just be silent on the year, instead of bombing out. */
|
|
} else {
|
|
/* year 1100 to 1999. will anybody need this?!? */
|
|
snprintf(fn,sizeof(fn), "digits/%d", (year / 100) );
|
|
res = wait_file(chan, ints, fn, lang);
|
|
if (!res) {
|
|
res = wait_file(chan,ints, "digits/hundred", lang);
|
|
if (!res && year % 100 != 0) {
|
|
res = ast_say_number(chan, (year % 100), ints, lang, (char *) NULL);
|
|
}
|
|
}
|
|
}
|
|
}
|
|
}
|
|
return res;
|
|
}
|
|
|
|
/* German syntax */
|
|
int ast_say_date_de(struct ast_channel *chan, time_t t, const char *ints, const char *lang)
|
|
{
|
|
struct tm tm;
|
|
char fn[256];
|
|
int res = 0;
|
|
ast_localtime(&t,&tm,NULL);
|
|
if (!res) {
|
|
snprintf(fn, sizeof(fn), "digits/day-%d", tm.tm_wday);
|
|
res = ast_streamfile(chan, fn, lang);
|
|
if (!res)
|
|
res = ast_waitstream(chan, ints);
|
|
}
|
|
if (!res)
|
|
res = ast_say_enumeration(chan, tm.tm_mday, ints, lang, (char * ) NULL);
|
|
if (!res)
|
|
res = ast_waitstream(chan, ints);
|
|
if (!res) {
|
|
snprintf(fn, sizeof(fn), "digits/mon-%d", tm.tm_mon);
|
|
res = ast_streamfile(chan, fn, lang);
|
|
if (!res)
|
|
res = ast_waitstream(chan, ints);
|
|
}
|
|
if (!res) {
|
|
/* Year */
|
|
int year = tm.tm_year + 1900;
|
|
if (year > 1999) { /* year 2000 and later */
|
|
res = ast_say_number(chan, year, ints, lang, (char *) NULL);
|
|
} else {
|
|
if (year < 1100) {
|
|
/* I'm not going to handle 1100 and prior */
|
|
/* We'll just be silent on the year, instead of bombing out. */
|
|
} else {
|
|
/* year 1100 to 1999. will anybody need this?!? */
|
|
/* say 1967 as 'neunzehn hundert sieben und sechzig' */
|
|
snprintf(fn,sizeof(fn), "digits/%d", (year / 100) );
|
|
res = wait_file(chan, ints, fn, lang);
|
|
if (!res) {
|
|
res = wait_file(chan,ints, "digits/hundred", lang);
|
|
if (!res && year % 100 != 0) {
|
|
res = ast_say_number(chan, (year % 100), ints, lang, (char *) NULL);
|
|
}
|
|
}
|
|
}
|
|
}
|
|
}
|
|
return res;
|
|
}
|
|
|
|
/* French syntax */
|
|
int ast_say_date_fr(struct ast_channel *chan, time_t t, const char *ints, const char *lang)
|
|
{
|
|
struct tm tm;
|
|
char fn[256];
|
|
int res = 0;
|
|
ast_localtime(&t,&tm,NULL);
|
|
if (!res) {
|
|
snprintf(fn, sizeof(fn), "digits/day-%d", tm.tm_wday);
|
|
res = ast_streamfile(chan, fn, lang);
|
|
if (!res)
|
|
res = ast_waitstream(chan, ints);
|
|
}
|
|
if (!res)
|
|
res = ast_say_number(chan, tm.tm_mday, ints, lang, (char * ) NULL);
|
|
if (!res)
|
|
res = ast_waitstream(chan, ints);
|
|
if (!res) {
|
|
snprintf(fn, sizeof(fn), "digits/mon-%d", tm.tm_mon);
|
|
res = ast_streamfile(chan, fn, lang);
|
|
if (!res)
|
|
res = ast_waitstream(chan, ints);
|
|
}
|
|
if (!res)
|
|
res = ast_say_number(chan, tm.tm_year + 1900, ints, lang, (char *) NULL);
|
|
return res;
|
|
}
|
|
|
|
/* Dutch syntax */
|
|
int ast_say_date_nl(struct ast_channel *chan, time_t t, const char *ints, const char *lang)
|
|
{
|
|
struct tm tm;
|
|
char fn[256];
|
|
int res = 0;
|
|
ast_localtime(&t,&tm,NULL);
|
|
if (!res) {
|
|
snprintf(fn, sizeof(fn), "digits/day-%d", tm.tm_wday);
|
|
res = ast_streamfile(chan, fn, lang);
|
|
if (!res)
|
|
res = ast_waitstream(chan, ints);
|
|
}
|
|
if (!res)
|
|
res = ast_say_number(chan, tm.tm_mday, ints, lang, (char * ) NULL);
|
|
if (!res) {
|
|
snprintf(fn, sizeof(fn), "digits/mon-%d", tm.tm_mon);
|
|
res = ast_streamfile(chan, fn, lang);
|
|
if (!res)
|
|
res = ast_waitstream(chan, ints);
|
|
}
|
|
if (!res)
|
|
res = ast_waitstream(chan, ints);
|
|
if (!res)
|
|
res = ast_say_number(chan, tm.tm_year + 1900, ints, lang, (char *) NULL);
|
|
return res;
|
|
}
|
|
|
|
/* Portuguese syntax */
|
|
int ast_say_date_pt(struct ast_channel *chan, time_t t, const char *ints, const char *lang)
|
|
{
|
|
struct tm tm;
|
|
char fn[256];
|
|
int res = 0;
|
|
ast_localtime(&t,&tm,NULL);
|
|
localtime_r(&t,&tm);
|
|
snprintf(fn, sizeof(fn), "digits/day-%d", tm.tm_wday);
|
|
if (!res)
|
|
res = wait_file(chan, ints, fn, lang);
|
|
if (!res)
|
|
res = ast_say_number(chan, tm.tm_mday, ints, lang, (char *) NULL);
|
|
if (!res)
|
|
res = wait_file(chan, ints, "digits/pt-de", lang);
|
|
snprintf(fn, sizeof(fn), "digits/mon-%d", tm.tm_mon);
|
|
if (!res)
|
|
res = wait_file(chan, ints, fn, lang);
|
|
if (!res)
|
|
res = wait_file(chan, ints, "digits/pt-de", lang);
|
|
if (!res)
|
|
res = ast_say_number(chan, tm.tm_year + 1900, ints, lang, (char *) NULL);
|
|
|
|
return res;
|
|
}
|
|
|
|
int ast_say_date_with_format(struct ast_channel *chan, time_t time, const char *ints, const char *lang, const char *format, const char *timezone)
|
|
{
|
|
if (!strcasecmp(lang, "en") ) { /* English syntax */
|
|
return(ast_say_date_with_format_en(chan, time, ints, lang, format, timezone));
|
|
} else if (!strcasecmp(lang, "da") ) { /* Danish syntax */
|
|
return(ast_say_date_with_format_da(chan, time, ints, lang, format, timezone));
|
|
} else if (!strcasecmp(lang, "de") ) { /* German syntax */
|
|
return(ast_say_date_with_format_de(chan, time, ints, lang, format, timezone));
|
|
} else if (!strcasecmp(lang, "es") || !strcasecmp(lang, "mx")) { /* Spanish syntax */
|
|
return(ast_say_date_with_format_es(chan, time, ints, lang, format, timezone));
|
|
} else if (!strcasecmp(lang, "he")) { /* Hebrew syntax */
|
|
return(ast_say_date_with_format_he(chan, time, ints, lang, format, timezone));
|
|
} else if (!strcasecmp(lang, "fr") ) { /* French syntax */
|
|
return(ast_say_date_with_format_fr(chan, time, ints, lang, format, timezone));
|
|
} else if (!strcasecmp(lang, "it") ) { /* Italian syntax */
|
|
return(ast_say_date_with_format_it(chan, time, ints, lang, format, timezone));
|
|
} else if (!strcasecmp(lang, "nl") ) { /* Dutch syntax */
|
|
return(ast_say_date_with_format_nl(chan, time, ints, lang, format, timezone));
|
|
} else if (!strcasecmp(lang, "pt") ) { /* Portuguese syntax */
|
|
return(ast_say_date_with_format_pt(chan, time, ints, lang, format, timezone));
|
|
} else if (!strcasecmp(lang, "tw") ) { /* Taiwanese syntax */
|
|
return(ast_say_date_with_format_tw(chan, time, ints, lang, format, timezone));
|
|
} else if (!strcasecmp(lang, "gr") ) { /* Greek syntax */
|
|
return(ast_say_date_with_format_gr(chan, time, ints, lang, format, timezone));
|
|
}
|
|
|
|
/* Default to English */
|
|
return(ast_say_date_with_format_en(chan, time, ints, lang, format, timezone));
|
|
}
|
|
|
|
/* English syntax */
|
|
int ast_say_date_with_format_en(struct ast_channel *chan, time_t time, const char *ints, const char *lang, const char *format, const char *timezone)
|
|
{
|
|
struct tm tm;
|
|
int res=0, offset, sndoffset;
|
|
char sndfile[256], nextmsg[256];
|
|
|
|
ast_localtime(&time,&tm,timezone);
|
|
|
|
for (offset=0 ; format[offset] != '\0' ; offset++) {
|
|
ast_log(LOG_DEBUG, "Parsing %c (offset %d) in %s\n", format[offset], offset, format);
|
|
switch (format[offset]) {
|
|
/* NOTE: if you add more options here, please try to be consistent with strftime(3) */
|
|
case '\'':
|
|
/* Literal name of a sound file */
|
|
sndoffset=0;
|
|
for (sndoffset=0 ; (format[++offset] != '\'') && (sndoffset < 256) ; sndoffset++)
|
|
sndfile[sndoffset] = format[offset];
|
|
sndfile[sndoffset] = '\0';
|
|
res = wait_file(chan,ints,sndfile,lang);
|
|
break;
|
|
case 'A':
|
|
case 'a':
|
|
/* Sunday - Saturday */
|
|
snprintf(nextmsg,sizeof(nextmsg), "digits/day-%d", tm.tm_wday);
|
|
res = wait_file(chan,ints,nextmsg,lang);
|
|
break;
|
|
case 'B':
|
|
case 'b':
|
|
case 'h':
|
|
/* January - December */
|
|
snprintf(nextmsg,sizeof(nextmsg), "digits/mon-%d", tm.tm_mon);
|
|
res = wait_file(chan,ints,nextmsg,lang);
|
|
break;
|
|
case 'm':
|
|
/* Month enumerated */
|
|
res = ast_say_enumeration(chan, (tm.tm_mon + 1), ints, lang, (char *) NULL);
|
|
break;
|
|
case 'd':
|
|
case 'e':
|
|
/* First - Thirtyfirst */
|
|
res = ast_say_enumeration(chan, tm.tm_mday, ints, lang, (char *) NULL);
|
|
break;
|
|
case 'Y':
|
|
/* Year */
|
|
if (tm.tm_year > 99) {
|
|
res = ast_say_number(chan, tm.tm_year + 1900, ints, lang, (char *) NULL);
|
|
} else {
|
|
if (tm.tm_year < 1) {
|
|
/* I'm not going to handle 1900 and prior */
|
|
/* We'll just be silent on the year, instead of bombing out. */
|
|
} else {
|
|
res = wait_file(chan,ints, "digits/19",lang);
|
|
if (!res) {
|
|
if (tm.tm_year <= 9) {
|
|
/* 1901 - 1909 */
|
|
res = wait_file(chan,ints, "digits/oh",lang);
|
|
if (!res) {
|
|
snprintf(nextmsg,sizeof(nextmsg), "digits/%d", tm.tm_year);
|
|
res = wait_file(chan,ints,nextmsg,lang);
|
|
}
|
|
} else if (tm.tm_year <= 20) {
|
|
/* 1910 - 1920 */
|
|
snprintf(nextmsg,sizeof(nextmsg), "digits/%d", tm.tm_year);
|
|
res = wait_file(chan,ints,nextmsg,lang);
|
|
} else {
|
|
/* 1921 - 1999 */
|
|
int ten, one;
|
|
ten = tm.tm_year / 10;
|
|
one = tm.tm_year % 10;
|
|
snprintf(nextmsg,sizeof(nextmsg), "digits/%d", ten * 10);
|
|
res = wait_file(chan,ints,nextmsg,lang);
|
|
if (!res) {
|
|
if (one != 0) {
|
|
snprintf(nextmsg,sizeof(nextmsg), "digits/%d", one);
|
|
res = wait_file(chan,ints,nextmsg,lang);
|
|
}
|
|
}
|
|
}
|
|
}
|
|
}
|
|
}
|
|
break;
|
|
case 'I':
|
|
case 'l':
|
|
/* 12-Hour */
|
|
if (tm.tm_hour == 0)
|
|
snprintf(nextmsg,sizeof(nextmsg), "digits/12");
|
|
else if (tm.tm_hour > 12)
|
|
snprintf(nextmsg,sizeof(nextmsg), "digits/%d", tm.tm_hour - 12);
|
|
else
|
|
snprintf(nextmsg,sizeof(nextmsg), "digits/%d", tm.tm_hour);
|
|
res = wait_file(chan,ints,nextmsg,lang);
|
|
break;
|
|
case 'H':
|
|
case 'k':
|
|
/* 24-Hour */
|
|
if (format[offset] == 'H') {
|
|
/* e.g. oh-eight */
|
|
if (tm.tm_hour < 10) {
|
|
res = wait_file(chan,ints, "digits/oh",lang);
|
|
}
|
|
} else {
|
|
/* e.g. eight */
|
|
if (tm.tm_hour == 0) {
|
|
res = wait_file(chan,ints, "digits/oh",lang);
|
|
}
|
|
}
|
|
if (!res) {
|
|
if (tm.tm_hour != 0) {
|
|
int remainder = tm.tm_hour;
|
|
if (tm.tm_hour > 20) {
|
|
res = wait_file(chan,ints, "digits/20",lang);
|
|
remainder -= 20;
|
|
}
|
|
if (!res) {
|
|
snprintf(nextmsg,sizeof(nextmsg), "digits/%d", remainder);
|
|
res = wait_file(chan,ints,nextmsg,lang);
|
|
}
|
|
}
|
|
}
|
|
break;
|
|
case 'M':
|
|
case 'N':
|
|
/* Minute */
|
|
if (tm.tm_min == 0) {
|
|
if (format[offset] == 'M') {
|
|
res = wait_file(chan, ints, "digits/oclock", lang);
|
|
} else {
|
|
res = wait_file(chan, ints, "digits/hundred", lang);
|
|
}
|
|
} else if (tm.tm_min < 10) {
|
|
res = wait_file(chan,ints, "digits/oh",lang);
|
|
if (!res) {
|
|
snprintf(nextmsg,sizeof(nextmsg), "digits/%d", tm.tm_min);
|
|
res = wait_file(chan,ints,nextmsg,lang);
|
|
}
|
|
} else if ((tm.tm_min < 21) || (tm.tm_min % 10 == 0)) {
|
|
snprintf(nextmsg,sizeof(nextmsg), "digits/%d", tm.tm_min);
|
|
res = wait_file(chan,ints,nextmsg,lang);
|
|
} else {
|
|
int ten, one;
|
|
ten = (tm.tm_min / 10) * 10;
|
|
one = (tm.tm_min % 10);
|
|
snprintf(nextmsg,sizeof(nextmsg), "digits/%d", ten);
|
|
res = wait_file(chan,ints,nextmsg,lang);
|
|
if (!res) {
|
|
/* Fifty, not fifty-zero */
|
|
if (one != 0) {
|
|
snprintf(nextmsg,sizeof(nextmsg), "digits/%d", one);
|
|
res = wait_file(chan,ints,nextmsg,lang);
|
|
}
|
|
}
|
|
}
|
|
break;
|
|
case 'P':
|
|
case 'p':
|
|
/* AM/PM */
|
|
if (tm.tm_hour > 11)
|
|
snprintf(nextmsg,sizeof(nextmsg), "digits/p-m");
|
|
else
|
|
snprintf(nextmsg,sizeof(nextmsg), "digits/a-m");
|
|
res = wait_file(chan,ints,nextmsg,lang);
|
|
break;
|
|
case 'Q':
|
|
/* Shorthand for "Today", "Yesterday", or ABdY */
|
|
{
|
|
struct timeval now;
|
|
struct tm tmnow;
|
|
time_t beg_today;
|
|
|
|
gettimeofday(&now,NULL);
|
|
ast_localtime(&now.tv_sec,&tmnow,timezone);
|
|
/* This might be slightly off, if we transcend a leap second, but never more off than 1 second */
|
|
/* In any case, it saves not having to do ast_mktime() */
|
|
beg_today = now.tv_sec - (tmnow.tm_hour * 3600) - (tmnow.tm_min * 60) - (tmnow.tm_sec);
|
|
if (beg_today < time) {
|
|
/* Today */
|
|
res = wait_file(chan,ints, "digits/today",lang);
|
|
} else if (beg_today - 86400 < time) {
|
|
/* Yesterday */
|
|
res = wait_file(chan,ints, "digits/yesterday",lang);
|
|
} else {
|
|
res = ast_say_date_with_format(chan, time, ints, lang, "ABdY", timezone);
|
|
}
|
|
}
|
|
break;
|
|
case 'q':
|
|
/* Shorthand for "" (today), "Yesterday", A (weekday), or ABdY */
|
|
{
|
|
struct timeval now;
|
|
struct tm tmnow;
|
|
time_t beg_today;
|
|
|
|
gettimeofday(&now,NULL);
|
|
ast_localtime(&now.tv_sec,&tmnow,timezone);
|
|
/* This might be slightly off, if we transcend a leap second, but never more off than 1 second */
|
|
/* In any case, it saves not having to do ast_mktime() */
|
|
beg_today = now.tv_sec - (tmnow.tm_hour * 3600) - (tmnow.tm_min * 60) - (tmnow.tm_sec);
|
|
if (beg_today < time) {
|
|
/* Today */
|
|
} else if ((beg_today - 86400) < time) {
|
|
/* Yesterday */
|
|
res = wait_file(chan,ints, "digits/yesterday",lang);
|
|
} else if (beg_today - 86400 * 6 < time) {
|
|
/* Within the last week */
|
|
res = ast_say_date_with_format(chan, time, ints, lang, "A", timezone);
|
|
} else {
|
|
res = ast_say_date_with_format(chan, time, ints, lang, "ABdY", timezone);
|
|
}
|
|
}
|
|
break;
|
|
case 'R':
|
|
res = ast_say_date_with_format(chan, time, ints, lang, "HM", timezone);
|
|
break;
|
|
case 'S':
|
|
/* Seconds */
|
|
if (tm.tm_sec == 0) {
|
|
snprintf(nextmsg,sizeof(nextmsg), "digits/%d", tm.tm_sec);
|
|
res = wait_file(chan,ints,nextmsg,lang);
|
|
} else if (tm.tm_sec < 10) {
|
|
res = wait_file(chan,ints, "digits/oh",lang);
|
|
if (!res) {
|
|
snprintf(nextmsg,sizeof(nextmsg), "digits/%d", tm.tm_sec);
|
|
res = wait_file(chan,ints,nextmsg,lang);
|
|
}
|
|
} else if ((tm.tm_sec < 21) || (tm.tm_sec % 10 == 0)) {
|
|
snprintf(nextmsg,sizeof(nextmsg), "digits/%d", tm.tm_sec);
|
|
res = wait_file(chan,ints,nextmsg,lang);
|
|
} else {
|
|
int ten, one;
|
|
ten = (tm.tm_sec / 10) * 10;
|
|
one = (tm.tm_sec % 10);
|
|
snprintf(nextmsg,sizeof(nextmsg), "digits/%d", ten);
|
|
res = wait_file(chan,ints,nextmsg,lang);
|
|
if (!res) {
|
|
/* Fifty, not fifty-zero */
|
|
if (one != 0) {
|
|
snprintf(nextmsg,sizeof(nextmsg), "digits/%d", one);
|
|
res = wait_file(chan,ints,nextmsg,lang);
|
|
}
|
|
}
|
|
}
|
|
break;
|
|
case 'T':
|
|
res = ast_say_date_with_format(chan, time, ints, lang, "HMS", timezone);
|
|
break;
|
|
case ' ':
|
|
case ' ':
|
|
/* Just ignore spaces and tabs */
|
|
break;
|
|
default:
|
|
/* Unknown character */
|
|
ast_log(LOG_WARNING, "Unknown character in datetime format %s: %c at pos %d\n", format, format[offset], offset);
|
|
}
|
|
/* Jump out on DTMF */
|
|
if (res) {
|
|
break;
|
|
}
|
|
}
|
|
return res;
|
|
}
|
|
|
|
/* Danish syntax */
|
|
int ast_say_date_with_format_da(struct ast_channel *chan, time_t time, const char *ints, const char *lang, const char *format, const char *timezone)
|
|
{
|
|
struct tm tm;
|
|
int res=0, offset, sndoffset;
|
|
char sndfile[256], nextmsg[256];
|
|
|
|
ast_localtime(&time,&tm,timezone);
|
|
|
|
for (offset=0 ; format[offset] != '\0' ; offset++) {
|
|
ast_log(LOG_DEBUG, "Parsing %c (offset %d) in %s\n", format[offset], offset, format);
|
|
switch (format[offset]) {
|
|
/* NOTE: if you add more options here, please try to be consistent with strftime(3) */
|
|
case '\'':
|
|
/* Literal name of a sound file */
|
|
sndoffset=0;
|
|
for (sndoffset=0 ; (format[++offset] != '\'') && (sndoffset < 256) ; sndoffset++)
|
|
sndfile[sndoffset] = format[offset];
|
|
sndfile[sndoffset] = '\0';
|
|
res = wait_file(chan,ints,sndfile,lang);
|
|
break;
|
|
case 'A':
|
|
case 'a':
|
|
/* Sunday - Saturday */
|
|
snprintf(nextmsg,sizeof(nextmsg), "digits/day-%d", tm.tm_wday);
|
|
res = wait_file(chan,ints,nextmsg,lang);
|
|
break;
|
|
case 'B':
|
|
case 'b':
|
|
case 'h':
|
|
/* January - December */
|
|
snprintf(nextmsg,sizeof(nextmsg), "digits/mon-%d", tm.tm_mon);
|
|
res = wait_file(chan,ints,nextmsg,lang);
|
|
break;
|
|
case 'm':
|
|
/* Month enumerated */
|
|
res = ast_say_enumeration(chan, (tm.tm_mon + 1), ints, lang, "m");
|
|
break;
|
|
case 'd':
|
|
case 'e':
|
|
/* First - Thirtyfirst */
|
|
res = ast_say_enumeration(chan, tm.tm_mday, ints, lang, "m");
|
|
break;
|
|
case 'Y':
|
|
/* Year */
|
|
{
|
|
int year = tm.tm_year + 1900;
|
|
if (year > 1999) { /* year 2000 and later */
|
|
res = ast_say_number(chan, year, ints, lang, (char *) NULL);
|
|
} else {
|
|
if (year < 1100) {
|
|
/* I'm not going to handle 1100 and prior */
|
|
/* We'll just be silent on the year, instead of bombing out. */
|
|
} else {
|
|
/* year 1100 to 1999. will anybody need this?!? */
|
|
/* say 1967 as 'nineteen hundred seven and sixty' */
|
|
snprintf(nextmsg,sizeof(nextmsg), "digits/%d", (year / 100) );
|
|
res = wait_file(chan,ints,nextmsg,lang);
|
|
if (!res) {
|
|
res = wait_file(chan,ints, "digits/hundred",lang);
|
|
if (!res && year % 100 != 0) {
|
|
res = ast_say_number(chan, (year % 100), ints, lang, (char *) NULL);
|
|
}
|
|
}
|
|
}
|
|
}
|
|
}
|
|
break;
|
|
case 'I':
|
|
case 'l':
|
|
/* 12-Hour */
|
|
res = wait_file(chan,ints,"digits/oclock",lang);
|
|
if (tm.tm_hour == 0)
|
|
snprintf(nextmsg,sizeof(nextmsg), "digits/12");
|
|
else if (tm.tm_hour > 12)
|
|
snprintf(nextmsg,sizeof(nextmsg), "digits/%d", tm.tm_hour - 12);
|
|
else
|
|
snprintf(nextmsg,sizeof(nextmsg), "digits/%d", tm.tm_hour);
|
|
if (!res) {
|
|
res = wait_file(chan,ints,nextmsg,lang);
|
|
}
|
|
break;
|
|
case 'H':
|
|
case 'k':
|
|
/* 24-Hour */
|
|
res = wait_file(chan,ints,"digits/oclock",lang);
|
|
if (!res) {
|
|
res = ast_say_number(chan, tm.tm_hour, ints, lang, (char *) NULL);
|
|
}
|
|
break;
|
|
case 'M':
|
|
/* Minute */
|
|
if (tm.tm_min > 0 || format[offset+ 1 ] == 'S' ) { /* zero 'digits/0' only if seconds follow (kind of a hack) */
|
|
res = ast_say_number(chan, tm.tm_min, ints, lang, "f");
|
|
}
|
|
if ( !res && format[offset + 1] == 'S' ) { /* minutes only if seconds follow (kind of a hack) */
|
|
if (tm.tm_min == 1) {
|
|
res = wait_file(chan,ints,"digits/minute",lang);
|
|
} else {
|
|
res = wait_file(chan,ints,"digits/minutes",lang);
|
|
}
|
|
}
|
|
break;
|
|
case 'P':
|
|
case 'p':
|
|
/* AM/PM */
|
|
if (tm.tm_hour > 11)
|
|
snprintf(nextmsg,sizeof(nextmsg), "digits/p-m");
|
|
else
|
|
snprintf(nextmsg,sizeof(nextmsg), "digits/a-m");
|
|
res = wait_file(chan,ints,nextmsg,lang);
|
|
break;
|
|
case 'Q':
|
|
/* Shorthand for "Today", "Yesterday", or AdBY */
|
|
{
|
|
struct timeval now;
|
|
struct tm tmnow;
|
|
time_t beg_today;
|
|
|
|
gettimeofday(&now,NULL);
|
|
ast_localtime(&now.tv_sec,&tmnow,timezone);
|
|
/* This might be slightly off, if we transcend a leap second, but never more off than 1 second */
|
|
/* In any case, it saves not having to do ast_mktime() */
|
|
beg_today = now.tv_sec - (tmnow.tm_hour * 3600) - (tmnow.tm_min * 60) - (tmnow.tm_sec);
|
|
if (beg_today < time) {
|
|
/* Today */
|
|
res = wait_file(chan,ints, "digits/today",lang);
|
|
} else if (beg_today - 86400 < time) {
|
|
/* Yesterday */
|
|
res = wait_file(chan,ints, "digits/yesterday",lang);
|
|
} else {
|
|
res = ast_say_date_with_format(chan, time, ints, lang, "AdBY", timezone);
|
|
}
|
|
}
|
|
break;
|
|
case 'q':
|
|
/* Shorthand for "" (today), "Yesterday", A (weekday), or AdBY */
|
|
{
|
|
struct timeval now;
|
|
struct tm tmnow;
|
|
time_t beg_today;
|
|
|
|
gettimeofday(&now,NULL);
|
|
ast_localtime(&now.tv_sec,&tmnow,timezone);
|
|
/* This might be slightly off, if we transcend a leap second, but never more off than 1 second */
|
|
/* In any case, it saves not having to do ast_mktime() */
|
|
beg_today = now.tv_sec - (tmnow.tm_hour * 3600) - (tmnow.tm_min * 60) - (tmnow.tm_sec);
|
|
if (beg_today < time) {
|
|
/* Today */
|
|
} else if ((beg_today - 86400) < time) {
|
|
/* Yesterday */
|
|
res = wait_file(chan,ints, "digits/yesterday",lang);
|
|
} else if (beg_today - 86400 * 6 < time) {
|
|
/* Within the last week */
|
|
res = ast_say_date_with_format(chan, time, ints, lang, "A", timezone);
|
|
} else {
|
|
res = ast_say_date_with_format(chan, time, ints, lang, "AdBY", timezone);
|
|
}
|
|
}
|
|
break;
|
|
case 'R':
|
|
res = ast_say_date_with_format(chan, time, ints, lang, "HM", timezone);
|
|
break;
|
|
case 'S':
|
|
/* Seconds */
|
|
res = wait_file(chan,ints, "digits/and",lang);
|
|
if (!res) {
|
|
res = ast_say_number(chan, tm.tm_sec, ints, lang, "f");
|
|
if (!res) {
|
|
res = wait_file(chan,ints, "digits/seconds",lang);
|
|
}
|
|
}
|
|
break;
|
|
case 'T':
|
|
res = ast_say_date_with_format(chan, time, ints, lang, "HMS", timezone);
|
|
break;
|
|
case ' ':
|
|
case ' ':
|
|
/* Just ignore spaces and tabs */
|
|
break;
|
|
default:
|
|
/* Unknown character */
|
|
ast_log(LOG_WARNING, "Unknown character in datetime format %s: %c at pos %d\n", format, format[offset], offset);
|
|
}
|
|
/* Jump out on DTMF */
|
|
if (res) {
|
|
break;
|
|
}
|
|
}
|
|
return res;
|
|
}
|
|
|
|
/* German syntax */
|
|
int ast_say_date_with_format_de(struct ast_channel *chan, time_t time, const char *ints, const char *lang, const char *format, const char *timezone)
|
|
{
|
|
struct tm tm;
|
|
int res=0, offset, sndoffset;
|
|
char sndfile[256], nextmsg[256];
|
|
|
|
ast_localtime(&time,&tm,timezone);
|
|
|
|
for (offset=0 ; format[offset] != '\0' ; offset++) {
|
|
ast_log(LOG_DEBUG, "Parsing %c (offset %d) in %s\n", format[offset], offset, format);
|
|
switch (format[offset]) {
|
|
/* NOTE: if you add more options here, please try to be consistent with strftime(3) */
|
|
case '\'':
|
|
/* Literal name of a sound file */
|
|
sndoffset=0;
|
|
for (sndoffset=0 ; (format[++offset] != '\'') && (sndoffset < 256) ; sndoffset++)
|
|
sndfile[sndoffset] = format[offset];
|
|
sndfile[sndoffset] = '\0';
|
|
res = wait_file(chan,ints,sndfile,lang);
|
|
break;
|
|
case 'A':
|
|
case 'a':
|
|
/* Sunday - Saturday */
|
|
snprintf(nextmsg,sizeof(nextmsg), "digits/day-%d", tm.tm_wday);
|
|
res = wait_file(chan,ints,nextmsg,lang);
|
|
break;
|
|
case 'B':
|
|
case 'b':
|
|
case 'h':
|
|
/* January - December */
|
|
snprintf(nextmsg,sizeof(nextmsg), "digits/mon-%d", tm.tm_mon);
|
|
res = wait_file(chan,ints,nextmsg,lang);
|
|
break;
|
|
case 'm':
|
|
/* Month enumerated */
|
|
res = ast_say_enumeration(chan, (tm.tm_mon + 1), ints, lang, "m");
|
|
break;
|
|
case 'd':
|
|
case 'e':
|
|
/* First - Thirtyfirst */
|
|
res = ast_say_enumeration(chan, tm.tm_mday, ints, lang, "m");
|
|
break;
|
|
case 'Y':
|
|
/* Year */
|
|
{
|
|
int year = tm.tm_year + 1900;
|
|
if (year > 1999) { /* year 2000 and later */
|
|
res = ast_say_number(chan, year, ints, lang, (char *) NULL);
|
|
} else {
|
|
if (year < 1100) {
|
|
/* I'm not going to handle 1100 and prior */
|
|
/* We'll just be silent on the year, instead of bombing out. */
|
|
} else {
|
|
/* year 1100 to 1999. will anybody need this?!? */
|
|
/* say 1967 as 'neunzehn hundert sieben und sechzig' */
|
|
snprintf(nextmsg,sizeof(nextmsg), "digits/%d", (year / 100) );
|
|
res = wait_file(chan,ints,nextmsg,lang);
|
|
if (!res) {
|
|
res = wait_file(chan,ints, "digits/hundred",lang);
|
|
if (!res && year % 100 != 0) {
|
|
res = ast_say_number(chan, (year % 100), ints, lang, (char *) NULL);
|
|
}
|
|
}
|
|
}
|
|
}
|
|
}
|
|
break;
|
|
case 'I':
|
|
case 'l':
|
|
/* 12-Hour */
|
|
if (tm.tm_hour == 0)
|
|
snprintf(nextmsg,sizeof(nextmsg), "digits/12");
|
|
else if (tm.tm_hour > 12)
|
|
snprintf(nextmsg,sizeof(nextmsg), "digits/%d", tm.tm_hour - 12);
|
|
else
|
|
snprintf(nextmsg,sizeof(nextmsg), "digits/%d", tm.tm_hour);
|
|
res = wait_file(chan,ints,nextmsg,lang);
|
|
if (!res) {
|
|
res = wait_file(chan,ints,"digits/oclock",lang);
|
|
}
|
|
break;
|
|
case 'H':
|
|
case 'k':
|
|
/* 24-Hour */
|
|
res = ast_say_number(chan, tm.tm_hour, ints, lang, (char *) NULL);
|
|
if (!res) {
|
|
res = wait_file(chan,ints,"digits/oclock",lang);
|
|
}
|
|
break;
|
|
case 'M':
|
|
/* Minute */
|
|
if (tm.tm_min > 0 || format[offset+ 1 ] == 'S' ) { /* zero 'digits/0' only if seconds follow (kind of a hack) */
|
|
res = ast_say_number(chan, tm.tm_min, ints, lang, "f");
|
|
}
|
|
if ( !res && format[offset + 1] == 'S' ) { /* minutes only if seconds follow (kind of a hack) */
|
|
if (tm.tm_min == 1) {
|
|
res = wait_file(chan,ints,"digits/minute",lang);
|
|
} else {
|
|
res = wait_file(chan,ints,"digits/minutes",lang);
|
|
}
|
|
}
|
|
break;
|
|
case 'P':
|
|
case 'p':
|
|
/* AM/PM */
|
|
if (tm.tm_hour > 11)
|
|
snprintf(nextmsg,sizeof(nextmsg), "digits/p-m");
|
|
else
|
|
snprintf(nextmsg,sizeof(nextmsg), "digits/a-m");
|
|
res = wait_file(chan,ints,nextmsg,lang);
|
|
break;
|
|
case 'Q':
|
|
/* Shorthand for "Today", "Yesterday", or AdBY */
|
|
{
|
|
struct timeval now;
|
|
struct tm tmnow;
|
|
time_t beg_today;
|
|
|
|
gettimeofday(&now,NULL);
|
|
ast_localtime(&now.tv_sec,&tmnow,timezone);
|
|
/* This might be slightly off, if we transcend a leap second, but never more off than 1 second */
|
|
/* In any case, it saves not having to do ast_mktime() */
|
|
beg_today = now.tv_sec - (tmnow.tm_hour * 3600) - (tmnow.tm_min * 60) - (tmnow.tm_sec);
|
|
if (beg_today < time) {
|
|
/* Today */
|
|
res = wait_file(chan,ints, "digits/today",lang);
|
|
} else if (beg_today - 86400 < time) {
|
|
/* Yesterday */
|
|
res = wait_file(chan,ints, "digits/yesterday",lang);
|
|
} else {
|
|
res = ast_say_date_with_format(chan, time, ints, lang, "AdBY", timezone);
|
|
}
|
|
}
|
|
break;
|
|
case 'q':
|
|
/* Shorthand for "" (today), "Yesterday", A (weekday), or AdBY */
|
|
{
|
|
struct timeval now;
|
|
struct tm tmnow;
|
|
time_t beg_today;
|
|
|
|
gettimeofday(&now,NULL);
|
|
ast_localtime(&now.tv_sec,&tmnow,timezone);
|
|
/* This might be slightly off, if we transcend a leap second, but never more off than 1 second */
|
|
/* In any case, it saves not having to do ast_mktime() */
|
|
beg_today = now.tv_sec - (tmnow.tm_hour * 3600) - (tmnow.tm_min * 60) - (tmnow.tm_sec);
|
|
if (beg_today < time) {
|
|
/* Today */
|
|
} else if ((beg_today - 86400) < time) {
|
|
/* Yesterday */
|
|
res = wait_file(chan,ints, "digits/yesterday",lang);
|
|
} else if (beg_today - 86400 * 6 < time) {
|
|
/* Within the last week */
|
|
res = ast_say_date_with_format(chan, time, ints, lang, "A", timezone);
|
|
} else {
|
|
res = ast_say_date_with_format(chan, time, ints, lang, "AdBY", timezone);
|
|
}
|
|
}
|
|
break;
|
|
case 'R':
|
|
res = ast_say_date_with_format(chan, time, ints, lang, "HM", timezone);
|
|
break;
|
|
case 'S':
|
|
/* Seconds */
|
|
res = wait_file(chan,ints, "digits/and",lang);
|
|
if (!res) {
|
|
res = ast_say_number(chan, tm.tm_sec, ints, lang, "f");
|
|
if (!res) {
|
|
res = wait_file(chan,ints, "digits/seconds",lang);
|
|
}
|
|
}
|
|
break;
|
|
case 'T':
|
|
res = ast_say_date_with_format(chan, time, ints, lang, "HMS", timezone);
|
|
break;
|
|
case ' ':
|
|
case ' ':
|
|
/* Just ignore spaces and tabs */
|
|
break;
|
|
default:
|
|
/* Unknown character */
|
|
ast_log(LOG_WARNING, "Unknown character in datetime format %s: %c at pos %d\n", format, format[offset], offset);
|
|
}
|
|
/* Jump out on DTMF */
|
|
if (res) {
|
|
break;
|
|
}
|
|
}
|
|
return res;
|
|
}
|
|
|
|
/* TODO: this probably is not the correct format for doxygen remarks */
|
|
|
|
/** ast_say_date_with_format_he Say formmated date in Hebrew
|
|
*
|
|
* @seealso ast_say_date_with_format_en for the details of the options
|
|
*
|
|
* Changes from the English version:
|
|
*
|
|
* * don't replicate in here the logic of ast_say_number_full_he
|
|
*
|
|
* * year is always 4-digit (because it's simpler)
|
|
*
|
|
* * added c, x, and X. Mainly for my tests
|
|
*
|
|
* * The standard "long" format used in Hebrew is AdBY, rather than ABdY
|
|
*
|
|
* TODO:
|
|
* * A "ha" is missing in the standard date format, before the 'd'.
|
|
* * The numbers of 3000--19000 are not handled well
|
|
**/
|
|
#define IL_DATE_STR "AdBY"
|
|
#define IL_TIME_STR "IMp"
|
|
#define IL_DATE_STR_FULL IL_DATE_STR " 'digits/at' " IL_TIME_STR
|
|
int ast_say_date_with_format_he(struct ast_channel *chan, time_t time,
|
|
const char *ints, const char *lang, const char *format,
|
|
const char *timezone)
|
|
{
|
|
/* TODO: This whole function is cut&paste from
|
|
* ast_say_date_with_format_en . Is that considered acceptable?
|
|
**/
|
|
struct tm tm;
|
|
int res=0, offset, sndoffset;
|
|
char sndfile[256], nextmsg[256];
|
|
|
|
ast_localtime(&time,&tm,timezone);
|
|
|
|
for (offset=0 ; format[offset] != '\0' ; offset++) {
|
|
ast_log(LOG_DEBUG, "Parsing %c (offset %d) in %s\n", format[offset], offset, format);
|
|
switch (format[offset]) {
|
|
/* NOTE: if you add more options here, please try to be consistent with strftime(3) */
|
|
case '\'':
|
|
/* Literal name of a sound file */
|
|
sndoffset=0;
|
|
for (sndoffset=0 ; (format[++offset] != '\'') && (sndoffset < 256) ; sndoffset++)
|
|
sndfile[sndoffset] = format[offset];
|
|
sndfile[sndoffset] = '\0';
|
|
res = wait_file(chan,ints,sndfile,lang);
|
|
break;
|
|
case 'A':
|
|
case 'a':
|
|
/* Sunday - Saturday */
|
|
snprintf(nextmsg,sizeof(nextmsg), "digits/day-%d", tm.tm_wday);
|
|
res = wait_file(chan,ints,nextmsg,lang);
|
|
break;
|
|
case 'B':
|
|
case 'b':
|
|
case 'h':
|
|
/* January - December */
|
|
snprintf(nextmsg,sizeof(nextmsg), "digits/mon-%d", tm.tm_mon);
|
|
res = wait_file(chan,ints,nextmsg,lang);
|
|
break;
|
|
case 'd':
|
|
case 'e': /* Day of the month */
|
|
/* I'm not sure exactly what the parameters
|
|
* audiofd and ctrlfd to
|
|
* ast_say_number_full_he mean, but it seems
|
|
* safe to pass -1 there.
|
|
*
|
|
* At least in one of the pathes :-(
|
|
*/
|
|
res = ast_say_number_full_he(chan, tm.tm_mday,
|
|
ints, lang, "m", -1, -1
|
|
);
|
|
break;
|
|
case 'Y': /* Year */
|
|
res = ast_say_number_full_he(chan, tm.tm_year+1900,
|
|
ints, lang, "f", -1, -1
|
|
);
|
|
break;
|
|
case 'I':
|
|
case 'l': /* 12-Hour */
|
|
{
|
|
int hour = tm.tm_hour;
|
|
hour = hour%12;
|
|
if (hour == 0) hour=12;
|
|
|
|
res = ast_say_number_full_he(chan, hour,
|
|
ints, lang, "f", -1, -1
|
|
);
|
|
}
|
|
break;
|
|
case 'H':
|
|
case 'k': /* 24-Hour */
|
|
/* With 'H' there is an 'oh' after a single-
|
|
* digit hour */
|
|
if ((format[offset] == 'H') &&
|
|
(tm.tm_hour <10)&&(tm.tm_hour>0)
|
|
) { /* e.g. oh-eight */
|
|
res = wait_file(chan,ints, "digits/oh",lang);
|
|
}
|
|
|
|
res = ast_say_number_full_he(chan, tm.tm_hour,
|
|
ints, lang, "f", -1, -1
|
|
);
|
|
break;
|
|
case 'M': /* Minute */
|
|
res = ast_say_number_full_he(chan, tm.tm_min,
|
|
ints, lang,"f", -1, -1
|
|
);
|
|
break;
|
|
case 'P':
|
|
case 'p':
|
|
/* AM/PM */
|
|
if (tm.tm_hour > 11)
|
|
snprintf(nextmsg,sizeof(nextmsg), "digits/p-m");
|
|
else
|
|
snprintf(nextmsg,sizeof(nextmsg), "digits/a-m");
|
|
res = wait_file(chan,ints,nextmsg,lang);
|
|
break;
|
|
case 'Q':
|
|
/* Shorthand for "Today", "Yesterday", or "date" */
|
|
case 'q':
|
|
/* Shorthand for "" (today), "Yesterday", A
|
|
* (weekday), or "date" */
|
|
{
|
|
struct timeval now;
|
|
struct tm tmnow;
|
|
time_t beg_today;
|
|
char todo = format[offset]; /* The letter to format*/
|
|
|
|
gettimeofday(&now,NULL);
|
|
ast_localtime(&now.tv_sec,&tmnow,timezone);
|
|
/* This might be slightly off, if we transcend a leap second, but never more off than 1 second */
|
|
/* In any case, it saves not having to do ast_mktime() */
|
|
beg_today = now.tv_sec - (tmnow.tm_hour * 3600) - (tmnow.tm_min * 60) - (tmnow.tm_sec);
|
|
if (beg_today < time) {
|
|
/* Today */
|
|
if (todo == 'Q') {
|
|
res = wait_file(chan,
|
|
ints,
|
|
"digits/today",
|
|
lang);
|
|
}
|
|
} else if (beg_today - 86400 < time) {
|
|
/* Yesterday */
|
|
res = wait_file(chan,ints, "digits/yesterday",lang);
|
|
} else if ((todo != 'Q') &&
|
|
(beg_today - 86400 * 6 < time))
|
|
{
|
|
/* Within the last week */
|
|
res = ast_say_date_with_format_he(chan,
|
|
time, ints, lang,
|
|
"A", timezone);
|
|
} else {
|
|
res = ast_say_date_with_format_he(chan,
|
|
time, ints, lang,
|
|
IL_DATE_STR, timezone);
|
|
}
|
|
}
|
|
break;
|
|
case 'R':
|
|
res = ast_say_date_with_format(chan, time, ints, lang, "HM", timezone);
|
|
break;
|
|
case 'S': /* Seconds */
|
|
res = ast_say_number_full_he(chan, tm.tm_sec,
|
|
ints, lang, "f", -1, -1
|
|
);
|
|
break;
|
|
case 'T':
|
|
res = ast_say_date_with_format(chan, time, ints, lang, "HMS", timezone);
|
|
break;
|
|
/* c, x, and X seem useful for testing. Not sure
|
|
* if thiey're good for the general public */
|
|
case 'c':
|
|
res = ast_say_date_with_format_he(chan, time,
|
|
ints, lang, IL_DATE_STR_FULL, timezone);
|
|
break;
|
|
case 'x':
|
|
res = ast_say_date_with_format_he(chan, time,
|
|
ints, lang, IL_DATE_STR, timezone);
|
|
break;
|
|
case 'X': /* Currently not locale-dependent...*/
|
|
res = ast_say_date_with_format_he(chan, time,
|
|
ints, lang, IL_TIME_STR, timezone);
|
|
break;
|
|
case ' ':
|
|
case ' ':
|
|
/* Just ignore spaces and tabs */
|
|
break;
|
|
default:
|
|
/* Unknown character */
|
|
ast_log(LOG_WARNING, "Unknown character in datetime format %s: %c at pos %d\n", format, format[offset], offset);
|
|
}
|
|
/* Jump out on DTMF */
|
|
if (res) {
|
|
break;
|
|
}
|
|
}
|
|
return res;
|
|
}
|
|
|
|
|
|
/* Spanish syntax */
|
|
int ast_say_date_with_format_es(struct ast_channel *chan, time_t time, const char *ints, const char *lang, const char *format, const char *timezone)
|
|
{
|
|
struct tm tm;
|
|
int res=0, offset, sndoffset;
|
|
char sndfile[256], nextmsg[256];
|
|
|
|
ast_localtime(&time,&tm,timezone);
|
|
|
|
for (offset=0 ; format[offset] != '\0' ; offset++) {
|
|
ast_log(LOG_DEBUG, "Parsing %c (offset %d) in %s\n", format[offset], offset, format);
|
|
switch (format[offset]) {
|
|
/* NOTE: if you add more options here, please try to be consistent with strftime(3) */
|
|
case '\'':
|
|
/* Literal name of a sound file */
|
|
sndoffset=0;
|
|
for (sndoffset=0 ; (format[++offset] != '\'') && (sndoffset < 256) ; sndoffset++)
|
|
sndfile[sndoffset] = format[offset];
|
|
sndfile[sndoffset] = '\0';
|
|
snprintf(nextmsg,sizeof(nextmsg), "%s", sndfile);
|
|
res = wait_file(chan,ints,nextmsg,lang);
|
|
break;
|
|
case 'A':
|
|
case 'a':
|
|
/* Sunday - Saturday */
|
|
snprintf(nextmsg,sizeof(nextmsg), "digits/day-%d", tm.tm_wday);
|
|
res = wait_file(chan,ints,nextmsg,lang);
|
|
break;
|
|
case 'B':
|
|
case 'b':
|
|
case 'h':
|
|
/* January - December */
|
|
snprintf(nextmsg,sizeof(nextmsg), "digits/mon-%d", tm.tm_mon);
|
|
res = wait_file(chan,ints,nextmsg,lang);
|
|
break;
|
|
case 'm':
|
|
/* First - Twelfth */
|
|
snprintf(nextmsg,sizeof(nextmsg), "digits/h-%d", tm.tm_mon +1);
|
|
res = wait_file(chan,ints,nextmsg,lang);
|
|
break;
|
|
case 'd':
|
|
case 'e':
|
|
/* First - Thirtyfirst */
|
|
res = ast_say_number(chan, tm.tm_mday, ints, lang, (char *) NULL);
|
|
break;
|
|
case 'Y':
|
|
/* Year */
|
|
res = ast_say_number(chan, tm.tm_year + 1900, ints, lang, (char *) NULL);
|
|
break;
|
|
case 'I':
|
|
case 'l':
|
|
/* 12-Hour */
|
|
if (tm.tm_hour == 0)
|
|
snprintf(nextmsg,sizeof(nextmsg), "digits/12");
|
|
else if (tm.tm_hour > 12)
|
|
snprintf(nextmsg,sizeof(nextmsg), "digits/%d", tm.tm_hour - 12);
|
|
else
|
|
snprintf(nextmsg,sizeof(nextmsg), "digits/%d", tm.tm_hour);
|
|
res = wait_file(chan,ints,nextmsg,lang);
|
|
break;
|
|
case 'H':
|
|
case 'k':
|
|
/* 24-Hour */
|
|
res = ast_say_number(chan, tm.tm_hour, ints, lang, NULL);
|
|
break;
|
|
case 'M':
|
|
/* Minute */
|
|
res = ast_say_number(chan, tm.tm_min, ints, lang, (char *) NULL);
|
|
break;
|
|
case 'P':
|
|
case 'p':
|
|
/* AM/PM */
|
|
if (tm.tm_hour > 12)
|
|
res = wait_file(chan, ints, "digits/p-m", lang);
|
|
else if (tm.tm_hour && tm.tm_hour < 12)
|
|
res = wait_file(chan, ints, "digits/a-m", lang);
|
|
break;
|
|
case 'Q':
|
|
/* Shorthand for "Today", "Yesterday", or ABdY */
|
|
{
|
|
struct timeval now;
|
|
struct tm tmnow;
|
|
time_t beg_today;
|
|
|
|
gettimeofday(&now,NULL);
|
|
ast_localtime(&now.tv_sec,&tmnow,timezone);
|
|
/* This might be slightly off, if we transcend a leap second, but never more off than 1 second */
|
|
/* In any case, it saves not having to do ast_mktime() */
|
|
beg_today = now.tv_sec - (tmnow.tm_hour * 3600) - (tmnow.tm_min * 60) - (tmnow.tm_sec);
|
|
if (beg_today < time) {
|
|
/* Today */
|
|
res = wait_file(chan,ints, "digits/today",lang);
|
|
} else if (beg_today - 86400 < time) {
|
|
/* Yesterday */
|
|
res = wait_file(chan,ints, "digits/yesterday",lang);
|
|
} else {
|
|
res = ast_say_date_with_format(chan, time, ints, lang, "'digits/es-el' Ad 'digits/es-de' B 'digits/es-de' Y", timezone);
|
|
}
|
|
}
|
|
break;
|
|
case 'q':
|
|
/* Shorthand for "" (today), "Yesterday", A (weekday), or ABdY */
|
|
{
|
|
struct timeval now;
|
|
struct tm tmnow;
|
|
time_t beg_today;
|
|
|
|
gettimeofday(&now,NULL);
|
|
ast_localtime(&now.tv_sec,&tmnow,timezone);
|
|
/* This might be slightly off, if we transcend a leap second, but never more off than 1 second */
|
|
/* In any case, it saves not having to do ast_mktime() */
|
|
beg_today = now.tv_sec - (tmnow.tm_hour * 3600) - (tmnow.tm_min * 60) - (tmnow.tm_sec);
|
|
if (beg_today < time) {
|
|
/* Today */
|
|
res = wait_file(chan,ints, "digits/today",lang);
|
|
} else if ((beg_today - 86400) < time) {
|
|
/* Yesterday */
|
|
res = wait_file(chan,ints, "digits/yesterday",lang);
|
|
} else if (beg_today - 86400 * 6 < time) {
|
|
/* Within the last week */
|
|
res = ast_say_date_with_format(chan, time, ints, lang, "A", timezone);
|
|
} else {
|
|
res = ast_say_date_with_format(chan, time, ints, lang, "'digits/es-el' Ad 'digits/es-de' B 'digits/es-de' Y", timezone);
|
|
}
|
|
}
|
|
break;
|
|
case 'R':
|
|
res = ast_say_date_with_format(chan, time, ints, lang, "H 'digits/y' M", timezone);
|
|
break;
|
|
case 'S':
|
|
/* Seconds */
|
|
if (tm.tm_sec == 0) {
|
|
snprintf(nextmsg,sizeof(nextmsg), "digits/%d", tm.tm_sec);
|
|
res = wait_file(chan,ints,nextmsg,lang);
|
|
} else if (tm.tm_sec < 10) {
|
|
res = wait_file(chan,ints, "digits/oh",lang);
|
|
if (!res) {
|
|
snprintf(nextmsg,sizeof(nextmsg), "digits/%d", tm.tm_sec);
|
|
res = wait_file(chan,ints,nextmsg,lang);
|
|
}
|
|
} else if ((tm.tm_sec < 21) || (tm.tm_sec % 10 == 0)) {
|
|
snprintf(nextmsg,sizeof(nextmsg), "digits/%d", tm.tm_sec);
|
|
res = wait_file(chan,ints,nextmsg,lang);
|
|
} else {
|
|
int ten, one;
|
|
ten = (tm.tm_sec / 10) * 10;
|
|
one = (tm.tm_sec % 10);
|
|
snprintf(nextmsg,sizeof(nextmsg), "digits/%d", ten);
|
|
res = wait_file(chan,ints,nextmsg,lang);
|
|
if (!res) {
|
|
/* Fifty, not fifty-zero */
|
|
if (one != 0) {
|
|
snprintf(nextmsg,sizeof(nextmsg), "digits/%d", one);
|
|
res = wait_file(chan,ints,nextmsg,lang);
|
|
}
|
|
}
|
|
}
|
|
break;
|
|
case 'T':
|
|
res = ast_say_date_with_format(chan, time, ints, lang, "HMS", timezone);
|
|
break;
|
|
case ' ':
|
|
case ' ':
|
|
/* Just ignore spaces and tabs */
|
|
break;
|
|
default:
|
|
/* Unknown character */
|
|
ast_log(LOG_WARNING, "Unknown character in datetime format %s: %c at pos %d\n", format, format[offset], offset);
|
|
}
|
|
/* Jump out on DTMF */
|
|
if (res) {
|
|
break;
|
|
}
|
|
}
|
|
return res;
|
|
}
|
|
|
|
/* French syntax
|
|
oclock = heure
|
|
*/
|
|
int ast_say_date_with_format_fr(struct ast_channel *chan, time_t time, const char *ints, const char *lang, const char *format, const char *timezone)
|
|
{
|
|
struct tm tm;
|
|
int res=0, offset, sndoffset;
|
|
char sndfile[256], nextmsg[256];
|
|
|
|
ast_localtime(&time,&tm,timezone);
|
|
|
|
for (offset=0 ; format[offset] != '\0' ; offset++) {
|
|
ast_log(LOG_DEBUG, "Parsing %c (offset %d) in %s\n", format[offset], offset, format);
|
|
switch (format[offset]) {
|
|
/* NOTE: if you add more options here, please try to be consistent with strftime(3) */
|
|
case '\'':
|
|
/* Literal name of a sound file */
|
|
sndoffset=0;
|
|
for (sndoffset=0 ; (format[++offset] != '\'') && (sndoffset < 256) ; sndoffset++)
|
|
sndfile[sndoffset] = format[offset];
|
|
sndfile[sndoffset] = '\0';
|
|
res = wait_file(chan,ints,sndfile,lang);
|
|
break;
|
|
case 'A':
|
|
case 'a':
|
|
/* Sunday - Saturday */
|
|
snprintf(nextmsg,sizeof(nextmsg), "digits/day-%d", tm.tm_wday);
|
|
res = wait_file(chan,ints,nextmsg,lang);
|
|
break;
|
|
case 'B':
|
|
case 'b':
|
|
case 'h':
|
|
/* January - December */
|
|
snprintf(nextmsg,sizeof(nextmsg), "digits/mon-%d", tm.tm_mon);
|
|
res = wait_file(chan,ints,nextmsg,lang);
|
|
break;
|
|
case 'm':
|
|
/* First - Twelfth */
|
|
snprintf(nextmsg,sizeof(nextmsg), "digits/h-%d", tm.tm_mon +1);
|
|
res = wait_file(chan,ints,nextmsg,lang);
|
|
break;
|
|
case 'd':
|
|
case 'e':
|
|
/* First */
|
|
if (tm.tm_mday == 1) {
|
|
snprintf(nextmsg,sizeof(nextmsg), "digits/h-%d", tm.tm_mday);
|
|
res = wait_file(chan,ints,nextmsg,lang);
|
|
} else {
|
|
res = ast_say_number(chan, tm.tm_mday, ints, lang, (char * ) NULL);
|
|
}
|
|
break;
|
|
case 'Y':
|
|
/* Year */
|
|
if (tm.tm_year > 99) {
|
|
res = wait_file(chan,ints, "digits/2",lang);
|
|
if (!res) {
|
|
res = wait_file(chan,ints, "digits/thousand",lang);
|
|
}
|
|
if (tm.tm_year > 100) {
|
|
if (!res) {
|
|
res = ast_say_number(chan, tm.tm_year - 100, ints, lang, (char * ) NULL);
|
|
}
|
|
}
|
|
} else {
|
|
if (tm.tm_year < 1) {
|
|
/* I'm not going to handle 1900 and prior */
|
|
/* We'll just be silent on the year, instead of bombing out. */
|
|
} else {
|
|
res = wait_file(chan,ints, "digits/thousand",lang);
|
|
if (!res) {
|
|
wait_file(chan,ints, "digits/9",lang);
|
|
wait_file(chan,ints, "digits/hundred",lang);
|
|
res = ast_say_number(chan, tm.tm_year, ints, lang, (char * ) NULL);
|
|
}
|
|
}
|
|
}
|
|
break;
|
|
case 'I':
|
|
case 'l':
|
|
/* 12-Hour */
|
|
if (tm.tm_hour == 0)
|
|
snprintf(nextmsg,sizeof(nextmsg), "digits/12");
|
|
else if (tm.tm_hour > 12)
|
|
snprintf(nextmsg,sizeof(nextmsg), "digits/%d", tm.tm_hour - 12);
|
|
else
|
|
snprintf(nextmsg,sizeof(nextmsg), "digits/%d", tm.tm_hour);
|
|
res = wait_file(chan,ints,nextmsg,lang);
|
|
if (!res)
|
|
res = wait_file(chan,ints, "digits/oclock",lang);
|
|
break;
|
|
case 'H':
|
|
case 'k':
|
|
/* 24-Hour */
|
|
res = ast_say_number(chan, tm.tm_hour, ints, lang, (char * ) NULL);
|
|
if (!res) {
|
|
if (format[offset] == 'H') {
|
|
res = wait_file(chan,ints, "digits/oclock",lang);
|
|
}
|
|
}
|
|
if (!res)
|
|
res = wait_file(chan,ints, "digits/oclock",lang);
|
|
break;
|
|
case 'M':
|
|
/* Minute */
|
|
if (tm.tm_min == 0) {
|
|
break;
|
|
}
|
|
res = ast_say_number(chan, tm.tm_min, ints, lang, (char * ) NULL);
|
|
break;
|
|
case 'P':
|
|
case 'p':
|
|
/* AM/PM */
|
|
if (tm.tm_hour > 11)
|
|
snprintf(nextmsg,sizeof(nextmsg), "digits/p-m");
|
|
else
|
|
snprintf(nextmsg,sizeof(nextmsg), "digits/a-m");
|
|
res = wait_file(chan,ints,nextmsg,lang);
|
|
break;
|
|
case 'Q':
|
|
/* Shorthand for "Today", "Yesterday", or AdBY */
|
|
{
|
|
struct timeval now;
|
|
struct tm tmnow;
|
|
time_t beg_today;
|
|
|
|
gettimeofday(&now,NULL);
|
|
ast_localtime(&now.tv_sec,&tmnow,timezone);
|
|
/* This might be slightly off, if we transcend a leap second, but never more off than 1 second */
|
|
/* In any case, it saves not having to do ast_mktime() */
|
|
beg_today = now.tv_sec - (tmnow.tm_hour * 3600) - (tmnow.tm_min * 60) - (tmnow.tm_sec);
|
|
if (beg_today < time) {
|
|
/* Today */
|
|
res = wait_file(chan,ints, "digits/today",lang);
|
|
} else if (beg_today - 86400 < time) {
|
|
/* Yesterday */
|
|
res = wait_file(chan,ints, "digits/yesterday",lang);
|
|
} else {
|
|
res = ast_say_date_with_format(chan, time, ints, lang, "AdBY", timezone);
|
|
}
|
|
}
|
|
break;
|
|
case 'q':
|
|
/* Shorthand for "" (today), "Yesterday", A (weekday), or AdBY */
|
|
{
|
|
struct timeval now;
|
|
struct tm tmnow;
|
|
time_t beg_today;
|
|
|
|
gettimeofday(&now,NULL);
|
|
ast_localtime(&now.tv_sec,&tmnow,timezone);
|
|
/* This might be slightly off, if we transcend a leap second, but never more off than 1 second */
|
|
/* In any case, it saves not having to do ast_mktime() */
|
|
beg_today = now.tv_sec - (tmnow.tm_hour * 3600) - (tmnow.tm_min * 60) - (tmnow.tm_sec);
|
|
if (beg_today < time) {
|
|
/* Today */
|
|
} else if ((beg_today - 86400) < time) {
|
|
/* Yesterday */
|
|
res = wait_file(chan,ints, "digits/yesterday",lang);
|
|
} else if (beg_today - 86400 * 6 < time) {
|
|
/* Within the last week */
|
|
res = ast_say_date_with_format(chan, time, ints, lang, "A", timezone);
|
|
} else {
|
|
res = ast_say_date_with_format(chan, time, ints, lang, "AdBY", timezone);
|
|
}
|
|
}
|
|
break;
|
|
case 'R':
|
|
res = ast_say_date_with_format(chan, time, ints, lang, "HM", timezone);
|
|
break;
|
|
case 'S':
|
|
/* Seconds */
|
|
res = ast_say_number(chan, tm.tm_hour, ints, lang, (char * ) NULL);
|
|
if (!res) {
|
|
res = wait_file(chan,ints, "digits/second",lang);
|
|
}
|
|
break;
|
|
case 'T':
|
|
res = ast_say_date_with_format(chan, time, ints, lang, "HMS", timezone);
|
|
break;
|
|
case ' ':
|
|
case ' ':
|
|
/* Just ignore spaces and tabs */
|
|
break;
|
|
default:
|
|
/* Unknown character */
|
|
ast_log(LOG_WARNING, "Unknown character in datetime format %s: %c at pos %d\n", format, format[offset], offset);
|
|
}
|
|
/* Jump out on DTMF */
|
|
if (res) {
|
|
break;
|
|
}
|
|
}
|
|
return res;
|
|
}
|
|
|
|
int ast_say_date_with_format_it(struct ast_channel *chan, time_t time, const char *ints, const char *lang, const char *format, const char *timezone)
|
|
{
|
|
struct tm tm;
|
|
int res=0, offset, sndoffset;
|
|
char sndfile[256], nextmsg[256];
|
|
|
|
ast_localtime(&time,&tm,timezone);
|
|
|
|
for (offset=0 ; format[offset] != '\0' ; offset++) {
|
|
ast_log(LOG_DEBUG, "Parsing %c (offset %d) in %s\n", format[offset], offset, format);
|
|
switch (format[offset]) {
|
|
/* NOTE: if you add more options here, please try to be consistent with strftime(3) */
|
|
case '\'':
|
|
/* Literal name of a sound file */
|
|
sndoffset=0;
|
|
for (sndoffset=0 ; (format[++offset] != '\'') && (sndoffset < 256) ; sndoffset++)
|
|
sndfile[sndoffset] = format[offset];
|
|
sndfile[sndoffset] = '\0';
|
|
res = wait_file(chan,ints,sndfile,lang);
|
|
break;
|
|
case 'A':
|
|
case 'a':
|
|
/* Sunday - Saturday */
|
|
snprintf(nextmsg,sizeof(nextmsg), "digits/day-%d", tm.tm_wday);
|
|
res = wait_file(chan,ints,nextmsg,lang);
|
|
break;
|
|
case 'B':
|
|
case 'b':
|
|
case 'h':
|
|
/* January - December */
|
|
snprintf(nextmsg,sizeof(nextmsg), "digits/mon-%d", tm.tm_mon);
|
|
res = wait_file(chan,ints,nextmsg,lang);
|
|
break;
|
|
case 'm':
|
|
/* First - Twelfth */
|
|
snprintf(nextmsg,sizeof(nextmsg), "digits/h-%d", tm.tm_mon +1);
|
|
res = wait_file(chan,ints,nextmsg,lang);
|
|
break;
|
|
case 'd':
|
|
case 'e':
|
|
/* First day of the month is spelled as ordinal */
|
|
if (tm.tm_mday == 1) {
|
|
snprintf(nextmsg,sizeof(nextmsg), "digits/h-%d", tm.tm_mday);
|
|
res = wait_file(chan,ints,nextmsg,lang);
|
|
} else {
|
|
if (!res) {
|
|
res = ast_say_number(chan, tm.tm_mday, ints, lang, (char *) NULL);
|
|
}
|
|
}
|
|
break;
|
|
case 'Y':
|
|
/* Year */
|
|
if (tm.tm_year > 99) {
|
|
res = wait_file(chan,ints, "digits/ore-2000",lang);
|
|
if (tm.tm_year > 100) {
|
|
if (!res) {
|
|
/* This works until the end of 2021 */
|
|
snprintf(nextmsg,sizeof(nextmsg), "digits/%d", tm.tm_year - 100);
|
|
res = wait_file(chan,ints,nextmsg,lang);
|
|
}
|
|
}
|
|
} else {
|
|
if (tm.tm_year < 1) {
|
|
/* I'm not going to handle 1900 and prior */
|
|
/* We'll just be silent on the year, instead of bombing out. */
|
|
} else {
|
|
res = wait_file(chan,ints, "digits/ore-1900",lang);
|
|
if ((!res) && (tm.tm_year != 0)) {
|
|
if (tm.tm_year <= 21) {
|
|
/* 1910 - 1921 */
|
|
snprintf(nextmsg,sizeof(nextmsg), "digits/%d", tm.tm_year);
|
|
res = wait_file(chan,ints,nextmsg,lang);
|
|
} else {
|
|
/* 1922 - 1999, but sounds badly in 1928, 1931, 1938, etc... */
|
|
int ten, one;
|
|
ten = tm.tm_year / 10;
|
|
one = tm.tm_year % 10;
|
|
snprintf(nextmsg,sizeof(nextmsg), "digits/%d", ten * 10);
|
|
res = wait_file(chan,ints,nextmsg,lang);
|
|
if (!res) {
|
|
if (one != 0) {
|
|
snprintf(nextmsg,sizeof(nextmsg), "digits/%d", one);
|
|
res = wait_file(chan,ints,nextmsg,lang);
|
|
}
|
|
}
|
|
}
|
|
}
|
|
}
|
|
}
|
|
break;
|
|
case 'I':
|
|
case 'l':
|
|
/* 12-Hour */
|
|
if (tm.tm_hour == 0)
|
|
snprintf(nextmsg,sizeof(nextmsg), "digits/12");
|
|
else if (tm.tm_hour > 12)
|
|
snprintf(nextmsg,sizeof(nextmsg), "digits/%d", tm.tm_hour - 12);
|
|
else
|
|
snprintf(nextmsg,sizeof(nextmsg), "digits/%d", tm.tm_hour);
|
|
res = wait_file(chan,ints,nextmsg,lang);
|
|
break;
|
|
case 'H':
|
|
case 'k':
|
|
/* 24-Hour */
|
|
if (tm.tm_hour == 0) {
|
|
res = wait_file(chan,ints, "digits/ore-mezzanotte",lang);
|
|
} else if (tm.tm_hour == 1) {
|
|
res = wait_file(chan,ints, "digits/ore-una",lang);
|
|
} else {
|
|
res = ast_say_number(chan, tm.tm_hour, ints, lang, (char *) NULL);
|
|
}
|
|
break;
|
|
case 'M':
|
|
/* Minute */
|
|
res = ast_say_number(chan, tm.tm_min, ints, lang, (char *) NULL);
|
|
break;
|
|
case 'P':
|
|
case 'p':
|
|
/* AM/PM */
|
|
if (tm.tm_hour > 11)
|
|
snprintf(nextmsg,sizeof(nextmsg), "digits/p-m");
|
|
else
|
|
snprintf(nextmsg,sizeof(nextmsg), "digits/a-m");
|
|
res = wait_file(chan,ints,nextmsg,lang);
|
|
break;
|
|
case 'Q':
|
|
/* Shorthand for "Today", "Yesterday", or ABdY */
|
|
{
|
|
struct timeval now;
|
|
struct tm tmnow;
|
|
time_t beg_today;
|
|
|
|
gettimeofday(&now,NULL);
|
|
ast_localtime(&now.tv_sec,&tmnow,timezone);
|
|
/* This might be slightly off, if we transcend a leap second, but never more off than 1 second */
|
|
/* In any case, it saves not having to do ast_mktime() */
|
|
beg_today = now.tv_sec - (tmnow.tm_hour * 3600) - (tmnow.tm_min * 60) - (tmnow.tm_sec);
|
|
if (beg_today < time) {
|
|
/* Today */
|
|
res = wait_file(chan,ints, "digits/today",lang);
|
|
} else if (beg_today - 86400 < time) {
|
|
/* Yesterday */
|
|
res = wait_file(chan,ints, "digits/yesterday",lang);
|
|
} else {
|
|
res = ast_say_date_with_format(chan, time, ints, lang, "AdB", timezone);
|
|
}
|
|
}
|
|
break;
|
|
case 'q':
|
|
/* Shorthand for "" (today), "Yesterday", A (weekday), or ABdY */
|
|
{
|
|
struct timeval now;
|
|
struct tm tmnow;
|
|
time_t beg_today;
|
|
|
|
gettimeofday(&now,NULL);
|
|
ast_localtime(&now.tv_sec,&tmnow,timezone);
|
|
/* This might be slightly off, if we transcend a leap second, but never more off than 1 second */
|
|
/* In any case, it saves not having to do ast_mktime() */
|
|
beg_today = now.tv_sec - (tmnow.tm_hour * 3600) - (tmnow.tm_min * 60) - (tmnow.tm_sec);
|
|
if (beg_today < time) {
|
|
/* Today */
|
|
} else if ((beg_today - 86400) < time) {
|
|
/* Yesterday */
|
|
res = wait_file(chan,ints, "digits/yesterday",lang);
|
|
} else if (beg_today - 86400 * 6 < time) {
|
|
/* Within the last week */
|
|
res = ast_say_date_with_format(chan, time, ints, lang, "A", timezone);
|
|
} else {
|
|
res = ast_say_date_with_format(chan, time, ints, lang, "AdB", timezone);
|
|
}
|
|
}
|
|
break;
|
|
case 'R':
|
|
res = ast_say_date_with_format(chan, time, ints, lang, "HM", timezone);
|
|
break;
|
|
case 'S':
|
|
/* Seconds */
|
|
if (tm.tm_sec == 0) {
|
|
snprintf(nextmsg,sizeof(nextmsg), "digits/%d", tm.tm_sec);
|
|
res = wait_file(chan,ints,nextmsg,lang);
|
|
} else if (tm.tm_sec < 10) {
|
|
res = wait_file(chan,ints, "digits/oh",lang);
|
|
if (!res) {
|
|
snprintf(nextmsg,sizeof(nextmsg), "digits/%d", tm.tm_sec);
|
|
res = wait_file(chan,ints,nextmsg,lang);
|
|
}
|
|
} else if ((tm.tm_sec < 21) || (tm.tm_sec % 10 == 0)) {
|
|
snprintf(nextmsg,sizeof(nextmsg), "digits/%d", tm.tm_sec);
|
|
res = wait_file(chan,ints,nextmsg,lang);
|
|
} else {
|
|
int ten, one;
|
|
ten = (tm.tm_sec / 10) * 10;
|
|
one = (tm.tm_sec % 10);
|
|
snprintf(nextmsg,sizeof(nextmsg), "digits/%d", ten);
|
|
res = wait_file(chan,ints,nextmsg,lang);
|
|
if (!res) {
|
|
/* Fifty, not fifty-zero */
|
|
if (one != 0) {
|
|
snprintf(nextmsg,sizeof(nextmsg), "digits/%d", one);
|
|
res = wait_file(chan,ints,nextmsg,lang);
|
|
}
|
|
}
|
|
}
|
|
break;
|
|
case 'T':
|
|
res = ast_say_date_with_format(chan, time, ints, lang, "HMS", timezone);
|
|
break;
|
|
case ' ':
|
|
case ' ':
|
|
/* Just ignore spaces and tabs */
|
|
break;
|
|
default:
|
|
/* Unknown character */
|
|
ast_log(LOG_WARNING, "Unknown character in datetime format %s: %c at pos %d\n", format, format[offset], offset);
|
|
}
|
|
/* Jump out on DTMF */
|
|
if (res) {
|
|
break;
|
|
}
|
|
}
|
|
return res;
|
|
}
|
|
|
|
/* Dutch syntax */
|
|
int ast_say_date_with_format_nl(struct ast_channel *chan, time_t time, const char *ints, const char *lang, const char *format, const char *timezone)
|
|
{
|
|
struct tm tm;
|
|
int res=0, offset, sndoffset;
|
|
char sndfile[256], nextmsg[256];
|
|
|
|
ast_localtime(&time,&tm,timezone);
|
|
|
|
for (offset=0 ; format[offset] != '\0' ; offset++) {
|
|
ast_log(LOG_DEBUG, "Parsing %c (offset %d) in %s\n", format[offset], offset, format);
|
|
switch (format[offset]) {
|
|
/* NOTE: if you add more options here, please try to be consistent with strftime(3) */
|
|
case '\'':
|
|
/* Literal name of a sound file */
|
|
sndoffset=0;
|
|
for (sndoffset=0 ; (format[++offset] != '\'') && (sndoffset < 256) ; sndoffset++)
|
|
sndfile[sndoffset] = format[offset];
|
|
sndfile[sndoffset] = '\0';
|
|
res = wait_file(chan,ints,sndfile,lang);
|
|
break;
|
|
case 'A':
|
|
case 'a':
|
|
/* Sunday - Saturday */
|
|
snprintf(nextmsg,sizeof(nextmsg), "digits/day-%d", tm.tm_wday);
|
|
res = wait_file(chan,ints,nextmsg,lang);
|
|
break;
|
|
case 'B':
|
|
case 'b':
|
|
case 'h':
|
|
/* January - December */
|
|
snprintf(nextmsg,sizeof(nextmsg), "digits/mon-%d", tm.tm_mon);
|
|
res = wait_file(chan,ints,nextmsg,lang);
|
|
break;
|
|
case 'm':
|
|
/* First - Twelfth */
|
|
snprintf(nextmsg,sizeof(nextmsg), "digits/h-%d", tm.tm_mon +1);
|
|
res = wait_file(chan,ints,nextmsg,lang);
|
|
break;
|
|
case 'd':
|
|
case 'e':
|
|
/* First - Thirtyfirst */
|
|
res = ast_say_number(chan, tm.tm_mday, ints, lang, NULL);
|
|
break;
|
|
case 'Y':
|
|
/* Year */
|
|
if (tm.tm_year > 99) {
|
|
res = wait_file(chan,ints, "digits/2",lang);
|
|
if (!res) {
|
|
res = wait_file(chan,ints, "digits/thousand",lang);
|
|
}
|
|
if (tm.tm_year > 100) {
|
|
if (!res) {
|
|
/* This works until the end of 2020 */
|
|
snprintf(nextmsg,sizeof(nextmsg), "digits/%d", tm.tm_year - 100);
|
|
res = wait_file(chan,ints,nextmsg,lang);
|
|
}
|
|
}
|
|
} else {
|
|
if (tm.tm_year < 1) {
|
|
/* I'm not going to handle 1900 and prior */
|
|
/* We'll just be silent on the year, instead of bombing out. */
|
|
} else {
|
|
res = wait_file(chan,ints, "digits/19",lang);
|
|
if (!res) {
|
|
if (tm.tm_year <= 9) {
|
|
/* 1901 - 1909 */
|
|
res = wait_file(chan,ints, "digits/oh",lang);
|
|
if (!res) {
|
|
snprintf(nextmsg,sizeof(nextmsg), "digits/%d", tm.tm_year);
|
|
res = wait_file(chan,ints,nextmsg,lang);
|
|
}
|
|
} else if (tm.tm_year <= 20) {
|
|
/* 1910 - 1920 */
|
|
snprintf(nextmsg,sizeof(nextmsg), "digits/%d", tm.tm_year);
|
|
res = wait_file(chan,ints,nextmsg,lang);
|
|
} else {
|
|
/* 1921 - 1999 */
|
|
int ten, one;
|
|
ten = tm.tm_year / 10;
|
|
one = tm.tm_year % 10;
|
|
snprintf(nextmsg,sizeof(nextmsg), "digits/%d", ten * 10);
|
|
res = wait_file(chan,ints,nextmsg,lang);
|
|
if (!res) {
|
|
if (one != 0) {
|
|
snprintf(nextmsg,sizeof(nextmsg), "digits/%d", one);
|
|
res = wait_file(chan,ints,nextmsg,lang);
|
|
}
|
|
}
|
|
}
|
|
}
|
|
}
|
|
}
|
|
break;
|
|
case 'I':
|
|
case 'l':
|
|
/* 12-Hour */
|
|
if (tm.tm_hour == 0)
|
|
snprintf(nextmsg,sizeof(nextmsg), "digits/12");
|
|
else if (tm.tm_hour > 12)
|
|
snprintf(nextmsg,sizeof(nextmsg), "digits/%d", tm.tm_hour - 12);
|
|
else
|
|
snprintf(nextmsg,sizeof(nextmsg), "digits/%d", tm.tm_hour);
|
|
res = wait_file(chan,ints,nextmsg,lang);
|
|
break;
|
|
case 'H':
|
|
case 'k':
|
|
/* 24-Hour */
|
|
res = ast_say_number(chan, tm.tm_hour, ints, lang, (char *) NULL);
|
|
if (!res) {
|
|
res = wait_file(chan,ints, "digits/nl-uur",lang);
|
|
}
|
|
break;
|
|
case 'M':
|
|
/* Minute */
|
|
res = ast_say_number(chan, tm.tm_min, ints, lang, (char *) NULL);
|
|
break;
|
|
case 'P':
|
|
case 'p':
|
|
/* AM/PM */
|
|
if (tm.tm_hour > 11)
|
|
snprintf(nextmsg,sizeof(nextmsg), "digits/p-m");
|
|
else
|
|
snprintf(nextmsg,sizeof(nextmsg), "digits/a-m");
|
|
res = wait_file(chan,ints,nextmsg,lang);
|
|
break;
|
|
case 'Q':
|
|
/* Shorthand for "Today", "Yesterday", or ABdY */
|
|
{
|
|
struct timeval now;
|
|
struct tm tmnow;
|
|
time_t beg_today;
|
|
|
|
gettimeofday(&now,NULL);
|
|
ast_localtime(&now.tv_sec,&tmnow,timezone);
|
|
/* This might be slightly off, if we transcend a leap second, but never more off than 1 second */
|
|
/* In any case, it saves not having to do ast_mktime() */
|
|
beg_today = now.tv_sec - (tmnow.tm_hour * 3600) - (tmnow.tm_min * 60) - (tmnow.tm_sec);
|
|
if (beg_today < time) {
|
|
/* Today */
|
|
res = wait_file(chan,ints, "digits/today",lang);
|
|
} else if (beg_today - 86400 < time) {
|
|
/* Yesterday */
|
|
res = wait_file(chan,ints, "digits/yesterday",lang);
|
|
} else {
|
|
res = ast_say_date_with_format(chan, time, ints, lang, "ABdY", timezone);
|
|
}
|
|
}
|
|
break;
|
|
case 'q':
|
|
/* Shorthand for "" (today), "Yesterday", A (weekday), or ABdY */
|
|
{
|
|
struct timeval now;
|
|
struct tm tmnow;
|
|
time_t beg_today;
|
|
|
|
gettimeofday(&now,NULL);
|
|
ast_localtime(&now.tv_sec,&tmnow,timezone);
|
|
/* This might be slightly off, if we transcend a leap second, but never more off than 1 second */
|
|
/* In any case, it saves not having to do ast_mktime() */
|
|
beg_today = now.tv_sec - (tmnow.tm_hour * 3600) - (tmnow.tm_min * 60) - (tmnow.tm_sec);
|
|
if (beg_today < time) {
|
|
/* Today */
|
|
} else if ((beg_today - 86400) < time) {
|
|
/* Yesterday */
|
|
res = wait_file(chan,ints, "digits/yesterday",lang);
|
|
} else if (beg_today - 86400 * 6 < time) {
|
|
/* Within the last week */
|
|
res = ast_say_date_with_format(chan, time, ints, lang, "A", timezone);
|
|
} else {
|
|
res = ast_say_date_with_format(chan, time, ints, lang, "ABdY", timezone);
|
|
}
|
|
}
|
|
break;
|
|
case 'R':
|
|
res = ast_say_date_with_format(chan, time, ints, lang, "HM", timezone);
|
|
break;
|
|
case 'S':
|
|
/* Seconds */
|
|
if (tm.tm_sec == 0) {
|
|
snprintf(nextmsg,sizeof(nextmsg), "digits/%d", tm.tm_sec);
|
|
res = wait_file(chan,ints,nextmsg,lang);
|
|
} else if (tm.tm_sec < 10) {
|
|
res = wait_file(chan,ints, "digits/oh",lang);
|
|
if (!res) {
|
|
snprintf(nextmsg,sizeof(nextmsg), "digits/%d", tm.tm_sec);
|
|
res = wait_file(chan,ints,nextmsg,lang);
|
|
}
|
|
} else if ((tm.tm_sec < 21) || (tm.tm_sec % 10 == 0)) {
|
|
snprintf(nextmsg,sizeof(nextmsg), "digits/%d", tm.tm_sec);
|
|
res = wait_file(chan,ints,nextmsg,lang);
|
|
} else {
|
|
int ten, one;
|
|
ten = (tm.tm_sec / 10) * 10;
|
|
one = (tm.tm_sec % 10);
|
|
snprintf(nextmsg,sizeof(nextmsg), "digits/%d", ten);
|
|
res = wait_file(chan,ints,nextmsg,lang);
|
|
if (!res) {
|
|
/* Fifty, not fifty-zero */
|
|
if (one != 0) {
|
|
snprintf(nextmsg,sizeof(nextmsg), "digits/%d", one);
|
|
res = wait_file(chan,ints,nextmsg,lang);
|
|
}
|
|
}
|
|
}
|
|
break;
|
|
case 'T':
|
|
res = ast_say_date_with_format(chan, time, ints, lang, "HMS", timezone);
|
|
break;
|
|
case ' ':
|
|
case ' ':
|
|
/* Just ignore spaces and tabs */
|
|
break;
|
|
default:
|
|
/* Unknown character */
|
|
ast_log(LOG_WARNING, "Unknown character in datetime format %s: %c at pos %d\n", format, format[offset], offset);
|
|
}
|
|
/* Jump out on DTMF */
|
|
if (res) {
|
|
break;
|
|
}
|
|
}
|
|
return res;
|
|
}
|
|
|
|
/* Portuguese syntax */
|
|
int ast_say_date_with_format_pt(struct ast_channel *chan, time_t time, const char *ints, const char *lang, const char *format, const char *timezone)
|
|
{
|
|
struct tm tm;
|
|
int res=0, offset, sndoffset;
|
|
char sndfile[256], nextmsg[256];
|
|
|
|
ast_localtime(&time,&tm,timezone);
|
|
|
|
for (offset=0 ; format[offset] != '\0' ; offset++) {
|
|
ast_log(LOG_DEBUG, "Parsing %c (offset %d) in %s\n", format[offset], offset, format);
|
|
switch (format[offset]) {
|
|
/* NOTE: if you add more options here, please try to be consistent with strftime(3) */
|
|
case '\'':
|
|
/* Literal name of a sound file */
|
|
sndoffset=0;
|
|
for (sndoffset=0 ; (format[++offset] != '\'') && (sndoffset < 256) ; sndoffset++)
|
|
sndfile[sndoffset] = format[offset];
|
|
sndfile[sndoffset] = '\0';
|
|
snprintf(nextmsg,sizeof(nextmsg), "%s", sndfile);
|
|
res = wait_file(chan,ints,nextmsg,lang);
|
|
break;
|
|
case 'A':
|
|
case 'a':
|
|
/* Sunday - Saturday */
|
|
snprintf(nextmsg,sizeof(nextmsg), "digits/day-%d", tm.tm_wday);
|
|
res = wait_file(chan,ints,nextmsg,lang);
|
|
break;
|
|
case 'B':
|
|
case 'b':
|
|
case 'h':
|
|
/* January - December */
|
|
snprintf(nextmsg,sizeof(nextmsg), "digits/mon-%d", tm.tm_mon);
|
|
res = wait_file(chan,ints,nextmsg,lang);
|
|
break;
|
|
case 'm':
|
|
/* First - Twelfth */
|
|
snprintf(nextmsg,sizeof(nextmsg), "digits/h-%d", tm.tm_mon +1);
|
|
res = wait_file(chan,ints,nextmsg,lang);
|
|
break;
|
|
case 'd':
|
|
case 'e':
|
|
/* First - Thirtyfirst */
|
|
res = ast_say_number(chan, tm.tm_mday, ints, lang, (char *) NULL);
|
|
break;
|
|
case 'Y':
|
|
/* Year */
|
|
res = ast_say_number(chan, tm.tm_year + 1900, ints, lang, (char *) NULL);
|
|
break;
|
|
case 'I':
|
|
case 'l':
|
|
/* 12-Hour */
|
|
if (tm.tm_hour == 0) {
|
|
if (format[offset] == 'I')
|
|
res = wait_file(chan, ints, "digits/pt-ah", lang);
|
|
if (!res)
|
|
res = wait_file(chan, ints, "digits/pt-meianoite", lang);
|
|
}
|
|
else if (tm.tm_hour == 12) {
|
|
if (format[offset] == 'I')
|
|
res = wait_file(chan, ints, "digits/pt-ao", lang);
|
|
if (!res)
|
|
res = wait_file(chan, ints, "digits/pt-meiodia", lang);
|
|
}
|
|
else {
|
|
if (format[offset] == 'I') {
|
|
res = wait_file(chan, ints, "digits/pt-ah", lang);
|
|
if ((tm.tm_hour % 12) != 1)
|
|
if (!res)
|
|
res = wait_file(chan, ints, "digits/pt-sss", lang);
|
|
}
|
|
if (!res)
|
|
res = ast_say_number(chan, (tm.tm_hour % 12), ints, lang, "f");
|
|
}
|
|
break;
|
|
case 'H':
|
|
case 'k':
|
|
/* 24-Hour */
|
|
res = ast_say_number(chan, -tm.tm_hour, ints, lang, NULL);
|
|
if (!res) {
|
|
if (tm.tm_hour != 0) {
|
|
int remainder = tm.tm_hour;
|
|
if (tm.tm_hour > 20) {
|
|
res = wait_file(chan,ints, "digits/20",lang);
|
|
remainder -= 20;
|
|
}
|
|
if (!res) {
|
|
snprintf(nextmsg,sizeof(nextmsg), "digits/%d", remainder);
|
|
res = wait_file(chan,ints,nextmsg,lang);
|
|
}
|
|
}
|
|
}
|
|
break;
|
|
case 'M':
|
|
/* Minute */
|
|
if (tm.tm_min == 0) {
|
|
res = wait_file(chan, ints, "digits/pt-hora", lang);
|
|
if (tm.tm_hour != 1)
|
|
if (!res)
|
|
res = wait_file(chan, ints, "digits/pt-sss", lang); } else {
|
|
res = wait_file(chan,ints,"digits/pt-e",lang);
|
|
if (!res)
|
|
res = ast_say_number(chan, tm.tm_min, ints, lang, (char *) NULL);
|
|
}
|
|
break;
|
|
case 'P':
|
|
case 'p':
|
|
/* AM/PM */
|
|
if (tm.tm_hour > 12)
|
|
res = wait_file(chan, ints, "digits/p-m", lang);
|
|
else if (tm.tm_hour && tm.tm_hour < 12)
|
|
res = wait_file(chan, ints, "digits/a-m", lang);
|
|
break;
|
|
case 'Q':
|
|
/* Shorthand for "Today", "Yesterday", or ABdY */
|
|
{
|
|
struct timeval now;
|
|
struct tm tmnow;
|
|
time_t beg_today;
|
|
|
|
gettimeofday(&now,NULL);
|
|
ast_localtime(&now.tv_sec,&tmnow,timezone);
|
|
/* This might be slightly off, if we transcend a leap second, but never more off than 1 second */
|
|
/* In any case, it saves not having to do ast_mktime() */
|
|
beg_today = now.tv_sec - (tmnow.tm_hour * 3600) - (tmnow.tm_min * 60) - (tmnow.tm_sec);
|
|
if (beg_today < time) {
|
|
/* Today */
|
|
res = wait_file(chan,ints, "digits/today",lang);
|
|
} else if (beg_today - 86400 < time) {
|
|
/* Yesterday */
|
|
res = wait_file(chan,ints, "digits/yesterday",lang);
|
|
} else {
|
|
res = ast_say_date_with_format(chan, time, ints, lang, "Ad 'digits/pt-de' B 'digits/pt-de' Y", timezone);
|
|
}
|
|
}
|
|
break;
|
|
case 'q':
|
|
/* Shorthand for "" (today), "Yesterday", A (weekday), or ABdY */
|
|
{
|
|
struct timeval now;
|
|
struct tm tmnow;
|
|
time_t beg_today;
|
|
|
|
gettimeofday(&now,NULL);
|
|
ast_localtime(&now.tv_sec,&tmnow,timezone);
|
|
/* This might be slightly off, if we transcend a leap second, but never more off than 1 second */
|
|
/* In any case, it saves not having to do ast_mktime() */
|
|
beg_today = now.tv_sec - (tmnow.tm_hour * 3600) - (tmnow.tm_min * 60) - (tmnow.tm_sec);
|
|
if (beg_today < time) {
|
|
/* Today */
|
|
} else if ((beg_today - 86400) < time) {
|
|
/* Yesterday */
|
|
res = wait_file(chan,ints, "digits/yesterday",lang);
|
|
} else if (beg_today - 86400 * 6 < time) {
|
|
/* Within the last week */
|
|
res = ast_say_date_with_format(chan, time, ints, lang, "A", timezone);
|
|
} else {
|
|
res = ast_say_date_with_format(chan, time, ints, lang, "Ad 'digits/pt-de' B 'digits/pt-de' Y", timezone);
|
|
}
|
|
}
|
|
break;
|
|
case 'R':
|
|
res = ast_say_date_with_format(chan, time, ints, lang, "H 'digits/pt-e' M", timezone);
|
|
break;
|
|
case 'S':
|
|
/* Seconds */
|
|
if (tm.tm_sec == 0) {
|
|
snprintf(nextmsg,sizeof(nextmsg), "digits/%d", tm.tm_sec);
|
|
res = wait_file(chan,ints,nextmsg,lang);
|
|
} else if (tm.tm_sec < 10) {
|
|
res = wait_file(chan,ints, "digits/oh",lang);
|
|
if (!res) {
|
|
snprintf(nextmsg,sizeof(nextmsg), "digits/%d", tm.tm_sec);
|
|
res = wait_file(chan,ints,nextmsg,lang);
|
|
}
|
|
} else if ((tm.tm_sec < 21) || (tm.tm_sec % 10 == 0)) {
|
|
snprintf(nextmsg,sizeof(nextmsg), "digits/%d", tm.tm_sec);
|
|
res = wait_file(chan,ints,nextmsg,lang);
|
|
} else {
|
|
int ten, one;
|
|
ten = (tm.tm_sec / 10) * 10;
|
|
one = (tm.tm_sec % 10);
|
|
snprintf(nextmsg,sizeof(nextmsg), "digits/%d", ten);
|
|
res = wait_file(chan,ints,nextmsg,lang);
|
|
if (!res) {
|
|
/* Fifty, not fifty-zero */
|
|
if (one != 0) {
|
|
snprintf(nextmsg,sizeof(nextmsg), "digits/%d", one);
|
|
res = wait_file(chan,ints,nextmsg,lang);
|
|
}
|
|
}
|
|
}
|
|
break;
|
|
case 'T':
|
|
res = ast_say_date_with_format(chan, time, ints, lang, "HMS", timezone);
|
|
break;
|
|
case ' ':
|
|
case ' ':
|
|
/* Just ignore spaces and tabs */
|
|
break;
|
|
default:
|
|
/* Unknown character */
|
|
ast_log(LOG_WARNING, "Unknown character in datetime format %s: %c at pos %d\n", format, format[offset], offset);
|
|
}
|
|
/* Jump out on DTMF */
|
|
if (res) {
|
|
break;
|
|
}
|
|
}
|
|
return res;
|
|
}
|
|
|
|
/* Taiwanese syntax */
|
|
int ast_say_date_with_format_tw(struct ast_channel *chan, time_t time, const char *ints, const char *lang, const char *format, const char *timezone)
|
|
{
|
|
struct tm tm;
|
|
int res=0, offset, sndoffset;
|
|
char sndfile[256], nextmsg[256];
|
|
|
|
ast_localtime(&time,&tm,timezone);
|
|
|
|
for (offset=0 ; format[offset] != '\0' ; offset++) {
|
|
ast_log(LOG_DEBUG, "Parsing %c (offset %d) in %s\n", format[offset], offset, format);
|
|
switch (format[offset]) {
|
|
/* NOTE: if you add more options here, please try to be consistent with strftime(3) */
|
|
case '\'':
|
|
/* Literal name of a sound file */
|
|
sndoffset=0;
|
|
for (sndoffset=0 ; (format[++offset] != '\'') && (sndoffset < 256) ; sndoffset++)
|
|
sndfile[sndoffset] = format[offset];
|
|
sndfile[sndoffset] = '\0';
|
|
res = wait_file(chan,ints,sndfile,lang);
|
|
break;
|
|
case 'A':
|
|
case 'a':
|
|
/* Sunday - Saturday */
|
|
snprintf(nextmsg,sizeof(nextmsg), "digits/day-%d", tm.tm_wday);
|
|
res = wait_file(chan,ints,nextmsg,lang);
|
|
break;
|
|
case 'B':
|
|
case 'b':
|
|
case 'h':
|
|
/* January - December */
|
|
snprintf(nextmsg,sizeof(nextmsg), "digits/mon-%d", tm.tm_mon);
|
|
res = wait_file(chan,ints,nextmsg,lang);
|
|
break;
|
|
case 'm':
|
|
/* First - Twelfth */
|
|
snprintf(nextmsg,sizeof(nextmsg), "digits/h-%d", tm.tm_mon +1);
|
|
res = wait_file(chan,ints,nextmsg,lang);
|
|
break;
|
|
case 'd':
|
|
case 'e':
|
|
/* First - Thirtyfirst */
|
|
if (!(tm.tm_mday % 10) || (tm.tm_mday < 10)) {
|
|
snprintf(nextmsg,sizeof(nextmsg), "digits/h-%d", tm.tm_mday);
|
|
res = wait_file(chan,ints,nextmsg,lang);
|
|
} else {
|
|
snprintf(nextmsg,sizeof(nextmsg), "digits/h-%dh", tm.tm_mday - (tm.tm_mday % 10));
|
|
res = wait_file(chan,ints,nextmsg,lang);
|
|
if(!res) {
|
|
snprintf(nextmsg,sizeof(nextmsg), "digits/h-%d", tm.tm_mday % 10);
|
|
res = wait_file(chan,ints,nextmsg,lang);
|
|
}
|
|
}
|
|
break;
|
|
case 'Y':
|
|
/* Year */
|
|
if (tm.tm_year > 99) {
|
|
res = wait_file(chan,ints, "digits/2",lang);
|
|
if (!res) {
|
|
res = wait_file(chan,ints, "digits/thousand",lang);
|
|
}
|
|
if (tm.tm_year > 100) {
|
|
if (!res) {
|
|
snprintf(nextmsg,sizeof(nextmsg), "digits/%d", (tm.tm_year - 100) / 10);
|
|
res = wait_file(chan,ints,nextmsg,lang);
|
|
if (!res) {
|
|
snprintf(nextmsg,sizeof(nextmsg), "digits/%d", (tm.tm_year - 100) % 10);
|
|
res = wait_file(chan,ints,nextmsg,lang);
|
|
}
|
|
}
|
|
}
|
|
if (!res) {
|
|
res = wait_file(chan,ints, "digits/year",lang);
|
|
}
|
|
} else {
|
|
if (tm.tm_year < 1) {
|
|
/* I'm not going to handle 1900 and prior */
|
|
/* We'll just be silent on the year, instead of bombing out. */
|
|
} else {
|
|
res = wait_file(chan,ints, "digits/1",lang);
|
|
if (!res) {
|
|
res = wait_file(chan,ints, "digits/9",lang);
|
|
}
|
|
if (!res) {
|
|
if (tm.tm_year <= 9) {
|
|
/* 1901 - 1909 */
|
|
res = wait_file(chan,ints, "digits/0",lang);
|
|
if (!res) {
|
|
snprintf(nextmsg,sizeof(nextmsg), "digits/%d", tm.tm_year);
|
|
res = wait_file(chan,ints,nextmsg,lang);
|
|
}
|
|
} else {
|
|
/* 1910 - 1999 */
|
|
snprintf(nextmsg,sizeof(nextmsg), "digits/%d", tm.tm_year / 10);
|
|
res = wait_file(chan,ints,nextmsg,lang);
|
|
if (!res) {
|
|
snprintf(nextmsg,sizeof(nextmsg), "digits/%d", tm.tm_year % 10);
|
|
res = wait_file(chan,ints,nextmsg,lang);
|
|
}
|
|
}
|
|
}
|
|
}
|
|
if (!res) {
|
|
res = wait_file(chan,ints, "digits/year",lang);
|
|
}
|
|
}
|
|
break;
|
|
case 'I':
|
|
case 'l':
|
|
/* 12-Hour */
|
|
if (tm.tm_hour == 0)
|
|
snprintf(nextmsg,sizeof(nextmsg), "digits/12");
|
|
else if (tm.tm_hour > 12)
|
|
snprintf(nextmsg,sizeof(nextmsg), "digits/%d", tm.tm_hour - 12);
|
|
else
|
|
snprintf(nextmsg,sizeof(nextmsg), "digits/%d", tm.tm_hour);
|
|
res = wait_file(chan,ints,nextmsg,lang);
|
|
if (!res) {
|
|
res = wait_file(chan,ints, "digits/oclock",lang);
|
|
}
|
|
break;
|
|
case 'H':
|
|
case 'k':
|
|
/* 24-Hour */
|
|
if (!(tm.tm_hour % 10) || tm.tm_hour < 10) {
|
|
if (tm.tm_hour < 10) {
|
|
res = wait_file(chan, ints, "digits/0", lang);
|
|
}
|
|
snprintf(nextmsg,sizeof(nextmsg), "digits/%d", tm.tm_hour);
|
|
res = wait_file(chan,ints,nextmsg,lang);
|
|
} else {
|
|
snprintf(nextmsg,sizeof(nextmsg), "digits/%d", tm.tm_hour - (tm.tm_hour % 10));
|
|
res = wait_file(chan,ints,nextmsg,lang);
|
|
if (!res) {
|
|
snprintf(nextmsg,sizeof(nextmsg), "digits/%d", tm.tm_hour % 10);
|
|
res = wait_file(chan,ints,nextmsg,lang);
|
|
}
|
|
}
|
|
if (!res) {
|
|
res = wait_file(chan,ints, "digits/oclock",lang);
|
|
}
|
|
break;
|
|
case 'M':
|
|
/* Minute */
|
|
if (!(tm.tm_min % 10) || tm.tm_min < 10) {
|
|
if (tm.tm_min < 10) {
|
|
res = wait_file(chan, ints, "digits/0", lang);
|
|
}
|
|
snprintf(nextmsg,sizeof(nextmsg), "digits/%d", tm.tm_min);
|
|
res = wait_file(chan,ints,nextmsg,lang);
|
|
} else {
|
|
snprintf(nextmsg,sizeof(nextmsg), "digits/%d", tm.tm_min - (tm.tm_min % 10));
|
|
res = wait_file(chan,ints,nextmsg,lang);
|
|
if (!res) {
|
|
snprintf(nextmsg,sizeof(nextmsg), "digits/%d", tm.tm_min % 10);
|
|
res = wait_file(chan,ints,nextmsg,lang);
|
|
}
|
|
}
|
|
if (!res) {
|
|
res = wait_file(chan,ints, "digits/minute",lang);
|
|
}
|
|
break;
|
|
case 'P':
|
|
case 'p':
|
|
/* AM/PM */
|
|
if (tm.tm_hour > 11)
|
|
snprintf(nextmsg,sizeof(nextmsg), "digits/p-m");
|
|
else
|
|
snprintf(nextmsg,sizeof(nextmsg), "digits/a-m");
|
|
res = wait_file(chan,ints,nextmsg,lang);
|
|
break;
|
|
case 'Q':
|
|
/* Shorthand for "Today", "Yesterday", or ABdY */
|
|
{
|
|
struct timeval now;
|
|
struct tm tmnow;
|
|
time_t beg_today;
|
|
|
|
gettimeofday(&now,NULL);
|
|
ast_localtime(&now.tv_sec,&tmnow,timezone);
|
|
/* This might be slightly off, if we transcend a leap second, but never more off than 1 second */
|
|
/* In any case, it saves not having to do ast_mktime() */
|
|
beg_today = now.tv_sec - (tmnow.tm_hour * 3600) - (tmnow.tm_min * 60) - (tmnow.tm_sec);
|
|
if (beg_today < time) {
|
|
/* Today */
|
|
res = wait_file(chan,ints, "digits/today",lang);
|
|
} else if (beg_today - 86400 < time) {
|
|
/* Yesterday */
|
|
res = wait_file(chan,ints, "digits/yesterday",lang);
|
|
} else {
|
|
res = ast_say_date_with_format(chan, time, ints, lang, "YBdA", timezone);
|
|
}
|
|
}
|
|
break;
|
|
case 'q':
|
|
/* Shorthand for "" (today), "Yesterday", A (weekday), or ABdY */
|
|
{
|
|
struct timeval now;
|
|
struct tm tmnow;
|
|
time_t beg_today;
|
|
|
|
gettimeofday(&now,NULL);
|
|
ast_localtime(&now.tv_sec,&tmnow,timezone);
|
|
/* This might be slightly off, if we transcend a leap second, but never more off than 1 second */
|
|
/* In any case, it saves not having to do ast_mktime() */
|
|
beg_today = now.tv_sec - (tmnow.tm_hour * 3600) - (tmnow.tm_min * 60) - (tmnow.tm_sec);
|
|
if (beg_today < time) {
|
|
/* Today */
|
|
} else if ((beg_today - 86400) < time) {
|
|
/* Yesterday */
|
|
res = wait_file(chan,ints, "digits/yesterday",lang);
|
|
} else if (beg_today - 86400 * 6 < time) {
|
|
/* Within the last week */
|
|
res = ast_say_date_with_format(chan, time, ints, lang, "A", timezone);
|
|
} else {
|
|
res = ast_say_date_with_format(chan, time, ints, lang, "YBdA", timezone);
|
|
}
|
|
}
|
|
break;
|
|
case 'R':
|
|
res = ast_say_date_with_format(chan, time, ints, lang, "HM", timezone);
|
|
break;
|
|
case 'S':
|
|
/* Seconds */
|
|
if (!(tm.tm_sec % 10) || tm.tm_sec < 10) {
|
|
if (tm.tm_sec < 10) {
|
|
res = wait_file(chan, ints, "digits/0", lang);
|
|
}
|
|
snprintf(nextmsg,sizeof(nextmsg), "digits/%d", tm.tm_sec);
|
|
res = wait_file(chan,ints,nextmsg,lang);
|
|
} else {
|
|
snprintf(nextmsg,sizeof(nextmsg), "digits/%d", tm.tm_sec - (tm.tm_sec % 10));
|
|
res = wait_file(chan,ints,nextmsg,lang);
|
|
if (!res) {
|
|
snprintf(nextmsg,sizeof(nextmsg), "digits/%d", tm.tm_sec % 10);
|
|
res = wait_file(chan,ints,nextmsg,lang);
|
|
}
|
|
}
|
|
if (!res) {
|
|
res = wait_file(chan,ints, "digits/second",lang);
|
|
}
|
|
break;
|
|
case 'T':
|
|
res = ast_say_date_with_format(chan, time, ints, lang, "HMS", timezone);
|
|
break;
|
|
case ' ':
|
|
case ' ':
|
|
/* Just ignore spaces and tabs */
|
|
break;
|
|
default:
|
|
/* Unknown character */
|
|
ast_log(LOG_WARNING, "Unknown character in datetime format %s: %c at pos %d\n", format, format[offset], offset);
|
|
}
|
|
/* Jump out on DTMF */
|
|
if (res) {
|
|
break;
|
|
}
|
|
}
|
|
return res;
|
|
}
|
|
|
|
int ast_say_time(struct ast_channel *chan, time_t t, const char *ints, const char *lang)
|
|
{
|
|
if (!strcasecmp(lang, "en") ) { /* English syntax */
|
|
return(ast_say_time_en(chan, t, ints, lang));
|
|
} else if (!strcasecmp(lang, "de") ) { /* German syntax */
|
|
return(ast_say_time_de(chan, t, ints, lang));
|
|
} else if (!strcasecmp(lang, "fr") ) { /* French syntax */
|
|
return(ast_say_time_fr(chan, t, ints, lang));
|
|
} else if (!strcasecmp(lang, "nl") ) { /* Dutch syntax */
|
|
return(ast_say_time_nl(chan, t, ints, lang));
|
|
} else if (!strcasecmp(lang, "pt") ) { /* Portuguese syntax */
|
|
return(ast_say_time_pt(chan, t, ints, lang));
|
|
} else if (!strcasecmp(lang, "tw") ) { /* Taiwanese syntax */
|
|
return(ast_say_time_tw(chan, t, ints, lang));
|
|
} else if (!strcasecmp(lang, "gr") ) { /* Greek syntax */
|
|
return(ast_say_time_gr(chan, t, ints, lang));
|
|
}
|
|
|
|
/* Default to English */
|
|
return(ast_say_time_en(chan, t, ints, lang));
|
|
}
|
|
|
|
/* English syntax */
|
|
int ast_say_time_en(struct ast_channel *chan, time_t t, const char *ints, const char *lang)
|
|
{
|
|
struct tm tm;
|
|
int res = 0;
|
|
int hour, pm=0;
|
|
localtime_r(&t,&tm);
|
|
hour = tm.tm_hour;
|
|
if (!hour)
|
|
hour = 12;
|
|
else if (hour == 12)
|
|
pm = 1;
|
|
else if (hour > 12) {
|
|
hour -= 12;
|
|
pm = 1;
|
|
}
|
|
if (!res)
|
|
res = ast_say_number(chan, hour, ints, lang, (char *) NULL);
|
|
|
|
if (tm.tm_min > 9) {
|
|
if (!res)
|
|
res = ast_say_number(chan, tm.tm_min, ints, lang, (char *) NULL);
|
|
} else if (tm.tm_min) {
|
|
if (!res)
|
|
res = ast_streamfile(chan, "digits/oh", lang);
|
|
if (!res)
|
|
res = ast_waitstream(chan, ints);
|
|
if (!res)
|
|
res = ast_say_number(chan, tm.tm_min, ints, lang, (char *) NULL);
|
|
} else {
|
|
if (!res)
|
|
res = ast_streamfile(chan, "digits/oclock", lang);
|
|
if (!res)
|
|
res = ast_waitstream(chan, ints);
|
|
}
|
|
if (pm) {
|
|
if (!res)
|
|
res = ast_streamfile(chan, "digits/p-m", lang);
|
|
} else {
|
|
if (!res)
|
|
res = ast_streamfile(chan, "digits/a-m", lang);
|
|
}
|
|
if (!res)
|
|
res = ast_waitstream(chan, ints);
|
|
return res;
|
|
}
|
|
|
|
/* German syntax */
|
|
int ast_say_time_de(struct ast_channel *chan, time_t t, const char *ints, const char *lang)
|
|
{
|
|
struct tm tm;
|
|
int res = 0;
|
|
localtime_r(&t,&tm);
|
|
if (!res)
|
|
res = ast_say_number(chan, tm.tm_hour, ints, lang, "n");
|
|
if (!res)
|
|
res = ast_streamfile(chan, "digits/oclock", lang);
|
|
if (!res)
|
|
res = ast_waitstream(chan, ints);
|
|
if (!res)
|
|
if (tm.tm_min > 0)
|
|
res = ast_say_number(chan, tm.tm_min, ints, lang, "f");
|
|
return res;
|
|
}
|
|
|
|
/* French syntax */
|
|
int ast_say_time_fr(struct ast_channel *chan, time_t t, const char *ints, const char *lang)
|
|
{
|
|
struct tm tm;
|
|
int res = 0;
|
|
localtime_r(&t,&tm);
|
|
|
|
res = ast_say_number(chan, tm.tm_hour, ints, lang, "f");
|
|
if (!res)
|
|
res = ast_streamfile(chan, "digits/oclock", lang);
|
|
if (tm.tm_min) {
|
|
if (!res)
|
|
res = ast_say_number(chan, tm.tm_min, ints, lang, (char *) NULL);
|
|
}
|
|
return res;
|
|
}
|
|
|
|
/* Dutch syntax */
|
|
int ast_say_time_nl(struct ast_channel *chan, time_t t, const char *ints, const char *lang)
|
|
{
|
|
struct tm tm;
|
|
int res = 0;
|
|
localtime_r(&t,&tm);
|
|
if (!res)
|
|
res = ast_say_number(chan, tm.tm_hour, ints, lang, (char *) NULL);
|
|
if (!res)
|
|
res = ast_streamfile(chan, "digits/nl-uur", lang);
|
|
if (!res)
|
|
res = ast_waitstream(chan, ints);
|
|
if (!res)
|
|
if (tm.tm_min > 0)
|
|
res = ast_say_number(chan, tm.tm_min, ints, lang, NULL);
|
|
return res;
|
|
}
|
|
|
|
/* Portuguese syntax */
|
|
int ast_say_time_pt(struct ast_channel *chan, time_t t, const char *ints, const char *lang)
|
|
{
|
|
struct tm tm;
|
|
int res = 0;
|
|
int hour;
|
|
localtime_r(&t,&tm);
|
|
hour = tm.tm_hour;
|
|
if (!res)
|
|
res = ast_say_number(chan, hour, ints, lang, "f");
|
|
if (tm.tm_min) {
|
|
if (!res)
|
|
res = wait_file(chan, ints, "digits/pt-e", lang);
|
|
if (!res)
|
|
res = ast_say_number(chan, tm.tm_min, ints, lang, (char *) NULL);
|
|
} else {
|
|
if (!res)
|
|
res = wait_file(chan, ints, "digits/pt-hora", lang);
|
|
if (tm.tm_hour != 1)
|
|
if (!res)
|
|
res = wait_file(chan, ints, "digits/pt-sss", lang);
|
|
}
|
|
if (!res)
|
|
res = ast_say_number(chan, hour, ints, lang, (char *) NULL);
|
|
return res;
|
|
}
|
|
|
|
/* Taiwanese syntax */
|
|
int ast_say_time_tw(struct ast_channel *chan, time_t t, const char *ints, const char *lang)
|
|
{
|
|
struct tm tm;
|
|
int res = 0;
|
|
int hour, pm=0;
|
|
localtime_r(&t,&tm);
|
|
hour = tm.tm_hour;
|
|
if (!hour)
|
|
hour = 12;
|
|
else if (hour == 12)
|
|
pm = 1;
|
|
else if (hour > 12) {
|
|
hour -= 12;
|
|
pm = 1;
|
|
}
|
|
if (pm) {
|
|
if (!res)
|
|
res = ast_streamfile(chan, "digits/p-m", lang);
|
|
} else {
|
|
if (!res)
|
|
res = ast_streamfile(chan, "digits/a-m", lang);
|
|
}
|
|
if (!res)
|
|
res = ast_waitstream(chan, ints);
|
|
if (!res)
|
|
res = ast_say_number(chan, hour, ints, lang, (char *) NULL);
|
|
if (!res)
|
|
res = ast_streamfile(chan, "digits/oclock", lang);
|
|
if (!res)
|
|
res = ast_waitstream(chan, ints);
|
|
if (!res)
|
|
res = ast_say_number(chan, tm.tm_min, ints, lang, (char *) NULL);
|
|
if (!res)
|
|
res = ast_streamfile(chan, "digits/minute", lang);
|
|
if (!res)
|
|
res = ast_waitstream(chan, ints);
|
|
return res;
|
|
}
|
|
|
|
int ast_say_datetime(struct ast_channel *chan, time_t t, const char *ints, const char *lang)
|
|
{
|
|
if (!strcasecmp(lang, "en") ) { /* English syntax */
|
|
return(ast_say_datetime_en(chan, t, ints, lang));
|
|
} else if (!strcasecmp(lang, "de") ) { /* German syntax */
|
|
return(ast_say_datetime_de(chan, t, ints, lang));
|
|
} else if (!strcasecmp(lang, "fr") ) { /* French syntax */
|
|
return(ast_say_datetime_fr(chan, t, ints, lang));
|
|
} else if (!strcasecmp(lang, "nl") ) { /* Dutch syntax */
|
|
return(ast_say_datetime_nl(chan, t, ints, lang));
|
|
} else if (!strcasecmp(lang, "pt") ) { /* Portuguese syntax */
|
|
return(ast_say_datetime_pt(chan, t, ints, lang));
|
|
} else if (!strcasecmp(lang, "tw") ) { /* Taiwanese syntax */
|
|
return(ast_say_datetime_tw(chan, t, ints, lang));
|
|
} else if (!strcasecmp(lang, "gr") ) { /* Greek syntax */
|
|
return(ast_say_datetime_gr(chan, t, ints, lang));
|
|
}
|
|
|
|
/* Default to English */
|
|
return(ast_say_datetime_en(chan, t, ints, lang));
|
|
}
|
|
|
|
/* English syntax */
|
|
int ast_say_datetime_en(struct ast_channel *chan, time_t t, const char *ints, const char *lang)
|
|
{
|
|
struct tm tm;
|
|
char fn[256];
|
|
int res = 0;
|
|
int hour, pm=0;
|
|
localtime_r(&t,&tm);
|
|
if (!res) {
|
|
snprintf(fn, sizeof(fn), "digits/day-%d", tm.tm_wday);
|
|
res = ast_streamfile(chan, fn, lang);
|
|
if (!res)
|
|
res = ast_waitstream(chan, ints);
|
|
}
|
|
if (!res) {
|
|
snprintf(fn, sizeof(fn), "digits/mon-%d", tm.tm_mon);
|
|
res = ast_streamfile(chan, fn, lang);
|
|
if (!res)
|
|
res = ast_waitstream(chan, ints);
|
|
}
|
|
if (!res)
|
|
res = ast_say_number(chan, tm.tm_mday, ints, lang, (char *) NULL);
|
|
|
|
hour = tm.tm_hour;
|
|
if (!hour)
|
|
hour = 12;
|
|
else if (hour == 12)
|
|
pm = 1;
|
|
else if (hour > 12) {
|
|
hour -= 12;
|
|
pm = 1;
|
|
}
|
|
if (!res)
|
|
res = ast_say_number(chan, hour, ints, lang, (char *) NULL);
|
|
|
|
if (tm.tm_min > 9) {
|
|
if (!res)
|
|
res = ast_say_number(chan, tm.tm_min, ints, lang, (char *) NULL);
|
|
} else if (tm.tm_min) {
|
|
if (!res)
|
|
res = ast_streamfile(chan, "digits/oh", lang);
|
|
if (!res)
|
|
res = ast_waitstream(chan, ints);
|
|
if (!res)
|
|
res = ast_say_number(chan, tm.tm_min, ints, lang, (char *) NULL);
|
|
} else {
|
|
if (!res)
|
|
res = ast_streamfile(chan, "digits/oclock", lang);
|
|
if (!res)
|
|
res = ast_waitstream(chan, ints);
|
|
}
|
|
if (pm) {
|
|
if (!res)
|
|
res = ast_streamfile(chan, "digits/p-m", lang);
|
|
} else {
|
|
if (!res)
|
|
res = ast_streamfile(chan, "digits/a-m", lang);
|
|
}
|
|
if (!res)
|
|
res = ast_waitstream(chan, ints);
|
|
if (!res)
|
|
res = ast_say_number(chan, tm.tm_year + 1900, ints, lang, (char *) NULL);
|
|
return res;
|
|
}
|
|
|
|
/* German syntax */
|
|
int ast_say_datetime_de(struct ast_channel *chan, time_t t, const char *ints, const char *lang)
|
|
{
|
|
struct tm tm;
|
|
int res = 0;
|
|
localtime_r(&t,&tm);
|
|
res = ast_say_date(chan, t, ints, lang);
|
|
if (!res)
|
|
ast_say_time(chan, t, ints, lang);
|
|
return res;
|
|
|
|
}
|
|
|
|
/* French syntax */
|
|
int ast_say_datetime_fr(struct ast_channel *chan, time_t t, const char *ints, const char *lang)
|
|
{
|
|
struct tm tm;
|
|
char fn[256];
|
|
int res = 0;
|
|
localtime_r(&t,&tm);
|
|
|
|
if (!res)
|
|
res = ast_say_number(chan, tm.tm_mday, ints, lang, (char *) NULL);
|
|
|
|
if (!res) {
|
|
snprintf(fn, sizeof(fn), "digits/day-%d", tm.tm_wday);
|
|
res = ast_streamfile(chan, fn, lang);
|
|
if (!res)
|
|
res = ast_waitstream(chan, ints);
|
|
}
|
|
if (!res) {
|
|
snprintf(fn, sizeof(fn), "digits/mon-%d", tm.tm_mon);
|
|
res = ast_streamfile(chan, fn, lang);
|
|
if (!res)
|
|
res = ast_waitstream(chan, ints);
|
|
}
|
|
|
|
if (!res)
|
|
res = ast_say_number(chan, tm.tm_hour, ints, lang, "f");
|
|
if (!res)
|
|
res = ast_streamfile(chan, "digits/oclock", lang);
|
|
if (tm.tm_min > 0) {
|
|
if (!res)
|
|
res = ast_say_number(chan, tm.tm_min, ints, lang, (char *) NULL);
|
|
}
|
|
if (!res)
|
|
res = ast_waitstream(chan, ints);
|
|
if (!res)
|
|
res = ast_say_number(chan, tm.tm_year + 1900, ints, lang, (char *) NULL);
|
|
return res;
|
|
}
|
|
|
|
/* Dutch syntax */
|
|
int ast_say_datetime_nl(struct ast_channel *chan, time_t t, const char *ints, const char *lang)
|
|
{
|
|
struct tm tm;
|
|
int res = 0;
|
|
localtime_r(&t,&tm);
|
|
res = ast_say_date(chan, t, ints, lang);
|
|
if (!res) {
|
|
res = ast_streamfile(chan, "digits/nl-om", lang);
|
|
if (!res)
|
|
res = ast_waitstream(chan, ints);
|
|
}
|
|
if (!res)
|
|
ast_say_time(chan, t, ints, lang);
|
|
return res;
|
|
}
|
|
|
|
/* Portuguese syntax */
|
|
int ast_say_datetime_pt(struct ast_channel *chan, time_t t, const char *ints, const char *lang)
|
|
{
|
|
struct tm tm;
|
|
char fn[256];
|
|
int res = 0;
|
|
int hour, pm=0;
|
|
localtime_r(&t,&tm);
|
|
if (!res) {
|
|
snprintf(fn, sizeof(fn), "digits/day-%d", tm.tm_wday);
|
|
res = ast_streamfile(chan, fn, lang);
|
|
if (!res)
|
|
res = ast_waitstream(chan, ints);
|
|
}
|
|
if (!res) {
|
|
snprintf(fn, sizeof(fn), "digits/mon-%d", tm.tm_mon);
|
|
res = ast_streamfile(chan, fn, lang);
|
|
if (!res)
|
|
res = ast_waitstream(chan, ints);
|
|
}
|
|
if (!res)
|
|
res = ast_say_number(chan, tm.tm_mday, ints, lang, (char *) NULL);
|
|
|
|
hour = tm.tm_hour;
|
|
if (!hour)
|
|
hour = 12;
|
|
else if (hour == 12)
|
|
pm = 1;
|
|
else if (hour > 12) {
|
|
hour -= 12;
|
|
pm = 1;
|
|
}
|
|
if (!res)
|
|
res = ast_say_number(chan, hour, ints, lang, (char *) NULL);
|
|
|
|
if (tm.tm_min > 9) {
|
|
if (!res)
|
|
res = ast_say_number(chan, tm.tm_min, ints, lang, (char *) NULL);
|
|
} else if (tm.tm_min) {
|
|
if (!res)
|
|
res = ast_streamfile(chan, "digits/oh", lang);
|
|
if (!res)
|
|
res = ast_waitstream(chan, ints);
|
|
if (!res)
|
|
res = ast_say_number(chan, tm.tm_min, ints, lang, (char *) NULL);
|
|
} else {
|
|
if (!res)
|
|
res = ast_streamfile(chan, "digits/oclock", lang);
|
|
if (!res)
|
|
res = ast_waitstream(chan, ints);
|
|
}
|
|
if (pm) {
|
|
if (!res)
|
|
res = ast_streamfile(chan, "digits/p-m", lang);
|
|
} else {
|
|
if (!res)
|
|
res = ast_streamfile(chan, "digits/a-m", lang);
|
|
}
|
|
if (!res)
|
|
res = ast_waitstream(chan, ints);
|
|
if (!res)
|
|
res = ast_say_number(chan, tm.tm_year + 1900, ints, lang, (char *) NULL);
|
|
return res;
|
|
}
|
|
|
|
/* Taiwanese syntax */
|
|
int ast_say_datetime_tw(struct ast_channel *chan, time_t t, const char *ints, const char *lang)
|
|
{
|
|
struct tm tm;
|
|
char fn[256];
|
|
int res = 0;
|
|
int hour, pm=0;
|
|
localtime_r(&t,&tm);
|
|
if (!res)
|
|
res = ast_say_number(chan, tm.tm_year + 1900, ints, lang, (char *) NULL);
|
|
if (!res) {
|
|
snprintf(fn, sizeof(fn), "digits/mon-%d", tm.tm_mon);
|
|
res = ast_streamfile(chan, fn, lang);
|
|
if (!res)
|
|
res = ast_waitstream(chan, ints);
|
|
}
|
|
if (!res)
|
|
res = ast_say_number(chan, tm.tm_mday, ints, lang, (char *) NULL);
|
|
if (!res) {
|
|
snprintf(fn, sizeof(fn), "digits/day-%d", tm.tm_wday);
|
|
res = ast_streamfile(chan, fn, lang);
|
|
if (!res)
|
|
res = ast_waitstream(chan, ints);
|
|
}
|
|
|
|
hour = tm.tm_hour;
|
|
if (!hour)
|
|
hour = 12;
|
|
else if (hour == 12)
|
|
pm = 1;
|
|
else if (hour > 12) {
|
|
hour -= 12;
|
|
pm = 1;
|
|
}
|
|
if (pm) {
|
|
if (!res)
|
|
res = ast_streamfile(chan, "digits/p-m", lang);
|
|
} else {
|
|
if (!res)
|
|
res = ast_streamfile(chan, "digits/a-m", lang);
|
|
}
|
|
if (!res)
|
|
res = ast_waitstream(chan, ints);
|
|
if (!res)
|
|
res = ast_say_number(chan, hour, ints, lang, (char *) NULL);
|
|
if (!res)
|
|
res = ast_streamfile(chan, "digits/oclock", lang);
|
|
if (!res)
|
|
res = ast_waitstream(chan, ints);
|
|
if (!res)
|
|
res = ast_say_number(chan, tm.tm_min, ints, lang, (char *) NULL);
|
|
if (!res)
|
|
res = ast_streamfile(chan, "digits/minute", lang);
|
|
if (!res)
|
|
res = ast_waitstream(chan, ints);
|
|
return res;
|
|
}
|
|
|
|
int ast_say_datetime_from_now(struct ast_channel *chan, time_t t, const char *ints, const char *lang)
|
|
{
|
|
if (!strcasecmp(lang, "en") ) { /* English syntax */
|
|
return(ast_say_datetime_from_now_en(chan, t, ints, lang));
|
|
} else if (!strcasecmp(lang, "fr") ) { /* French syntax */
|
|
return(ast_say_datetime_from_now_fr(chan, t, ints, lang));
|
|
} else if (!strcasecmp(lang, "pt") ) { /* Portuguese syntax */
|
|
return(ast_say_datetime_from_now_pt(chan, t, ints, lang));
|
|
}
|
|
|
|
/* Default to English */
|
|
return(ast_say_datetime_from_now_en(chan, t, ints, lang));
|
|
}
|
|
|
|
/* English syntax */
|
|
int ast_say_datetime_from_now_en(struct ast_channel *chan, time_t t, const char *ints, const char *lang)
|
|
{
|
|
int res=0;
|
|
time_t nowt;
|
|
int daydiff;
|
|
struct tm tm;
|
|
struct tm now;
|
|
char fn[256];
|
|
|
|
time(&nowt);
|
|
|
|
localtime_r(&t,&tm);
|
|
localtime_r(&nowt,&now);
|
|
daydiff = now.tm_yday - tm.tm_yday;
|
|
if ((daydiff < 0) || (daydiff > 6)) {
|
|
/* Day of month and month */
|
|
if (!res) {
|
|
snprintf(fn, sizeof(fn), "digits/mon-%d", tm.tm_mon);
|
|
res = ast_streamfile(chan, fn, lang);
|
|
if (!res)
|
|
res = ast_waitstream(chan, ints);
|
|
}
|
|
if (!res)
|
|
res = ast_say_number(chan, tm.tm_mday, ints, lang, (char *) NULL);
|
|
|
|
} else if (daydiff) {
|
|
/* Just what day of the week */
|
|
if (!res) {
|
|
snprintf(fn, sizeof(fn), "digits/day-%d", tm.tm_wday);
|
|
res = ast_streamfile(chan, fn, lang);
|
|
if (!res)
|
|
res = ast_waitstream(chan, ints);
|
|
}
|
|
} /* Otherwise, it was today */
|
|
if (!res)
|
|
res = ast_say_time(chan, t, ints, lang);
|
|
return res;
|
|
}
|
|
|
|
/* French syntax */
|
|
int ast_say_datetime_from_now_fr(struct ast_channel *chan, time_t t, const char *ints, const char *lang)
|
|
{
|
|
int res=0;
|
|
time_t nowt;
|
|
int daydiff;
|
|
struct tm tm;
|
|
struct tm now;
|
|
char fn[256];
|
|
|
|
time(&nowt);
|
|
|
|
localtime_r(&t,&tm);
|
|
localtime_r(&nowt,&now);
|
|
daydiff = now.tm_yday - tm.tm_yday;
|
|
if ((daydiff < 0) || (daydiff > 6)) {
|
|
/* Day of month and month */
|
|
if (!res) {
|
|
snprintf(fn, sizeof(fn), "digits/mon-%d", tm.tm_mon);
|
|
res = ast_streamfile(chan, fn, lang);
|
|
if (!res)
|
|
res = ast_waitstream(chan, ints);
|
|
}
|
|
if (!res)
|
|
res = ast_say_number(chan, tm.tm_mday, ints, lang, (char *) NULL);
|
|
|
|
} else if (daydiff) {
|
|
/* Just what day of the week */
|
|
if (!res) {
|
|
snprintf(fn, sizeof(fn), "digits/day-%d", tm.tm_wday);
|
|
res = ast_streamfile(chan, fn, lang);
|
|
if (!res)
|
|
res = ast_waitstream(chan, ints);
|
|
}
|
|
} /* Otherwise, it was today */
|
|
if (!res)
|
|
res = ast_say_time(chan, t, ints, lang);
|
|
return res;
|
|
}
|
|
|
|
/* Portuguese syntax */
|
|
int ast_say_datetime_from_now_pt(struct ast_channel *chan, time_t t, const char *ints, const char *lang)
|
|
{
|
|
int res=0;
|
|
time_t nowt;
|
|
int daydiff;
|
|
struct tm tm;
|
|
struct tm now;
|
|
char fn[256];
|
|
|
|
time(&nowt);
|
|
|
|
localtime_r(&t,&tm);
|
|
localtime_r(&nowt,&now);
|
|
daydiff = now.tm_yday - tm.tm_yday;
|
|
if ((daydiff < 0) || (daydiff > 6)) {
|
|
/* Day of month and month */
|
|
if (!res)
|
|
res = ast_say_number(chan, tm.tm_mday, ints, lang, (char *) NULL);
|
|
if (!res)
|
|
res = wait_file(chan, ints, "digits/pt-de", lang);
|
|
snprintf(fn, sizeof(fn), "digits/mon-%d", tm.tm_mon);
|
|
if (!res)
|
|
res = wait_file(chan, ints, fn, lang);
|
|
|
|
} else if (daydiff) {
|
|
/* Just what day of the week */
|
|
snprintf(fn, sizeof(fn), "digits/day-%d", tm.tm_wday);
|
|
if (!res)
|
|
res = wait_file(chan, ints, fn, lang);
|
|
} /* Otherwise, it was today */
|
|
snprintf(fn, sizeof(fn), "digits/pt-ah");
|
|
if (!res)
|
|
res = wait_file(chan, ints, fn, lang);
|
|
if (tm.tm_hour != 1)
|
|
if (!res)
|
|
res = wait_file(chan, ints, "digits/pt-sss", lang);
|
|
if (!res)
|
|
res = ast_say_time(chan, t, ints, lang);
|
|
return res;
|
|
}
|
|
|
|
|
|
/*********************************** GREEK SUPPORT ***************************************/
|
|
|
|
|
|
/*
|
|
* digits/female-[1..4] : "Mia, dyo , treis, tessereis"
|
|
*/
|
|
static int gr_say_number_female(int num, struct ast_channel *chan, const char *ints, const char *lang){
|
|
int tmp;
|
|
int left;
|
|
int res;
|
|
char fn[256] = "";
|
|
|
|
/* ast_log(LOG_DEBUG, "\n\n Saying number female %s %d \n\n",lang, num); */
|
|
if (num < 5) {
|
|
snprintf(fn, sizeof(fn), "digits/female-%d", num);
|
|
res = wait_file(chan, ints, fn, lang);
|
|
} else if (num < 13) {
|
|
res = ast_say_number(chan, num, ints, lang, (char *) NULL);
|
|
} else if (num <100 ) {
|
|
tmp = (num/10) * 10;
|
|
left = num - tmp;
|
|
snprintf(fn, sizeof(fn), "digits/%d", tmp);
|
|
res = ast_streamfile(chan, fn, lang);
|
|
if (!res)
|
|
res = ast_waitstream(chan, ints);
|
|
if (left)
|
|
gr_say_number_female(left, chan, ints, lang);
|
|
|
|
} else {
|
|
return -1;
|
|
}
|
|
return res;
|
|
}
|
|
|
|
|
|
|
|
/*
|
|
* A list of the files that you need to create
|
|
-> digits/xilia = "xilia"
|
|
-> digits/myrio = "ekatomyrio"
|
|
-> digits/thousands = "xiliades"
|
|
-> digits/millions = "ektatomyria"
|
|
-> digits/[1..12] :: A pronunciation of th digits form 1 to 12 e.g. "tria"
|
|
-> digits/[10..100] :: A pronunciation of the tens from 10 to 90
|
|
e,g 80 = "ogdonta"
|
|
Here we must note that we use digits/tens/100 to utter "ekato"
|
|
and digits/hundred-100 to utter "ekaton"
|
|
-> digits/hundred-[100...1000] :: A pronunciation of hundreds from 100 to 1000 e.g 400 =
|
|
"terakosia". Here again we use hundreds/1000 for "xilia"
|
|
and digits/thousnds for "xiliades"
|
|
*/
|
|
|
|
static int ast_say_number_full_gr(struct ast_channel *chan, int num, const char *ints, const char *language,int audiofd, int ctrlfd)
|
|
{
|
|
int res = 0;
|
|
char fn[256] = "";
|
|
int i=0;
|
|
|
|
|
|
if (!num) {
|
|
snprintf(fn, sizeof(fn), "digits/0");
|
|
res = ast_streamfile(chan, fn, chan->language);
|
|
if (!res)
|
|
return ast_waitstream(chan, ints);
|
|
}
|
|
|
|
while(!res && num ) {
|
|
i++;
|
|
if (num < 13) {
|
|
snprintf(fn, sizeof(fn), "digits/%d", num);
|
|
num = 0;
|
|
} else if (num <= 100) {
|
|
/* 13 < num <= 100 */
|
|
snprintf(fn, sizeof(fn), "digits/%d", (num /10) * 10);
|
|
num -= ((num / 10) * 10);
|
|
} else if (num < 200) {
|
|
/* 100 < num < 200 */
|
|
snprintf(fn, sizeof(fn), "digits/hundred-100");
|
|
num -= ((num / 100) * 100);
|
|
}else if (num < 1000) {
|
|
/* 200 < num < 1000 */
|
|
snprintf(fn, sizeof(fn), "digits/hundred-%d", (num/100)*100);
|
|
num -= ((num / 100) * 100);
|
|
}else if (num < 2000){
|
|
snprintf(fn, sizeof(fn), "digits/xilia");
|
|
num -= ((num / 1000) * 1000);
|
|
}
|
|
else {
|
|
/* num > 1000 */
|
|
if (num < 1000000) {
|
|
res = ast_say_number_full_gr(chan, (num / 1000), ints, chan->language, audiofd, ctrlfd);
|
|
if (res)
|
|
return res;
|
|
num = num % 1000;
|
|
snprintf(fn, sizeof(fn), "digits/thousands");
|
|
} else {
|
|
if (num < 1000000000) { /* 1,000,000,000 */
|
|
res = ast_say_number_full_gr(chan, (num / 1000000), ints, chan->language ,audiofd, ctrlfd);
|
|
if (res)
|
|
return res;
|
|
num = num % 1000000;
|
|
snprintf(fn, sizeof(fn), "digits/millions");
|
|
} else {
|
|
ast_log(LOG_DEBUG, "Number '%d' is too big for me\n", num);
|
|
res = -1;
|
|
}
|
|
}
|
|
}
|
|
if (!res) {
|
|
if(!ast_streamfile(chan, fn, language)) {
|
|
if ((audiofd > -1) && (ctrlfd > -1))
|
|
res = ast_waitstream_full(chan, ints, audiofd, ctrlfd);
|
|
else
|
|
res = ast_waitstream(chan, ints);
|
|
}
|
|
ast_stopstream(chan);
|
|
}
|
|
}
|
|
return res;
|
|
}
|
|
|
|
|
|
/*
|
|
* The format is weekday - day - month -year
|
|
*
|
|
* A list of the files that you need to create
|
|
* digits/day-[1..7] : "Deytera .. Paraskeyh"
|
|
* digits/months/1..12 : "Ianouariou .. Dekembriou"
|
|
Attention the months are in
|
|
"gekinh klhsh"
|
|
*/
|
|
|
|
|
|
static int ast_say_date_gr(struct ast_channel *chan, time_t t, const char *ints, const char *lang)
|
|
{
|
|
struct tm tm;
|
|
|
|
char fn[256];
|
|
int res = 0;
|
|
|
|
|
|
ast_localtime(&t,&tm,NULL);
|
|
/* W E E K - D A Y */
|
|
if (!res) {
|
|
snprintf(fn, sizeof(fn), "digits/day-%d", tm.tm_wday);
|
|
res = ast_streamfile(chan, fn, lang);
|
|
if (!res)
|
|
res = ast_waitstream(chan, ints);
|
|
}
|
|
/* D A Y */
|
|
if (!res) {
|
|
gr_say_number_female(tm.tm_mday, chan, ints, lang);
|
|
}
|
|
/* M O N T H */
|
|
if (!res) {
|
|
snprintf(fn, sizeof(fn), "digits/mon-%d", tm.tm_mon);
|
|
res = ast_streamfile(chan, fn, lang);
|
|
if (!res)
|
|
res = ast_waitstream(chan, ints);
|
|
}
|
|
/* Y E A R */
|
|
if (!res)
|
|
res = ast_say_number(chan, tm.tm_year + 1900, ints, lang, (char *) NULL);
|
|
return res;
|
|
}
|
|
|
|
|
|
|
|
/* A list of the files that you need to create
|
|
* digits/female/1..4 : "Mia, dyo , treis, tesseris "
|
|
* digits/kai : "KAI"
|
|
* didgits : "h wra"
|
|
* digits/p-m : "meta meshmbrias"
|
|
* digits/a-m : "pro meshmbrias"
|
|
*/
|
|
|
|
static int ast_say_time_gr(struct ast_channel *chan, time_t t, const char *ints, const char *lang)
|
|
{
|
|
|
|
struct tm tm;
|
|
int res = 0;
|
|
int hour, pm=0;
|
|
|
|
localtime_r(&t,&tm);
|
|
hour = tm.tm_hour;
|
|
|
|
if (!hour)
|
|
hour = 12;
|
|
else if (hour == 12)
|
|
pm = 1;
|
|
else if (hour > 12) {
|
|
hour -= 12;
|
|
pm = 1;
|
|
}
|
|
|
|
res = gr_say_number_female(hour, chan, ints, lang);
|
|
if (tm.tm_min) {
|
|
if (!res)
|
|
res = ast_streamfile(chan, "digits/kai", lang);
|
|
if (!res)
|
|
res = ast_waitstream(chan, ints);
|
|
if (!res)
|
|
res = ast_say_number(chan, tm.tm_min, ints, lang, (char *) NULL);
|
|
} else {
|
|
if (!res)
|
|
res = ast_streamfile(chan, "digits/hwra", lang);
|
|
if (!res)
|
|
res = ast_waitstream(chan, ints);
|
|
}
|
|
if (pm) {
|
|
if (!res)
|
|
res = ast_streamfile(chan, "digits/p-m", lang);
|
|
} else {
|
|
if (!res)
|
|
res = ast_streamfile(chan, "digits/a-m", lang);
|
|
}
|
|
if (!res)
|
|
res = ast_waitstream(chan, ints);
|
|
return res;
|
|
}
|
|
|
|
|
|
|
|
static int ast_say_datetime_gr(struct ast_channel *chan, time_t t, const char *ints, const char *lang)
|
|
{
|
|
struct tm tm;
|
|
char fn[256];
|
|
int res = 0;
|
|
localtime_r(&t,&tm);
|
|
|
|
|
|
/* W E E K - D A Y */
|
|
if (!res) {
|
|
snprintf(fn, sizeof(fn), "digits/day-%d", tm.tm_wday);
|
|
res = ast_streamfile(chan, fn, lang);
|
|
if (!res)
|
|
res = ast_waitstream(chan, ints);
|
|
}
|
|
/* D A Y */
|
|
if (!res) {
|
|
gr_say_number_female(tm.tm_mday, chan, ints, lang);
|
|
}
|
|
/* M O N T H */
|
|
if (!res) {
|
|
snprintf(fn, sizeof(fn), "digits/mon-%d", tm.tm_mon);
|
|
res = ast_streamfile(chan, fn, lang);
|
|
if (!res)
|
|
res = ast_waitstream(chan, ints);
|
|
}
|
|
|
|
res = ast_say_time_gr(chan, t, ints, lang);
|
|
return res;
|
|
}
|
|
|
|
static int ast_say_date_with_format_gr(struct ast_channel *chan, time_t time, const char *ints, const char *lang, const char *format, const char *timezone)
|
|
{
|
|
|
|
struct tm tm;
|
|
int res=0, offset, sndoffset;
|
|
char sndfile[256], nextmsg[256];
|
|
|
|
ast_localtime(&time,&tm,timezone);
|
|
|
|
for (offset=0 ; format[offset] != '\0' ; offset++) {
|
|
ast_log(LOG_DEBUG, "Parsing %c (offset %d) in %s\n", format[offset], offset, format);
|
|
switch (format[offset]) {
|
|
/* NOTE: if you add more options here, please try to be consistent with strftime(3) */
|
|
case '\'':
|
|
/* Literal name of a sound file */
|
|
sndoffset=0;
|
|
for (sndoffset=0 ; (format[++offset] != '\'') && (sndoffset < 256) ; sndoffset++)
|
|
sndfile[sndoffset] = format[offset];
|
|
sndfile[sndoffset] = '\0';
|
|
res = wait_file(chan,ints,sndfile,lang);
|
|
break;
|
|
case 'A':
|
|
case 'a':
|
|
/* Sunday - Saturday */
|
|
snprintf(nextmsg,sizeof(nextmsg), "digits/day-%d", tm.tm_wday);
|
|
res = wait_file(chan,ints,nextmsg,lang);
|
|
break;
|
|
case 'B':
|
|
case 'b':
|
|
case 'h':
|
|
/* January - December */
|
|
snprintf(nextmsg,sizeof(nextmsg), "digits/mon-%d", tm.tm_mon);
|
|
res = wait_file(chan,ints,nextmsg,lang);
|
|
break;
|
|
case 'd':
|
|
case 'e':
|
|
/* first - thirtyfirst */
|
|
gr_say_number_female(tm.tm_mday, chan, ints, lang);
|
|
break;
|
|
case 'Y':
|
|
/* Year */
|
|
|
|
ast_say_number_full_gr(chan, 1900+tm.tm_year, ints, chan->language, -1, -1);
|
|
break;
|
|
case 'I':
|
|
case 'l':
|
|
/* 12-Hour */
|
|
if (tm.tm_hour == 0)
|
|
gr_say_number_female(12, chan, ints, lang);
|
|
else if (tm.tm_hour > 12)
|
|
gr_say_number_female(tm.tm_hour - 12, chan, ints, lang);
|
|
else
|
|
gr_say_number_female(tm.tm_hour, chan, ints, lang);
|
|
break;
|
|
case 'H':
|
|
case 'k':
|
|
/* 24-Hour */
|
|
gr_say_number_female(tm.tm_hour, chan, ints, lang);
|
|
break;
|
|
case 'M':
|
|
/* Minute */
|
|
if (tm.tm_min) {
|
|
if (!res)
|
|
res = ast_streamfile(chan, "digits/kai", lang);
|
|
if (!res)
|
|
res = ast_waitstream(chan, ints);
|
|
if (!res)
|
|
res = ast_say_number_full_gr(chan, tm.tm_min, ints, lang, -1, -1);
|
|
} else {
|
|
if (!res)
|
|
res = ast_streamfile(chan, "digits/oclock", lang);
|
|
if (!res)
|
|
res = ast_waitstream(chan, ints);
|
|
}
|
|
break;
|
|
case 'P':
|
|
case 'p':
|
|
/* AM/PM */
|
|
if (tm.tm_hour > 11)
|
|
snprintf(nextmsg,sizeof(nextmsg), "digits/p-m");
|
|
else
|
|
snprintf(nextmsg,sizeof(nextmsg), "digits/a-m");
|
|
res = wait_file(chan,ints,nextmsg,lang);
|
|
break;
|
|
case 'Q':
|
|
/* Shorthand for "Today", "Yesterday", or ABdY */
|
|
{
|
|
struct timeval now;
|
|
struct tm tmnow;
|
|
time_t beg_today;
|
|
|
|
gettimeofday(&now,NULL);
|
|
ast_localtime(&now.tv_sec,&tmnow,timezone);
|
|
/* This might be slightly off, if we transcend a leap second, but never more off than 1 second */
|
|
/* In any case, it saves not having to do ast_mktime() */
|
|
beg_today = now.tv_sec - (tmnow.tm_hour * 3600) - (tmnow.tm_min * 60) - (tmnow.tm_sec);
|
|
if (beg_today < time) {
|
|
/* Today */
|
|
res = wait_file(chan,ints, "digits/today",lang);
|
|
} else if (beg_today - 86400 < time) {
|
|
/* Yesterday */
|
|
res = wait_file(chan,ints, "digits/yesterday",lang);
|
|
} else {
|
|
res = ast_say_date_with_format(chan, time, ints, lang, "AdBY", timezone);
|
|
}
|
|
}
|
|
break;
|
|
case 'q':
|
|
/* Shorthand for "" (today), "Yesterday", A (weekday), or ABdY */
|
|
{
|
|
struct timeval now;
|
|
struct tm tmnow;
|
|
time_t beg_today;
|
|
|
|
gettimeofday(&now,NULL);
|
|
ast_localtime(&now.tv_sec,&tmnow,timezone);
|
|
/* This might be slightly off, if we transcend a leap second, but never more off than 1 second */
|
|
/* In any case, it saves not having to do ast_mktime() */
|
|
beg_today = now.tv_sec - (tmnow.tm_hour * 3600) - (tmnow.tm_min * 60) - (tmnow.tm_sec);
|
|
if (beg_today < time) {
|
|
/* Today */
|
|
} else if ((beg_today - 86400) < time) {
|
|
/* Yesterday */
|
|
res = wait_file(chan,ints, "digits/yesterday",lang);
|
|
} else if (beg_today - 86400 * 6 < time) {
|
|
/* Within the last week */
|
|
res = ast_say_date_with_format(chan, time, ints, lang, "A", timezone);
|
|
} else {
|
|
res = ast_say_date_with_format(chan, time, ints, lang, "AdBY", timezone);
|
|
}
|
|
}
|
|
break;
|
|
case 'R':
|
|
res = ast_say_date_with_format(chan, time, ints, lang, "HM", timezone);
|
|
break;
|
|
case 'S':
|
|
/* Seconds */
|
|
snprintf(nextmsg,sizeof(nextmsg), "digits/kai");
|
|
res = wait_file(chan,ints,nextmsg,lang);
|
|
if (!res)
|
|
res = ast_say_number_full_gr(chan, tm.tm_sec, ints, lang, -1, -1);
|
|
if (!res)
|
|
snprintf(nextmsg,sizeof(nextmsg), "digits/seconds");
|
|
res = wait_file(chan,ints,nextmsg,lang);
|
|
break;
|
|
case 'T':
|
|
res = ast_say_date_with_format(chan, time, ints, lang, "HMS", timezone);
|
|
break;
|
|
case ' ':
|
|
case ' ':
|
|
/* Just ignore spaces and tabs */
|
|
break;
|
|
default:
|
|
/* Unknown character */
|
|
ast_log(LOG_WARNING, "Unknown character in datetime format %s: %c at pos %d\n", format, format[offset], offset);
|
|
}
|
|
/* Jump out on DTMF */
|
|
if (res) {
|
|
break;
|
|
}
|
|
}
|
|
return res;
|
|
}
|