- tz - lists.iana.org

Re: August 13 C Standard Rationale on mktime
by ado Sept. 1, 1987

Sept. 1, 1987

> . . .the sole purpose of mktime is for doing arithmetic on time_t's. . . > if the caller has to set tm_isdst before calling mktime(). > The use that most people want for mktime() is to be able to parse > a string (or whatever) into the tm_hour fields, etc, and then produce > a time_t .. having to second guess whether dst applied at that particular > time is never going to work. . . Well, much of the time it will work if you're prepared to check the values that mktime drops into the "struct tm" passed to it, and if you've saved away the values you passed in. You guess one way or the other on tm_isdst and call mktime. If mktime changes tm_hour on you, you guess "the other" way and try again. Second guessing, yes; third guessing, no. Of course the Standard is mum about the need to do this, and about the need to handle other problems mentioned by kre. --ado

1 0

August 13 C Standard Rationale on mktime; seismo's retirement
by ado Sept. 1, 1987

Sept. 1, 1987

The attached material appears on pages 99 and 100 of the 13 August 1987 "Rationale for Draft Proposed American National Standard for Information Systems Programming Language C". Note that mktime is supposed to examine tm_isdst when converting a "struct tm" to a "time_t". Coding errors in previous versions of the Rationale have also been corrected, although I believe that the when = *localtime(now); line should read when = *localtime(&now); Seismo's retirement from the mail forwarding business makes it unlikely that the time zone mailing list can be continued; this should not be a problem since volume has been nonexistent of late. If all else fails, try comp.std.c. --ado 4.12.2.3 The mktime function mktime was invented by the Committee to complete the set of time functions. With this function it becomes possible to perform portable calculations involving clock times and broken-down times. The rules on the ranges of the fields within the *timeptr record are crafted to permit useful arithmetic to be done. For instance, here is a paradigm for continuing some loop for an hour: #include <time.h> struct tm when; time_t now; time_t deadline; /* ... */ now = time(0); when = *localtime(now); when.tm_hour += 1; /* result is in the range [1,24] */ deadline = mktime(&when); printf("Loop will finish: %s\n", asctime(&when)); while (difftime(time(0), deadline) > 0) whatever(); The specification of mktime guarantees that the addition to the tm_hour field produces the correct result even when the new value of tm_hour is 24, i.e., a value outside the range ever returned by a library function in a struct tm object. One of the reasons for adding this function is to replace the capability to do such arithmetic which is lost when a programmer cannot depend on time_t being an integral multiple of some known time unit. Several readers of earlier versions of this Rationale have pointed out apparent problems in this example if now is just before a transition into or out of daylight saviings time. However, when.tm_isdst tags what sort of time was the basis of the calculation. Implementors, take heed.

2 1

Re: Timezone stuff and System V Verification Suite
by ado Aug. 7, 1987

Aug. 7, 1987

The problem at hand: the mod.sources version of the time zone compiler sets tzname[1] to "EWT" for the Eastern US; this causes problems with SVVS. I had a realization this morning. A while back I added a "-s" flag to the time zone compiler to let folks produce time zone information files with whose time(2) values were limited to numbers that have the same value whether you take them to be signed or unsigned. This was done to let time zone files be used in network setups where some machines on the network wanted to have signed time_t's and others wanted to have unsigned time_t's. Providentially, if you use the -s option you should get files that let localtime be SVVS compliant. Not only is this in line with kre's suggestion that the work should be done in the compiler rather than in localtime--it's already in place. I've attached diffs for zic.c and zic.8; apologies to those who've seen them before (but note that I've added a new note about SVVS to zic.8). --ado ------------------------------------------------------------------------------- *** 3.1/zic.c Fri Aug 7 10:42:22 1987 --- 3.6/zic.c Fri Aug 7 10:42:39 1987 *************** *** 4,10 **** #ifndef lint #ifndef NOID ! static char sccsid[] = "@(#)zic.c 3.1"; #endif /* !NOID */ #endif /* !lint */ --- 4,10 ---- #ifndef lint #ifndef NOID ! static char sccsid[] = "@(#)zic.c 3.6"; #endif /* !NOID */ #endif /* !lint */ *************** *** 355,361 **** usage() { (void) fprintf(stderr, ! "%s: usage is %s [ -v ] [ -l localtime ] [ -d directory ] [ filename ... ]\n", progname, progname); exit(1); } --- 355,361 ---- usage() { (void) fprintf(stderr, ! "%s: usage is %s [ -s ] [ -v ] [ -l localtime ] [ -d directory ] [ filename ... ]\n", progname, progname); exit(1); } *************** *** 362,367 **** --- 362,368 ---- static char * lcltime = NULL; static char * directory = NULL; + static int sflag = FALSE; main(argc, argv) int argc; *************** *** 374,380 **** umask(umask(022) | 022); #endif progname = argv[0]; ! while ((c = getopt(argc, argv, "d:l:v")) != EOF) switch (c) { default: usage(); --- 375,381 ---- umask(umask(022) | 022); #endif progname = argv[0]; ! while ((c = getopt(argc, argv, "d:l:vs")) != EOF) switch (c) { default: usage(); *************** *** 401,406 **** --- 402,410 ---- case 'v': noise = TRUE; break; + case 's': + sflag = TRUE; + break; } if (optind == argc - 1 && strcmp(argv[optind], "=") == 0) usage(); /* usage message by request */ *************** *** 465,470 **** --- 469,476 ---- tt_signed = FALSE; min_time = 0; max_time = ~(time_t) 0; + if (sflag) + max_time >>= 1; } else { tt_signed = TRUE; min_time = bit; *************** *** 471,476 **** --- 477,484 ---- max_time = bit; ++max_time; max_time = -max_time; + if (sflag) + min_time = 0; } min_year = TM_YEAR_BASE + gmtime(&min_time)->tm_year; max_year = TM_YEAR_BASE + gmtime(&max_time)->tm_year; *************** *** 687,693 **** inrule(fields, nfields) register char ** fields; { ! struct rule r; if (nfields != RULE_FIELDS) { error("wrong number of fields on Rule line"); --- 695,701 ---- inrule(fields, nfields) register char ** fields; { ! static struct rule r; if (nfields != RULE_FIELDS) { error("wrong number of fields on Rule line"); *************** *** 757,763 **** register char ** fields; { register char * cp; ! struct zone z; register int i_gmtoff, i_rule, i_format; register int i_untilyear, i_untilmonth; register int i_untilday, i_untiltime; --- 765,771 ---- register char ** fields; { register char * cp; ! static struct zone z; register int i_gmtoff, i_rule, i_format; register int i_untilyear, i_untilmonth; register int i_untilday, i_untiltime; ------------------------------------------------------------------------------- *** 3.1/zic.8 Fri Aug 7 10:41:29 1987 --- 3.4/zic.8 Fri Aug 7 10:41:32 1987 *************** *** 12,17 **** --- 12,19 ---- .B \-l .I localtime ] [ + .B \-s + ] [ .I filename \&... ] .SH DESCRIPTION *************** *** 43,48 **** --- 45,55 ---- of years representable by .IR time (2) values. + .TP + .B \-s + Limit time values stored in output files to values that are the same + whether they're taken to be signed or unsigned. + You can use this option to generate SVVS-compatible files. .sp Input lines are made up of fields. Fields are separated from one another by any number of white space characters. *************** *** 98,104 **** field. .TP .B TYPE ! Gives the type of year in which the year applies. If .B TYPE is --- 105,111 ---- field. .TP .B TYPE ! Gives the type of year in which the rule applies. If .B TYPE is *************** *** 155,161 **** field. .TP .B AT ! Gives the time of day at which the rule takes affect. Recognized forms include: .nf .in +.5i --- 162,168 ---- field. .TP .B AT ! Gives the time of day at which the rule takes effect. Recognized forms include: .nf .in +.5i *************** *** 291,294 **** /etc/zoneinfo standard directory used for created files .SH "SEE ALSO" newctime(3), tzfile(5), zdump(8) ! .. @(#)zic.8 3.1 --- 298,301 ---- /etc/zoneinfo standard directory used for created files .SH "SEE ALSO" newctime(3), tzfile(5), zdump(8) ! .. @(#)zic.8 3.4

1 0

Re: Timezone stuff and System V Verification Suite
by seismo!scubed!ncr-sd!ncrcae!sauron!wescott July 24, 1987

July 24, 1987

> I think the "right" fix is to remove the "Link" lines from the > "northamerica" file (which provides the EST5EDT aliases) and > put them in the "systemv" file, where the data should be an > accurate System V emulation. > ... > Or perhaps the "blessed" 2 zone names should be marked in the > data file to avoid surprises. That's an idea. We (NCR) are really looking for the best of both worlds. We like the flexibility of the PD ctime but need to pass the SVVS as well as provide compatibility for existing customers. The only problem has been that of tzname(). > Both the Sys V and BSD standard timezone name stuff is trash, > it all needs to be discarded completely. And the matter gets worse with SysVr3v1 (from ctime(3c) man page): [start of quote] tzset() uses the contents of the environment variable TZ to override the value of the different external variables. The syntax of TZ can be described as follows: TZ -> zone | zone signed_time | zone signed_time zone | zone signed_time zone dst zone -> letter letter letter signed_time -> sign time | time time -> hour | hour : minute | hour : minute : second dst -> signed_time | signed_time ; dst_date , dst_date | ; dst_date , dst_date dst_date -> julian | julian / time [...] For example, the setting for New Jersey in 1986 could be "EST5EDT4;117/2:00:00,299/2:00:00" [...] A southern hemisphere setting such as the Cook Islands could be "KDT9:30KST10:00;64/5:00,303/20:00" [end of quote] There is a new extern long to go along with timezone called "altzone" which holds the value of the daylight saving time "timezone" offset. The SVID, however, does not reflect this "improvement" nor apparently will the SVVS. The "Future Directions" section of the SVID does mention the fractional hours/sign component in the middle of the TZ variable. The ctime(3c) man page also describes two new routines that allow user control of the formatting of the string as well as using non-english names (under the control of the env. var. LANGUAGE). -Mike Wescott wescott(a)ncrcae.Columbia.NCR.COM

2 1

Timezone stuff and System V Verification Suite
by ado July 19, 1987

July 19, 1987

Here are excerpts from two notes from Mike Wescott (wescott(a)ncrcae.Columbia.NCR.COM) I'd appreciate comments on the changes Mike suggests. (Lines preceded by "> " in the second note are Mike quoting me.) --ado Date: Thu, 9 Jul 87 15:29:36 pdt Message-Id: <8707092229.AA00730(a)SCUBED.ARPA> Subject: Re: v08i085: Public Domain localtime We have uncovered a problem with the PD ctime with respect to the System V Verification Suite and SVID compatibility. The problem is with tzset(). Try the little program, xxx.c: #include <time.h> main() { tzset(); printf("%s\n%s\n", tzname[0], tzname[1]); } and try the following: TZ=PST8PDT ./xxx it produces: PST PWT and TZ=EST5EDT ./xxx produces EST EWT The problem is that tzname[1] become the (obsolete) EWT. The fix, I believe, would be to use the most recently used abbreviations to the value of time(0). Diff for such a fix follows and passes SVVS. -Mike Wescott wescott(a)ncrcae.Columbia.NCR.COM *** localtime.c.or Thu Jul 9 09:57:27 1987 --- localtime.c Thu Jul 9 14:31:38 1987 *************** *** 90,95 **** --- 90,96 ---- { register int i; register int fid; + time_t now; if (name == 0 && (name = TZDEFAULT) == 0) return -1; *************** *** 182,191 **** timezone = -s.ttis[0].tt_gmtoff; daylight = 0; #endif /* USG_COMPAT */ ! for (i = 1; i < s.typecnt; ++i) { register struct ttinfo * ttisp; ! ! ttisp = &s.ttis[i]; if (ttisp->tt_isdst) { tzname[1] = &s.chars[ttisp->tt_abbrind]; #ifdef USG_COMPAT --- 183,194 ---- timezone = -s.ttis[0].tt_gmtoff; daylight = 0; #endif /* USG_COMPAT */ ! now = time(0); ! for (i = 1; i < s.timecnt; ++i) { register struct ttinfo * ttisp; ! if (now < s.ats[i]) ! break; ! ttisp = &s.ttis[s.types[i]]; if (ttisp->tt_isdst) { tzname[1] = &s.chars[ttisp->tt_abbrind]; #ifdef USG_COMPAT ------------------------------------------------------------------------------- Date: Tue, 14 Jul 87 07:22:19 pdt Message-Id: <8707141422.AA27049(a)SCUBED.ARPA> Subject: Re: v08i085: Public Domain localtime > Many thanks for the note on the problem with SVVS. > Would it be all right with you if I passed the note on to other interested > folks? No problem, I sent it to you with that in mind. The code could probably be made a little more efficient but it shouldn't have to run too often. > Also--the time zone files described in the "systemv" file were designed to > provide System V compatibility. It looks as if you're in a position to check > on whether they actually do; if you could, I'd appreciate it. We ran SVVS (SysVr2 not r3 yet) and the only thing it complained about was the problem with the tzset() function. We are compiling with USG, USG_COMPAT, and STD_INSPIRED defined. I'll check the "systemv" file but it doesn't seem to be problem as far as SVVS goes.

2 1

Re: what should be in POSIX vs X3J11
by seismo!celerity!dreamit!jjw May 19, 1987

May 19, 1987

In response to mail from seismo!nbires!vianet!devine Sat May 9 10:19:15 1987 > In my last message I proposed dividing the many date/time related > functions by the rule of 'is it Unix-based or language-based?'. And I > meant by a call being in Man Section 3 as roughly the same as the latter. I don't think the division should be based on whether the function is documented in section 2 or 3 of the Unix manual, that is mostly an artifact of the Unix implementation. Instead, Something should be placed into X3J11 based on whether a standard conforming C program should be able to use the function. Something should be placed into the POSIX standard based on whether all programs running under a conforming operating system should be able to use the function. In general I think POSIX should restrict itself to low level functions (such as "time()") and should avoid specifying higher level functions (such as "ctime()"). X3J11, however, should specify both low and high level functions where that makes sense. Because of the relationship between C and Unix there will be a lot of overlap between these two and every effort should be made to keep them the same. I believe this mailing list is a reasonable place to discuss whether the various timing functions belong in X3J11, POSIX or both. It will then be up to those who are members of the standards committees to bring any results of such a discussion to the attention of the committees. I would organize "time" functions as follows: POSIX only both X3J11 only stime() time()** everything else utime()* times() time_t The reason for leaving stime() out of X3J11 is that setting a system wide time and date is not something that is done by many programs and The reason for leaving utime() out of X3J11 is that the implications of and reasons for setting file times is likely to be operating systems dependant. It is also possible to argue that if programs can read these values they should be able to set them too. * file access times (what they are and when they are set) is probably outside the purview of this mailing list. ** POSIX should probably have a "gettimeofday" call which provides better granularity than seconds. ftime() has the problem of including timezone information which I don't believe belongs in the operating system interface. -- J. J. Whelan

1 0

Colorado DST info
by seismo!nbires!vianet!devine May 19, 1987

May 19, 1987

The full state House has passed a resolution to have year-round DST for Colorado. The proposal to move the state to Central time was defeated. Basically, the argument against this second proposal boiled down to "This state has mountains. Why should we move out of the Mountain timezone?". The State Senate must approve the resolution before the US Dept of Transportation will be asked to approve it. Bob

1 0

POSIX & X3J11
by seismo!meccts.mecc.mn.org!ahby May 11, 1987

May 11, 1987

In a recent posting, Mark notes that this group has no real pull with either standards body, and that they will decide what they decide. Well, although I cannot speak for X3J11 (I am a voting member, but do not have time to attend the meetings), I can say that IEEE P1003.1 is always interested in any contributions that ANY group or individual cares to make. I have been recording most of the discussion about what items you think should be where, and I am willing to submit these to the Working Group for consideration at the June meeting in Seattle. However, time is short! (no pun intended). The June meeting is the 22nd, and that is the LAST meeting before full ballot will take place. Although anyone may comment after the standard goes to ballot, balloting objections will get first consideration after we go full use ballot. Right now, all comments are being considered. I know the reasons why the time stuff was dropped from the draft some meetings ago (I believe it was in September), and personally think that they were not good reasons. I believe that the group would really welcome a formal propsal with specific language for some time functions. I strongly recommend that some or all of the people on this mailing list get together and draft such a proposal. --- Shane P. McCarron UUCP ihnp4!meccts!ahby, ahby(a)MECC.MN.ORG MECC Technical Services ATT (612) 481-3589 (C) Copyright 1987 Shane P. McCarron Redistribution allowed only if your recipients can redistribute

1 0

Re: what should be in POSIX vs X3J11
by seismo!cbpavo.MIS.OH.ATT.COM!mark May 10, 1987

May 10, 1987

I guess I have a fundamental problem with this whole discussion. This mailing list is not, as far as I know, officially sanctioned by either X3J11 or P1003. We can pontificate all we want about what belongs in ANSI C or in POSIX, but these decisions are made by X3J11 and P1003, not by us. Other than being interested in time zone issues, do we really have any kind of sanction, delegation, or authority from either committee? Mark

1 0

Re: what should be in POSIX vs X3J11
by seismo!nbires!vianet!devine May 9, 1987

May 9, 1987

Unix and C are very tightly interwoven. Trying to separate them to form two standards is very much a Gordian (Ritchian?) Knot. In my last message I proposed dividing the many date/time related functions by the rule of 'is it Unix-based or language-based?'. And I meant by a call being in Man Section 3 as roughly the same as the latter. This message has more detail. Putting the choices into columns: Unix-based either-one language-based time() ftime() asctime() times() tzset() ctime() stime() time_t gmtime() utime() localtime() difftime() mktime() struct tm X3J11's date formatting functions Do these groupings make sense? Where should the 'either-one' functions be placed? Arguments can be made for putting them on both ends of the spectrum. Here are a couple of examples to show how various vendors with wildly different OSs dealt with the problem of what to put in their C libraries. VMS: Base OS has a 64-bit representation of time that is the number of 100 nanoseconds periods since 00:00 on November 17, 1858. HAS: ctime(), ftime() [but TZ and DST are zero], localtime() [isdst=0] time() [seconds since 00:00 UT 1/1/70], and times() [the sys fields=0] MISSING: asctime(), gmtime(), stime(), utime() DOS: Base OS keeps track of time since 1/1/80 HAS: asctime(), ctime(), difftime(), ftime(), gmtime(), localtime(), tzset(), time() [returns seconds since 00:00 UT 1/1/70], utime(), and times() MISSING: stime() DOS's functions won't give back a formatted string for any time before 1/1/80, ie, long t=12345; ctime(&t) gives back 1/1/80 even though t is referring to a day in 1970.

1 0