- tz - lists.iana.org

Cond jump or move depends on uninit'd value error on 64-bit Linux
by Ravindra Oct. 1, 2010

Oct. 1, 2010

I'm seeing two problems in localtime.c: 1. When I compile localtime.c using gcc 4.1.2 for 32-bit, I get following warning: $ gcc -m32 -g -c -o localtime.o localtime.c localtime.c: In function âdiffer_by_repeatâ: localtime.c:326: warning: comparison is always false due to limited range of data type 2. valgrind reports following error for some timezones (only when localtime.c is compiled as 64-bit binary, it does not happen when compiled as 32-bit binary): ==26671== Conditional jump or move depends on uninitialised value(s) ==26671== at 0x401CDA: typesequiv (localtime.c:588) ==26671== by 0x401B85: tzload (localtime.c:562) ==26671== by 0x405553: tz_validate (localtime.c:2017) ==26671== by 0x40560D: main (localtime.c:2040) One can reproduce it using following valgrind command and patch: valgrind -v --tool=memcheck --leak-check=full --leak-resolution=med --show- reachable=yes --error-limit=no --workaround-gcc296-bugs=yes --run-libc- freeres=no --log-file=valgrind-tzvalid ./tzvalid /<path/to/my/zoneinfo/EST5EDT> NOTE: I can consistently reproduce it for EST5EDT timezone and I don't get this error for GMT. Patch for localtime.c for reproducing above error: --- org/localtime.c 2010-08-16 16:27:05.000000000 +0530 +++ localtime.c 2010-10-01 17:34:45.999089000 +0530 @@ -2004,6 +2004,47 @@ #endif /* defined CMUCS */ +static int +tz_validate(const char * name) +{ + struct state state = {0}; + time_t now; + struct tm tm; + + time(&now); + tzload(gmt, &state, TRUE); + if (0 == tzload(name, &state, TRUE)) { + localsub(&now, 0L, &tm); + return TRUE; + } + if (0 == tzparse(name, &state, FALSE)) { + localsub(&now, 0L, &tm); + return TRUE; + } + return FALSE; +} + +/* + * Compile it simply as "$gcc -m64 -g -o tzvalid localtime.c" + */ +int main(int argc, char ** argv) { + + if (argc != 2) { + printf("Usage: ./tzvalid <timezone_file_path>\n"); + return 2; + } + + char * tz = argv[1]; + + if (tz_validate(tz)) { + printf("timezone %s is valid\n", tz); + return 0; + } else { + printf("timezone %s is not valid\n", tz); + return 1; + } +} + /* ** XXX--is the below the right way to conditionalize?? */ Thanks in advance, Ravindra

1 0

Re: zic and zoneinfo.zip
by Paul Eggert Oct. 1, 2010

Oct. 1, 2010

On 09/28/2010 11:15 PM, Ravindra wrote: > That's my appln code. It validates a timezone by loading it. It calls > tzload() like below: Can you please send a complete, self-contained test case illustrating the bug? Snippets of code from an otherwise unpublished help are nice, but they typically aren't enough to reproduce the problem.

2 2

environment independent time functions...
by christos＠zoulas.com Sept. 30, 2010

Sept. 30, 2010

Hello, What I want to achieve is to be able to use the time functions in a multi-threaded program that can handle multiple timezones at the same time. Right now, this is impossible, because the notion of timezone is expressed through the environment. I was careful to re-use the existing code and not change existing functionality so the following routines are just an add-on. I modified tzcode 2010l to add the following functions (as defined in timeextra.h). First I expose an opaque timezone type: typedef struct __state *timezone_t; which is no other than the internal state. Then I provide a way to allocate the handle on per timezone name and to free it. const timezone_t tzalloc(const char *); void tzfree(const timezone_t); Note that the handle is const, because I might want to enhance this in the future to use reference counting instead of allocating a new one per invocation. Then I provide a way to return the two (dst and nondst) strings (like tzname[0], and tzname[1]) with: const char * tzgetname(const timezone_t, int); The following functions are also provided, which are like the _r counterparts only they have an extra timezone argument: struct tm * localtime_z(const time_t *, struct tm *, const timezone_t); char * ctime_z(const time_t *, char *, const timezone_t); time_t timelocal_z(struct tm *, const timezone_t); time_t mktime_z(struct tm *, const timezone_t); time_t time2posix_z(time_t, const timezone_t); time_t posix2time_z(time_t, const timezone_t); The changes to the source are minimal (the diffs are slightly larger because I made things use ansi prototypes and removed 'register '. I don't care about the names or the signatures much, just something I came up as a proof of concept. Here's example code: #include "timeextra.h" #include <stdlib.h> #include <stdio.h> int main(int argc, char *argv[]) { size_t i; char buf[60]; time_t t = time(NULL); timezone_t *tz = malloc(argc * sizeof(*tz)); for (i = 1; i < argc; i++) { tz[i] = tzalloc(argv[i]); if (tz[i] == NULL) err(1, "Cannot find timezone %s", argv[i]); } for (i = 1; i < argc; i++) { printf("[%s] [%s] %s", tzgetname(tz[i], 0), tzgetname(tz[i], 1), ctime_z(&t, buf, tz[i])); tzfree(tz[i]); } return 0; } Has anyone done this work or similar? Do you think others might want something like this? Are you interested in taking a look at the changes/incorporating them? Do you have any other suggestions? Best Regards, christos

1 0

New TZ data for Fiji
by Franck Martin Sept. 30, 2010

Sept. 30, 2010

http://www.fiji.gov.fj/index.php?option=com_content&view=article&id=2438:da… Franck Martin http://www.avonsys.com/ http://www.facebook.com/Avonsys twitter: FranckMartin Avonsys Check your domain reputation: http://gurl.im/b69d4o

1 0

Re: Bug in zic.c?
by Arthur David Olson Sept. 29, 2010

Sept. 29, 2010

That is indeed a bug--in particular, in the DOWLEQ (<=) code. The bug wasn't triggered up to now since all the distributed rules use DOWGEQ (>=). Below is a proposed fix (and slight simplification). Note that the checks are needed to ensure that rules specified in the zic input are representable by POSIX environment variables. --ado ------- zic.c ------- *** /tmp/geta23934 Wed Sep 29 08:45:01 2010 --- /tmp/getb23934 Wed Sep 29 08:45:01 2010 *************** *** 3,9 **** ** 2006-07-17 by Arthur David Olson. */ ! static char elsieid[] = "@(#)zic.c 8.22"; #include "private.h" #include "locale.h" --- 3,9 ---- ** 2006-07-17 by Arthur David Olson. */ ! static char elsieid[] = "@(#)zic.c 8.23"; #include "private.h" #include "locale.h" *************** *** 1881,1896 **** register int week; if (rp->r_dycode == DC_DOWGEQ) { ! week = 1 + rp->r_dayofmonth / DAYSPERWEEK; ! if ((week - 1) * DAYSPERWEEK + 1 != rp->r_dayofmonth) return -1; } else if (rp->r_dycode == DC_DOWLEQ) { if (rp->r_dayofmonth == len_months[1][rp->r_month]) week = 5; else { ! week = 1 + rp->r_dayofmonth / DAYSPERWEEK; ! if (week * DAYSPERWEEK - 1 != rp->r_dayofmonth) return -1; } } else return -1; /* "cannot happen" */ (void) sprintf(result, "M%d.%d.%d", --- 1881,1896 ---- register int week; if (rp->r_dycode == DC_DOWGEQ) { ! if ((rp->r_dayofmonth % DAYSPERWEEK) != 1) return -1; + week = 1 + rp->r_dayofmonth / DAYSPERWEEK; } else if (rp->r_dycode == DC_DOWLEQ) { if (rp->r_dayofmonth == len_months[1][rp->r_month]) week = 5; else { ! if ((rp->r_dayofmonth % DAYSPERWEEK) != 0) return -1; + week = rp->r_dayofmonth / DAYSPERWEEK; } } else return -1; /* "cannot happen" */ (void) sprintf(result, "M%d.%d.%d",

1 0

FW: Bug in zic.c?
by Olson, Arthur David (NIH/NCI) [E] Sept. 29, 2010

Sept. 29, 2010

I'm forwarding this message from Lei Liu, who is not on the time zone mailing list. Those of you who are on the list, please direct replies appropriately. --ado -----Original Message----- From: Liu, Lei [mailto:lei.liu2@windriver.com] Sent: Wednesday, September 29, 2010 4:17 To: tz(a)lecserver.nci.nih.gov Subject: Bug in zic.c? Hi, I have a question about the code in zic.c. In function stringrule I get: week = 1 + rp->r_dayofmonth / DAYSPERWEEK; if ((week - 1) * DAYSPERWEEK + 1 != rp->r_dayofmonth) return -1; //... week = 1 + rp->r_dayofmonth / DAYSPERWEEK; if (week * DAYSPERWEEK - 1 != rp->r_dayofmonth) return -1; What do these two if check here? The first one passes only when rp->r_dayofmonth mod 7 equals to 1. And the second one passes only when rp->r_dayofmonth mode 7 equals to 6. But IMO rp->r_dayofmonth can be arbitrary valid day number of the month. This question came up with a problem of compiling following input file: #RULE NAME FROM TO TYPE IN ON AT SAVE LTRS Rule localtime minimum maximum - Mar Sun<=14 2:00 1:0 PDT Rule localtime minimum maximum - Nov Sun<=7 2:00 0:0 UTC # ZONE NAME GMTOFF RULES/SAVE FORMAT Zone localtime 0:0 localtime %s zic prints out: "fault.txt", line 6: too many transitions?! (rule from "fault.txt", line 2) After I removed the two if check, this file can be successfully processed. Is this problem caused by a bug in zic.c or the invalid input data. Thanks. Lei

1 0

Bug in zic.c?
by Liu, Lei Sept. 29, 2010

Sept. 29, 2010

Hi, I have a question about the code in zic.c. In function stringrule I get: week = 1 + rp->r_dayofmonth / DAYSPERWEEK; if ((week - 1) * DAYSPERWEEK + 1 != rp->r_dayofmonth) return -1; //... week = 1 + rp->r_dayofmonth / DAYSPERWEEK; if (week * DAYSPERWEEK - 1 != rp->r_dayofmonth) return -1; What do these two if check here? The first one passes only when rp->r_dayofmonth mod 7 equals to 1. And the second one passes only when rp->r_dayofmonth mode 7 equals to 6. But IMO rp->r_dayofmonth can be arbitrary valid day number of the month. This question came up with a problem of compiling following input file: #RULE NAME FROM TO TYPE IN ON AT SAVE LTRS Rule localtime minimum maximum - Mar Sun<=14 2:00 1:0 PDT Rule localtime minimum maximum - Nov Sun<=7 2:00 0:0 UTC # ZONE NAME GMTOFF RULES/SAVE FORMAT Zone localtime 0:0 localtime %s zic prints out: "fault.txt", line 6: too many transitions?! (rule from "fault.txt", line 2) After I removed the two if check, this file can be successfully processed. Is this problem caused by a bug in zic.c or the invalid input data. Thanks. Lei

1 0

FW: FW: zic and zoneinfo.zip
by Olson, Arthur David (NIH/NCI) [E] Sept. 28, 2010

Sept. 28, 2010

I'm forwarding this message from Ravindra, who is not on the time zone mailing list. Those of you who are on the list, please direct replies appropriately. --ado -----Original Message----- From: Ravindra [mailto:rsahlot@yahoo.com] Sent: Monday, September 27, 2010 8:09 To: 'tz(a)elsie.nci.nih.gov'; Olson, Arthur David (NIH/NCI) [E] Subject: Re: FW: zic and zoneinfo.zip More testing on this revealed that it happens only when I use tzdata from version "2010l" on a 64-bit platform (localtime.c compiled as 64-bit code). valgrind does not report the memory error if localtime.c is compiled as 32-bit code or if I use tzdata2007k with tzcode2010l. Looks like, tzcode2010l and tzdata2010l do not get well together on 64-bit? Thanks, Ravindra --- On Sat, 9/25/10, Olson, Arthur David (NIH/NCI) [E] <olsona(a)dc37a.nci.nih.gov> wrote: > From: Olson, Arthur David (NIH/NCI) [E] <olsona(a)dc37a.nci.nih.gov> > Subject: FW: zic and zoneinfo.zip > To: "'tz(a)elsie.nci.nih.gov'" <tz(a)elsie.nci.nih.gov> > Cc: "'rsahlot(a)yahoo.com'" <rsahlot(a)yahoo.com> > Date: Saturday, September 25, 2010, 6:11 PM > I'm forwarding this message from > Ravindra, who is not on the time zone mailing list. Those of > you who are on the list, please direct replies > appropriately. > > (Note that the line numbers below are indeed different from > those in the 2010l on elsie.nci.nih.gov, and that there's no > "tz.c" file with a "tz_validate" function in 2010l on > elsie.nci.nih.gov.) > > > --ado > > -----Original Message----- > From: Ravindra [mailto:rsahlot@yahoo.com] > > Sent: Saturday, September 25, 2010 8:01 > To: Olson, Arthur David (NIH/NCI) [E] > Subject: RE: zic and zoneinfo.zip > > Thanks Arthur! > > Is there a place where I get pre-built zic binary for > Linux? > > I got the doubt because I could build 32-bit zic as well as > 64-bit zic from source code, and I was not sure whether to > build zic as a 32-bit binary or 64-bit binary. What I'm > getting from your response is I can use either version of > zic and use the generated timezone data files on either > platform. > > BTW, I'm looking at various possibilities because I'm using > the code and data from version "2010l" and I'm hitting > following memory error in valgrind log: > > ==5207== Thread 27: > ==5207== Conditional jump or move depends on uninitialised > value(s) > ==5207== at 0x1B9F988: typesequiv > (localtime.c:522) > ==5207== by 0x1B9F830: tzload > (localtime.c:493) > ==5207== by 0x1BA261B: tz_validate > (tz.c:1717) > > Below call is hitting above memory error (line numbers may > not match with original code because we have integrated it > in our code): > > 1712 struct > state state = {0}; > 1713 time_t > > now; > 1714 struct tm > tm; > 1715 > 1716 time(&now); > *** 1717 tzload(gmt, > &state, TRUE, 0); > ... > > tzload code: > ... > 486 if > (sp->timecnt > 1) { > 487 > for (i = 1; i < sp->timecnt; ++i) > 488 > if (typesequiv(sp, > sp->types[i], sp->types[0]) && > 489 > > differ_by_repeat(sp->ats[i], > sp->ats[0])) { > 490 > > sp->goback = > TRUE; > 491 > > break; > 492 > > } > **** 493 > for (i = sp->timecnt - 2; i >= 0; > --i) > 494 > if (typesequiv(sp, > sp->types[sp->timecnt - 1], > 495 > > sp->types[i]) && > 496 > > differ_by_repeat(sp->ats[sp->timecnt > - 1], > 497 > > sp->ats[i])) { > 498 > > sp->goahead = > TRUE; > 499 > > break; > 500 > } > 501 } > ... > 511 static int > 512 typesequiv(const struct state * const sp, const int a, > const int b) > 513 { > 514 register > int result; > 515 > 516 if (sp == NULL > || > 517 > a < 0 || a >= sp->typecnt || > 518 > b < 0 || b >= sp->typecnt) > 519 > result = FALSE; > 520 else { > 521 > register const struct ttinfo * ap = > &sp->ttis[a]; > *** 522 > register const struct ttinfo * bp = > &sp->ttis[b]; > 523 > result = ap->tt_gmtoff == > bp->tt_gmtoff && > 524 > ap->tt_isdst == > bp->tt_isdst && > 525 > ap->tt_ttisstd == > bp->tt_ttisstd && > 526 > ap->tt_ttisgmt == > bp->tt_ttisgmt && > 527 > > strcmp(&sp->chars[ap->tt_abbrind], > 528 > > &sp->chars[bp->tt_abbrind]) == > 0; > 529 } > 530 return result; > 531 } > > It will be great if you are aware of fix for this or if you > could help me how to proceed on this. I can explore this > further but I'm new to this code. > > Thanks in advance, > Ravindra > > --- On Fri, 9/24/10, Olson, Arthur David (NIH/NCI) [E] > <olsona(a)dc37a.nci.nih.gov> > wrote: > > > From: Olson, Arthur David (NIH/NCI) [E] <olsona(a)dc37a.nci.nih.gov> > > Subject: RE: zic and zoneinfo.zip > > To: "Ravindra" <rsahlot(a)yahoo.com> > > Date: Friday, September 24, 2010, 6:40 PM > > The current version of zic generates, > > for each time zone, a binary file that includes both > 32-bit > > and 64-bit versions of the zone's data; the binary > files > > should be usable on both 32- and 64-bit systems. > > > > --ado > > > >

3 4

FW: zic and zoneinfo.zip
by Olson, Arthur David (NIH/NCI) [E] Sept. 28, 2010

Sept. 28, 2010

I'm forwarding this message from Ravindra, who is not on the time zone mailing list. Those of you who are on the list, please direct replies appropriately. (Note that the line numbers below are indeed different from those in the 2010l on elsie.nci.nih.gov, and that there's no "tz.c" file with a "tz_validate" function in 2010l on elsie.nci.nih.gov.) --ado -----Original Message----- From: Ravindra [mailto:rsahlot@yahoo.com] Sent: Saturday, September 25, 2010 8:01 To: Olson, Arthur David (NIH/NCI) [E] Subject: RE: zic and zoneinfo.zip Thanks Arthur! Is there a place where I get pre-built zic binary for Linux? I got the doubt because I could build 32-bit zic as well as 64-bit zic from source code, and I was not sure whether to build zic as a 32-bit binary or 64-bit binary. What I'm getting from your response is I can use either version of zic and use the generated timezone data files on either platform. BTW, I'm looking at various possibilities because I'm using the code and data from version "2010l" and I'm hitting following memory error in valgrind log: ==5207== Thread 27: ==5207== Conditional jump or move depends on uninitialised value(s) ==5207== at 0x1B9F988: typesequiv (localtime.c:522) ==5207== by 0x1B9F830: tzload (localtime.c:493) ==5207== by 0x1BA261B: tz_validate (tz.c:1717) Below call is hitting above memory error (line numbers may not match with original code because we have integrated it in our code): 1712 struct state state = {0}; 1713 time_t now; 1714 struct tm tm; 1715 1716 time(&now); *** 1717 tzload(gmt, &state, TRUE, 0); ... tzload code: ... 486 if (sp->timecnt > 1) { 487 for (i = 1; i < sp->timecnt; ++i) 488 if (typesequiv(sp, sp->types[i], sp->types[0]) && 489 differ_by_repeat(sp->ats[i], sp->ats[0])) { 490 sp->goback = TRUE; 491 break; 492 } **** 493 for (i = sp->timecnt - 2; i >= 0; --i) 494 if (typesequiv(sp, sp->types[sp->timecnt - 1], 495 sp->types[i]) && 496 differ_by_repeat(sp->ats[sp->timecnt - 1], 497 sp->ats[i])) { 498 sp->goahead = TRUE; 499 break; 500 } 501 } ... 511 static int 512 typesequiv(const struct state * const sp, const int a, const int b) 513 { 514 register int result; 515 516 if (sp == NULL || 517 a < 0 || a >= sp->typecnt || 518 b < 0 || b >= sp->typecnt) 519 result = FALSE; 520 else { 521 register const struct ttinfo * ap = &sp->ttis[a]; *** 522 register const struct ttinfo * bp = &sp->ttis[b]; 523 result = ap->tt_gmtoff == bp->tt_gmtoff && 524 ap->tt_isdst == bp->tt_isdst && 525 ap->tt_ttisstd == bp->tt_ttisstd && 526 ap->tt_ttisgmt == bp->tt_ttisgmt && 527 strcmp(&sp->chars[ap->tt_abbrind], 528 &sp->chars[bp->tt_abbrind]) == 0; 529 } 530 return result; 531 } It will be great if you are aware of fix for this or if you could help me how to proceed on this. I can explore this further but I'm new to this code. Thanks in advance, Ravindra --- On Fri, 9/24/10, Olson, Arthur David (NIH/NCI) [E] <olsona(a)dc37a.nci.nih.gov> wrote: > From: Olson, Arthur David (NIH/NCI) [E] <olsona(a)dc37a.nci.nih.gov> > Subject: RE: zic and zoneinfo.zip > To: "Ravindra" <rsahlot(a)yahoo.com> > Date: Friday, September 24, 2010, 6:40 PM > The current version of zic generates, > for each time zone, a binary file that includes both 32-bit > and 64-bit versions of the zone's data; the binary files > should be usable on both 32- and 64-bit systems. > > --ado

5 8

FW: FW: Time Zone in Paris
by Olson, Arthur David (NIH/NCI) [E] Sept. 25, 2010

Sept. 25, 2010

I'm forwarding this message from Katia, who is not the time zone mailing list. Those of you who are on the list, please direct replies appropriately. --ado -----Original Message----- From: Katia [mailto:katia@astrologiainlinea.it] Sent: Saturday, September 25, 2010 6:44 To: 'Paul Eggert' Cc: tz(a)lecserver.nci.nih.gov Subject: R: FW: Time Zone in Paris Thank you for your reply and for having put my question in the mailing lists to which I will enjoy soon. The problem is that according to the downloaded version of the database on the Google site for updates http://code.google.com/p/tzdata/ in 1971 the time zone in Paris wasn't one hour ahead of UTC. Here is part of Paris file in XML version <from>1945-04-02T01:00:00Z</from> <to>1945-09-16T01:00:00Z</to> <dst>true</dst> <offset>7200</offset> <name>WEMT</name> </period> - <period> <from>1945-09-16T01:00:00Z</from> <to>1976-03-28T01:00:00Z</to> <dst>false</dst> <offset>0</offset> <name>WET</name> </period> - <period> Is that file correct ? Katia -----Messaggio originale----- Da: Paul Eggert [mailto:eggert@cs.ucla.edu] Inviato: giovedì 23 settembre 2010 06:37 A: katia(a)astrologiainlinea.it Cc: tz(a)elsie.nci.nih.gov Oggetto: Re: FW: Time Zone in Paris Katia wrote: > With regard to the time zone of Paris from 1945 to 1976 it results a > strange data that conflicts over other resources credited as Shank, > according to your database in 1971 the Time zone Paris was 0. Sorry, I don't see that problem. According to the tz database, the time zone in Paris was one hour ahead of UTC during that period. This agrees with Shanks & Pottenger 2003 (ISBN 0-035127-88-7); I just now checked my printed copy. It also agrees with the Ephemerides Astronomiques for 1998 from Bureau des Longitudes, according to email that we received on 1997-12-12 from Denis Excoffier. These two sources disagree only about exactly when this long period of UTC+1 ended. Shanks & Pottenger say that it ended on 1976-03-28 at 02:00 local time, whereas Excoffier wrote that it ended at 00:00 UTC, i.e., at 01:00 local time. Currently the tz database agrees with Excoffier, but we always welcome any more-authoritative sources, such as reliable newspapers published at the time.

2 1