2 * parsetime.c - parse time for at(1)
3 * Copyright (C) 1993, 1994 Thomas Koenig
5 * modifications for English-language times
6 * Copyright (C) 1993 David Parsons
8 * A lot of modifications and extensions
9 * (including the new syntax being useful for RRDB)
10 * Copyright (C) 1999 Oleg Cherevko (aka Olwi Deer)
12 * severe structural damage inflicted by Tobi Oetiker in 1999
14 * Redistribution and use in source and binary forms, with or without
15 * modification, are permitted provided that the following conditions
17 * 1. Redistributions of source code must retain the above copyright
18 * notice, this list of conditions and the following disclaimer.
19 * 2. The name of the author(s) may not be used to endorse or promote
20 * products derived from this software without specific prior written
23 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR(S) ``AS IS'' AND ANY EXPRESS OR
24 * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
25 * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
26 * IN NO EVENT SHALL THE AUTHOR(S) BE LIABLE FOR ANY DIRECT, INDIRECT,
27 * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
28 * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
29 * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
30 * THEORY OF LIABILITY, WETHER IN CONTRACT, STRICT LIABILITY, OR TORT
31 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF
32 * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
35 /* NOTE: nothing in here is thread-safe!!!! Not even the localtime
39 * The BNF-like specification of the time syntax parsed is below:
41 * As usual, [ X ] means that X is optional, { X } means that X may
42 * be either omitted or specified as many times as needed,
43 * alternatives are separated by |, brackets are used for grouping.
44 * (# marks the beginning of comment that extends to the end of line)
46 * TIME-SPECIFICATION ::= TIME-REFERENCE [ OFFSET-SPEC ] |
48 * ( START | END ) OFFSET-SPEC
50 * TIME-REFERENCE ::= NOW | TIME-OF-DAY-SPEC [ DAY-SPEC-1 ] |
51 * [ TIME-OF-DAY-SPEC ] DAY-SPEC-2
53 * TIME-OF-DAY-SPEC ::= NUMBER (':') NUMBER [am|pm] | # HH:MM
54 * 'noon' | 'midnight' | 'teatime'
56 * DAY-SPEC-1 ::= NUMBER '/' NUMBER '/' NUMBER | # MM/DD/[YY]YY
57 * NUMBER '.' NUMBER '.' NUMBER | # DD.MM.[YY]YY
58 * NUMBER # Seconds since 1970
61 * DAY-SPEC-2 ::= MONTH-NAME NUMBER [NUMBER] | # Month DD [YY]YY
62 * 'yesterday' | 'today' | 'tomorrow' |
66 * OFFSET-SPEC ::= '+'|'-' NUMBER TIME-UNIT { ['+'|'-'] NUMBER TIME-UNIT }
68 * TIME-UNIT ::= SECONDS | MINUTES | HOURS |
69 * DAYS | WEEKS | MONTHS | YEARS
73 * START ::= 'start' | 's'
76 * SECONDS ::= 'seconds' | 'second' | 'sec' | 's'
77 * MINUTES ::= 'minutes' | 'minute' | 'min' | 'm'
78 * HOURS ::= 'hours' | 'hour' | 'hr' | 'h'
79 * DAYS ::= 'days' | 'day' | 'd'
80 * WEEKS ::= 'weeks' | 'week' | 'wk' | 'w'
81 * MONTHS ::= 'months' | 'month' | 'mon' | 'm'
82 * YEARS ::= 'years' | 'year' | 'yr' | 'y'
84 * MONTH-NAME ::= 'jan' | 'january' | 'feb' | 'february' | 'mar' | 'march' |
85 * 'apr' | 'april' | 'may' | 'jun' | 'june' | 'jul' | 'july' |
86 * 'aug' | 'august' | 'sep' | 'september' | 'oct' | 'october' |
87 * 'nov' | 'november' | 'dec' | 'december'
89 * DAY-OF-WEEK ::= 'sunday' | 'sun' | 'monday' | 'mon' | 'tuesday' | 'tue' |
90 * 'wednesday' | 'wed' | 'thursday' | 'thu' | 'friday' | 'fri' |
94 * As you may note, there is an ambiguity with respect to
95 * the 'm' time unit (which can mean either minutes or months).
96 * To cope with this, code tries to read users mind :) by applying
97 * certain heuristics. There are two of them:
99 * 1. If 'm' is used in context of (i.e. right after the) years,
100 * months, weeks, or days it is assumed to mean months, while
101 * in the context of hours, minutes, and seconds it means minutes.
102 * (e.g., in -1y6m or +3w1m 'm' means 'months', while in
103 * -3h20m or +5s2m 'm' means 'minutes')
105 * 2. Out of context (i.e. right after the '+' or '-' sign) the
106 * meaning of 'm' is guessed from the number it directly follows.
107 * Currently, if the number absolute value is below 25 it is assumed
108 * that 'm' means months, otherwise it is treated as minutes.
109 * (e.g., -25m == -25 minutes, while +24m == +24 months)
117 #include "rrd_tool.h"
120 /* Structures and unions */
123 MIDNIGHT, NOON, TEATIME,
124 PM, AM, YESTERDAY, TODAY, TOMORROW, NOW, START, END,
125 SECONDS, MINUTES, HOURS, DAYS, WEEKS, MONTHS, YEARS,
127 NUMBER, PLUS, MINUS, DOT, COLON, SLASH, ID, JUNK,
128 JAN, FEB, MAR, APR, MAY, JUN,
129 JUL, AUG, SEP, OCT, NOV, DEC,
130 SUN, MON, TUE, WED, THU, FRI, SAT
133 /* the below is for plus_minus() */
134 #define PREVIOUS_OP (-1)
136 /* parse translation table - table driven parsers can be your FRIEND!
138 struct SpecialToken {
139 char *name; /* token name */
140 int value; /* token id */
142 static struct SpecialToken VariousWords[] = {
143 { "midnight", MIDNIGHT }, /* 00:00:00 of today or tomorrow */
144 { "noon", NOON }, /* 12:00:00 of today or tomorrow */
145 { "teatime", TEATIME }, /* 16:00:00 of today or tomorrow */
146 { "am", AM }, /* morning times for 0-12 clock */
147 { "pm", PM }, /* evening times for 0-12 clock */
148 { "tomorrow", TOMORROW },
149 { "yesterday", YESTERDAY },
178 { "september", SEP },
188 { "wednesday", WED },
196 { NULL, 0 } /*** SENTINEL ***/
199 static struct SpecialToken TimeMultipliers[] = {
200 { "second", SECONDS }, /* seconds multiplier */
201 { "seconds", SECONDS }, /* (pluralized) */
202 { "sec", SECONDS }, /* (generic) */
203 { "s", SECONDS }, /* (short generic) */
204 { "minute", MINUTES }, /* minutes multiplier */
205 { "minutes", MINUTES }, /* (pluralized) */
206 { "min", MINUTES }, /* (generic) */
207 { "m", MONTHS_MINUTES }, /* (short generic) */
208 { "hour", HOURS }, /* hours ... */
209 { "hours", HOURS }, /* (pluralized) */
210 { "hr", HOURS }, /* (generic) */
211 { "h", HOURS }, /* (short generic) */
212 { "day", DAYS }, /* days ... */
213 { "days", DAYS }, /* (pluralized) */
214 { "d", DAYS }, /* (short generic) */
215 { "week", WEEKS }, /* week ... */
216 { "weeks", WEEKS }, /* (pluralized) */
217 { "wk", WEEKS }, /* (generic) */
218 { "w", WEEKS }, /* (short generic) */
219 { "month", MONTHS }, /* week ... */
220 { "months", MONTHS }, /* (pluralized) */
221 { "mon", MONTHS }, /* (generic) */
222 { "year", YEARS }, /* year ... */
223 { "years", YEARS }, /* (pluralized) */
224 { "yr", YEARS }, /* (generic) */
225 { "y", YEARS }, /* (short generic) */
226 { NULL, 0 } /*** SENTINEL ***/
229 /* File scope variables */
231 /* context dependent list of specials for parser to recognize,
232 * required for us to be able distinguish between 'mon' as 'month'
233 * and 'mon' as 'monday'
235 static struct SpecialToken *Specials;
237 static const char **scp; /* scanner - pointer at arglist */
238 static char scc; /* scanner - count of remaining arguments */
239 static const char *sct; /* scanner - next char pointer in current argument */
240 static int need; /* scanner - need to advance to next argument */
242 static char *sc_token=NULL; /* scanner - token buffer */
243 static size_t sc_len; /* scanner - length of token buffer */
244 static int sc_tokid; /* scanner - token id */
246 /* Local functions */
247 static void EnsureMemFree (void);
249 static void EnsureMemFree (void)
259 * A hack to compensate for the lack of the C++ exceptions
261 * Every function func that might generate parsing "exception"
262 * should return TIME_OK (aka NULL) or pointer to the error message,
263 * and should be called like this: try(func(args));
265 * if the try is not successful it will reset the token pointer ...
267 * [NOTE: when try(...) is used as the only statement in the "if-true"
268 * part of the if statement that also has an "else" part it should be
269 * either enclosed in the curly braces (despite the fact that it looks
270 * like a single statement) or NOT followed by the ";"]
282 * The panic() function was used in the original code to die, we redefine
283 * it as macro to start the chain of ascending returns that in conjunction
284 * with the try(b) above will simulate a sort of "exception handling"
292 * ve() and e() are used to set the return error,
293 * the most appropriate use for these is inside panic(...)
295 #define MAX_ERR_MSG_LEN 1024
296 static char errmsg[ MAX_ERR_MSG_LEN ];
299 ve ( char *fmt, va_list ap )
301 #ifdef HAVE_VSNPRINTF
302 vsnprintf( errmsg, MAX_ERR_MSG_LEN, fmt, ap );
304 vsprintf( errmsg, fmt, ap );
321 /* Compare S1 and S2, ignoring case, returning less than, equal to or
322 greater than zero if S1 is lexicographically less than,
323 equal to or greater than S2. -- copied from GNU libc*/
325 mystrcasecmp (s1, s2)
329 const unsigned char *p1 = (const unsigned char *) s1;
330 const unsigned char *p2 = (const unsigned char *) s2;
331 unsigned char c1, c2;
338 c1 = tolower (*p1++);
339 c2 = tolower (*p2++);
349 * parse a token, checking if it's something special to us
352 parse_token(char *arg)
356 for (i=0; Specials[i].name != NULL; i++)
357 if (mystrcasecmp(Specials[i].name, arg) == 0)
358 return sc_tokid = Specials[i].value;
360 /* not special - must be some random id */
361 return sc_tokid = ID;
367 * init_scanner() sets up the scanner to eat arguments
370 init_scanner(int argc, const char **argv)
377 sc_len += strlen(*argv++);
379 sc_token = (char *) malloc(sc_len*sizeof(char));
380 if( sc_token == NULL )
381 return "Failed to allocate memory";
386 * token() fetches a token from the input stream
394 memset(sc_token, '\0', sc_len);
398 /* if we need to read another argument, walk along the argument list;
399 * when we fall off the arglist, we'll just return EOF forever
409 /* eat whitespace now - if we walk off the end of the argument,
410 * we'll continue, which puts us up at the top of the while loop
411 * to fetch the next argument in
413 while (isspace((unsigned char)*sct) || *sct == '_' || *sct == ',' )
420 /* preserve the first character of the new token
422 sc_token[0] = *sct++;
424 /* then see what it is
426 if (isdigit((unsigned char)(sc_token[0]))) {
427 while (isdigit((unsigned char)(*sct)))
428 sc_token[++idx] = *sct++;
429 sc_token[++idx] = '\0';
430 return sc_tokid = NUMBER;
432 else if (isalpha((unsigned char)(sc_token[0]))) {
433 while (isalpha((unsigned char)(*sct)))
434 sc_token[++idx] = *sct++;
435 sc_token[++idx] = '\0';
436 return parse_token(sc_token);
438 else switch(sc_token[0]) {
439 case ':': return sc_tokid = COLON;
440 case '.': return sc_tokid = DOT;
441 case '+': return sc_tokid = PLUS;
442 case '-': return sc_tokid = MINUS;
443 case '/': return sc_tokid = SLASH;
445 /*OK, we did not make it ... */
447 return sc_tokid = EOF;
454 * expect2() gets a token and complains if it's not the token we want
457 expect2(int desired, char *complain_fmt, ...)
460 va_start( ap, complain_fmt );
461 if (token() != desired) {
462 panic(ve( complain_fmt, ap ));
471 * plus_minus() is used to parse a single NUMBER TIME-UNIT pair
472 * for the OFFSET-SPEC.
473 * It also applies those m-guessing heuristics.
476 plus_minus(struct rrd_time_value *ptv, int doop)
478 static int op = PLUS;
479 static int prev_multiplier = -1;
485 try(expect2(NUMBER,"There should be number after '%c'", op == PLUS ? '+' : '-'));
486 prev_multiplier = -1; /* reset months-minutes guessing mechanics */
488 /* if doop is < 0 then we repeat the previous op
489 * with the prefetched number */
491 delta = atoi(sc_token);
493 if( token() == MONTHS_MINUTES )
495 /* hard job to guess what does that -5m means: -5mon or -5min? */
496 switch(prev_multiplier)
512 if( delta < 6 ) /* it may be some other value but in the context
513 * of RRD who needs less than 6 min deltas? */
519 prev_multiplier = sc_tokid;
522 ptv->tm.tm_year += (op == PLUS) ? delta : -delta;
525 ptv->tm.tm_mon += (op == PLUS) ? delta : -delta;
531 ptv->tm.tm_mday += (op == PLUS) ? delta : -delta;
534 ptv->offset += (op == PLUS) ? delta*60*60 : -delta*60*60;
537 ptv->offset += (op == PLUS) ? delta*60 : -delta*60;
540 ptv->offset += (op == PLUS) ? delta : -delta;
542 default: /*default unit is seconds */
543 ptv->offset += (op == PLUS) ? delta : -delta;
546 panic(e("well-known time unit expected after %d", delta));
548 return TIME_OK; /* to make compiler happy :) */
553 * tod() computes the time of day (TIME-OF-DAY-SPEC)
556 tod(struct rrd_time_value *ptv)
558 int hour, minute = 0;
560 /* save token status in case we must abort */
562 const char *sct_sv = sct;
563 int sc_tokid_sv = sc_tokid;
565 tlen = strlen(sc_token);
567 /* first pick out the time of day - we assume a HH (COLON|DOT) MM time
573 hour = atoi(sc_token);
576 if (sc_tokid == SLASH || sc_tokid == DOT) {
577 /* guess we are looking at a date */
580 sc_tokid = sc_tokid_sv;
581 sprintf (sc_token,"%d", hour);
584 if (sc_tokid == COLON ) {
586 "Parsing HH:MM syntax, expecting MM as number, got none"));
587 minute = atoi(sc_token);
589 panic(e("parsing HH:MM syntax, got MM = %d (>59!)", minute ));
594 /* check if an AM or PM specifier was given
596 if (sc_tokid == AM || sc_tokid == PM) {
598 panic(e("there cannot be more than 12 AM or PM hours"));
600 if (sc_tokid == PM) {
601 if (hour != 12) /* 12:xx PM is 12:xx, not 24:xx */
604 if (hour == 12) /* 12:xx AM is 00:xx, not 12:xx */
609 else if (hour > 23) {
610 /* guess it was not a time then ... */
613 sc_tokid = sc_tokid_sv;
614 sprintf (sc_token,"%d", hour);
617 ptv->tm.tm_hour = hour;
618 ptv->tm.tm_min = minute;
620 if (ptv->tm.tm_hour == 24) {
629 * assign_date() assigns a date, adjusting year as appropriate
632 assign_date(struct rrd_time_value *ptv, long mday, long mon, long year)
638 panic(e("invalid year %d (should be either 00-99 or >1900)",
641 } else if( year >= 0 && year < 38 ) {
642 year += 100; /* Allow year 2000-2037 to be specified as */
643 } /* 00-37 until the problem of 2038 year will */
644 /* arise for unices with 32-bit time_t :) */
646 panic(e("won't handle dates before epoch (01/01/1970), sorry"));
649 ptv->tm.tm_mday = mday;
650 ptv->tm.tm_mon = mon;
651 ptv->tm.tm_year = year;
657 * day() picks apart DAY-SPEC-[12]
660 day(struct rrd_time_value *ptv)
662 /* using time_t seems to help portability with 64bit oses */
663 time_t mday=0, wday, mon, year = ptv->tm.tm_year;
670 case TODAY: /* force ourselves to stay in today - no further processing */
678 case JAN: case FEB: case MAR: case APR: case MAY: case JUN:
679 case JUL: case AUG: case SEP: case OCT: case NOV: case DEC:
680 /* do month mday [year]
682 mon = (sc_tokid-JAN);
684 "the day of the month should follow month name"));
685 mday = atol(sc_token);
686 if (token() == NUMBER) {
687 year = atol(sc_token);
691 year = ptv->tm.tm_year;
692 try(assign_date(ptv, mday, mon, year));
695 case SUN: case MON: case TUE:
696 case WED: case THU: case FRI:
698 /* do a particular day of the week
700 wday = (sc_tokid-SUN);
701 ptv->tm.tm_mday += (wday - ptv->tm.tm_wday);
705 mday = ptv->tm.tm_mday;
706 mday += (wday - ptv->tm.tm_wday);
707 ptv->tm.tm_wday = wday;
709 try(assign_date(ptv, mday, ptv->tm.tm_mon, ptv->tm.tm_year));
714 /* get numeric <sec since 1970>, MM/DD/[YY]YY, or DD.MM.[YY]YY
716 tlen = strlen(sc_token);
717 mon = atol(sc_token);
718 if (mon > 10*365*24*60*60) {
719 ptv->tm=*localtime(&mon);
724 if (mon > 19700101 && mon < 24000101){ /*works between 1900 and 2400 */
725 char cmon[3],cmday[3],cyear[5];
726 strncpy(cyear,sc_token,4);cyear[4]='\0';
728 strncpy(cmon,&(sc_token[4]),2);cmon[2]='\0';
730 strncpy(cmday,&(sc_token[6]),2);cmday[2]='\0';
736 if (mon <= 31 && (sc_tokid == SLASH || sc_tokid == DOT)) {
739 try(expect2(NUMBER,"there should be %s number after '%c'",
740 sep == DOT ? "month" : "day", sep == DOT ? '.' : '/'));
741 mday = atol(sc_token);
742 if (token() == sep) {
743 try(expect2(NUMBER,"there should be year number after '%c'",
744 sep == DOT ? '.' : '/'));
745 year = atol(sc_token);
749 /* flip months and days for European timing
760 if(mon < 0 || mon > 11 ) {
761 panic(e("did you really mean month %d?", mon+1));
763 if(mday < 1 || mday > 31) {
764 panic(e("I'm afraid that %d is not a valid day of the month",
767 try(assign_date(ptv, mday, mon, year));
774 /* Global functions */
778 * parsetime() is the external interface that takes tspec, parses
779 * it and puts the result in the rrd_time_value structure *ptv.
780 * It can return either absolute times (these are ensured to be
781 * correct) or relative time references that are expected to be
782 * added to some absolute time value and then normalized by
783 * mktime() The return value is either TIME_OK (aka NULL) or
784 * the pointer to the error message in the case of problems
787 parsetime(const char *tspec, struct rrd_time_value *ptv)
789 time_t now = time(NULL);
791 /* this MUST be initialized to zero for midnight/noon/teatime */
793 Specials = VariousWords; /* initialize special words context */
795 try(init_scanner( 1, &tspec ));
797 /* establish the default time reference */
798 ptv->type = ABSOLUTE_TIME;
800 ptv->tm = *localtime(&now);
801 ptv->tm.tm_isdst = -1; /* mk time can figure this out for us ... */
807 break; /* jump to OFFSET-SPEC part */
810 ptv->type = RELATIVE_TO_START_TIME;
813 ptv->type = RELATIVE_TO_END_TIME;
824 int time_reference = sc_tokid;
826 if( sc_tokid == PLUS || sc_tokid == MINUS )
828 if( time_reference != NOW ) {
829 panic(e("'start' or 'end' MUST be followed by +|- offset"));
832 if( sc_tokid != EOF ) {
833 panic(e("if 'now' is followed by a token it must be +|- offset"));
838 /* Only absolute time specifications below */
843 /* fix month parsing */
844 case JAN: case FEB: case MAR: case APR: case MAY: case JUN:
845 case JUL: case AUG: case SEP: case OCT: case NOV: case DEC:
847 if (sc_tokid != NUMBER) break;
851 /* evil coding for TEATIME|NOON|MIDNIGHT - we've initialized
852 * hr to zero up above, then fall into this case in such a
853 * way so we add +12 +4 hours to it for teatime, +12 hours
854 * to it for noon, and nothing at all for midnight, then
855 * set our rettime to that hour before leaping into the
865 /* if (ptv->tm.tm_hour >= hr) {
868 } */ /* shifting does not makes sense here ... noon is noon */
869 ptv->tm.tm_hour = hr;
876 panic(e("unparsable time: %s%s",sc_token,sct));
878 } /* ugly case statement */
881 * the OFFSET-SPEC part
883 * (NOTE, the sc_tokid was prefetched for us by the previous code)
885 if( sc_tokid == PLUS || sc_tokid == MINUS ) {
886 Specials = TimeMultipliers; /* switch special words context */
887 while( sc_tokid == PLUS || sc_tokid == MINUS ||
888 sc_tokid == NUMBER ) {
889 if( sc_tokid == NUMBER ) {
890 try(plus_minus(ptv, PREVIOUS_OP ));
892 try(plus_minus(ptv, sc_tokid));
893 token(); /* We will get EOF eventually but that's OK, since
894 token() will return us as many EOFs as needed */
898 /* now we should be at EOF */
899 if( sc_tokid != EOF ) {
900 panic(e("unparsable trailing text: '...%s%s'", sc_token, sct));
903 ptv->tm.tm_isdst = -1; /* for mktime to guess DST status */
904 if( ptv->type == ABSOLUTE_TIME )
905 if( mktime( &ptv->tm ) == -1 ) { /* normalize & check */
906 /* can happen for "nonexistent" times, e.g. around 3am */
907 /* when winter -> summer time correction eats a hour */
908 panic(e("the specified time is incorrect (out of range?)"));
915 int proc_start_end (struct rrd_time_value *start_tv,
916 struct rrd_time_value *end_tv,
919 if (start_tv->type == RELATIVE_TO_END_TIME && /* same as the line above */
920 end_tv->type == RELATIVE_TO_START_TIME) {
921 rrd_set_error("the start and end times cannot be specified "
922 "relative to each other");
926 if (start_tv->type == RELATIVE_TO_START_TIME) {
927 rrd_set_error("the start time cannot be specified relative to itself");
931 if (end_tv->type == RELATIVE_TO_END_TIME) {
932 rrd_set_error("the end time cannot be specified relative to itself");
936 if( start_tv->type == RELATIVE_TO_END_TIME) {
938 *end = mktime(&(end_tv->tm)) + end_tv->offset;
939 tmtmp = *localtime(end); /* reinit end including offset */
940 tmtmp.tm_mday += start_tv->tm.tm_mday;
941 tmtmp.tm_mon += start_tv->tm.tm_mon;
942 tmtmp.tm_year += start_tv->tm.tm_year;
943 *start = mktime(&tmtmp) + start_tv->offset;
945 *start = mktime(&(start_tv->tm)) + start_tv->offset;
947 if (end_tv->type == RELATIVE_TO_START_TIME) {
949 *start = mktime(&(start_tv->tm)) + start_tv->offset;
950 tmtmp = *localtime(start);
951 tmtmp.tm_mday += end_tv->tm.tm_mday;
952 tmtmp.tm_mon += end_tv->tm.tm_mon;
953 tmtmp.tm_year += end_tv->tm.tm_year;
954 *end = mktime(&tmtmp) + end_tv->offset;
956 *end = mktime(&(end_tv->tm)) + end_tv->offset;
959 } /* proc_start_end */