dave@lsuc.UUCP (David Sherman) (02/08/87)
(Note: this is cross-posted to several groups to increase its chances of being seen by those who will want it. Followups are directed to comp.bugs.misc.) Both the U.S. and Canada (well, Ontario and at least some of the other provinces) are adopting the new daylight savings time rules, switching to DST the first Sunday in April instead of the last, beginning in 1987. Below is a diff for /usr/src/libc/gen/ctime.c for v7 systems; if you're running v7 you should be able to feed this article right into "patch". Once you make the change to ctime.c, anything which calls ctime(3), localtime(3) or getdate(3) should be recompiled. I've gone through the source on our system, and found all the commands that call these routines. To be absolutely correct for the future, all of these should be recompiled after the ctime change is installed. Some of the requirements are pretty minor, however; things like adb and awk use ctime for diagnotic purposes only, I believe. Here are the commands I'll be changing. This is under Perkin-Elmer Edition VII Workbench. Other systems will differ; grep for getdate, localtime and cdate through your source files. /usr/src/cmd (basic v7 stuff): make adb tp xsend spool/lpd awk learn uucp/* tar refer mail ac at atrun calendar cron date ar find info iostat login ls pr prof sa who write /usr/src/ucb (stuff which came from Berkeley): ex/exrecover snake daytime tod finger script ucbmail /usr/sys/cmd (v7 and Berkeley stuff which is hardware-specific): dmesg dump dumpdir fsck restor restor512 w /usr/src/local (stuff off the net, etc.): clock hdate logacct type whopast notes diff cpio news month dumpdates last lastdown lastlog logcount lu talloc msgs Now, to the ctime.c diffs: The changed file will have #define NorthAm but not USA, which is correct for Canada. If you are in the USA, add #define USA as well. (The 1974-75 change didn't apply in Canada.) (I also put in a minor speedup posted to the net a while back; the test of year==74||year==75 is changed to year<=75&&year>=74, which it a bit faster for dates between 1975 and 1987.) Thanks to Mark Brader, {lsuc,sq}!msb, for the DST changes. *** ctime.c.old Thu Mar 6 19:58:52 1986 --- ctime.c Sat Feb 7 21:13:55 1987 *************** *** 1,4 ! #define USA /* #define Australia */ --- 1,4 ----- ! #define NorthAm /* North America */ /* #define USA #define Australia *************** *** 1,5 #define USA /* #define Australia */ --- 1,6 ----- #define NorthAm /* North America */ /* + #define USA #define Australia */ *************** *** 26,31 * * The routine calls the system to determine the local * timezone and whether Daylight Saving Time is permitted locally. #ifdef USA * (DST is then determined by the current US standard rules) * There is a table that accounts for the peculiarities --- 27,34 ----- * * The routine calls the system to determine the local * timezone and whether Daylight Saving Time is permitted locally. + #ifdef NorthAm + * (DST is then determined by the current US and Canadian standard rules) #ifdef USA * There is a table that accounts for the peculiarities * undergone by daylight time in 1974-1975. *************** *** 27,33 * The routine calls the system to determine the local * timezone and whether Daylight Saving Time is permitted locally. #ifdef USA - * (DST is then determined by the current US standard rules) * There is a table that accounts for the peculiarities * undergone by daylight time in 1974-1975. #endif --- 30,35 ----- #ifdef NorthAm * (DST is then determined by the current US and Canadian standard rules) #ifdef USA * There is a table that accounts for the peculiarities * undergone by daylight time in 1974-1975. #endif *************** *** 31,36 * There is a table that accounts for the peculiarities * undergone by daylight time in 1974-1975. #endif #ifdef Australia * (DST is then determined according to the Summer Time Act 1973 (Vic)) #endif --- 33,39 ----- * There is a table that accounts for the peculiarities * undergone by daylight time in 1974-1975. #endif + #endif #ifdef Australia * (DST is then determined according to the Summer Time Act 1973 (Vic)) #endif *************** *** 61,67 #define DS_GAP 1 /* DS time difference is 1 hour */ #endif ! #ifdef USA /* * daylight saving rules (copied from Bell original) */ --- 64,70 ----- #define DS_GAP 1 /* DS time difference is 1 hour */ #endif ! #ifdef NorthAm /* * daylight saving rules (copied from Bell original) */ *************** *** 66,72 * daylight saving rules (copied from Bell original) */ #define HEMI && /* northern hemisphere */ ! #define DS_BEGIN 119 /* last possible start for DS, (last day in Apr) */ #define DS_END 303 /* last possible end day for DS, (last day in Oct) */ #define HOUR_ON 2 /* 2am - 3am is lost when DS starts */ #define HOUR_OFF 1 /* 1am - 2am runs twice when DS ends (??) */ --- 69,77 ----- * daylight saving rules (copied from Bell original) */ #define HEMI && /* northern hemisphere */ ! #define DS_BEGIN 119 /* old last possible start for DS, (last day in Apr) */ ! #define DS_NBEGIN 96 /* new last possible start for DS, (Apr 7) */ ! /* (new rule begins in 1987) */ #define DS_END 303 /* last possible end day for DS, (last day in Oct) */ #define HOUR_ON 2 /* 2am - 3am is lost when DS starts */ #define HOUR_OFF 1 /* 1am - 2am runs twice when DS ends (??) */ *************** *** 140,145 dayno = ct->tm_yday; daylbegin = DS_BEGIN; daylend = DS_END; #ifdef USA if (ct->tm_year==74 || ct->tm_year==75) { daylbegin = daytab[ct->tm_year-74].daylb; --- 145,152 ----- dayno = ct->tm_yday; daylbegin = DS_BEGIN; daylend = DS_END; + #ifdef NorthAm + if (ct->tm_year >= 87) daylbegin = DS_NBEGIN; #ifdef USA else if (ct->tm_year<=75 && ct->tm_year>=74) { daylbegin = daytab[ct->tm_year-74].daylb; *************** *** 141,147 daylbegin = DS_BEGIN; daylend = DS_END; #ifdef USA ! if (ct->tm_year==74 || ct->tm_year==75) { daylbegin = daytab[ct->tm_year-74].daylb; daylend = daytab[ct->tm_year-74].dayle; } --- 148,154 ----- #ifdef NorthAm if (ct->tm_year >= 87) daylbegin = DS_NBEGIN; #ifdef USA ! else if (ct->tm_year<=75 && ct->tm_year>=74) { daylbegin = daytab[ct->tm_year-74].daylb; daylend = daytab[ct->tm_year-74].dayle; } *************** *** 145,150 daylbegin = daytab[ct->tm_year-74].daylb; daylend = daytab[ct->tm_year-74].dayle; } #endif daylbegin = sunday(ct, daylbegin); daylend = sunday(ct, daylend); --- 152,158 ----- daylbegin = daytab[ct->tm_year-74].daylb; daylend = daytab[ct->tm_year-74].dayle; } + #endif #endif daylbegin = sunday(ct, daylbegin); daylend = sunday(ct, daylend); David Sherman The Law Society of Upper Canada Toronto -- { seismo!mnetor cbosgd!utcs watmath decvax!utcsri ihnp4!utzoo } !lsuc!dave
edward@ukecc.UUCP (02/09/87)
In article <1565@lsuc.UUCP> dave@lsuc.UUCP (David Sherman) writes: >Both the U.S. and Canada (well, Ontario and at least some of the >other provinces) are adopting the new daylight savings time rules, >switching to DST the first Sunday in April instead of the last, >beginning in 1987. Below is a diff for /usr/src/libc/gen/ctime.c >for v7 systems; if you're running v7 you should be able to feed >this article right into "patch". I think we've all seen this coming, but shouldn't the writers of our respective flavors of UNIX be providing the fixes. I'm not knocking David's good intentions, but the DST switch can be viewed as a MAJOR software bug worthy of an OFFICIAL update. Guess it's time to call AT&T... -- Edward C. Bennett UUCP: cbosgd!ukma!ukecc!edward CSNET: edward@engr.uky.csnet "Goodnight M.A." BITNET: edward@ukma.BITNET Change the world. Make a new friend today.
jgy@hropus.UUCP (02/10/87)
> Now, to the ctime.c diffs: > > The changed file will have #define NorthAm but not USA, which is > correct for Canada. If you are in the USA, add #define USA as well. > (The 1974-75 change didn't apply in Canada.) > > (I also put in a minor speedup posted to the net a while back; > the test of year==74||year==75 is changed to year<=75&&year>=74, > which it a bit faster for dates between 1975 and 1987.) > If you really want a speed-up make the call to 'tzset' in localtime() dependent on a static variable not being initialized, this provides some time savings for things like long "ls"'s etc..
guy@gorodish.UUCP (02/10/87)
>I think we've all seen this coming, but shouldn't the writers of our >respective flavors of UNIX be providing the fixes. I'm not knocking >David's good intentions, but the DST switch can be viewed as a MAJOR >software bug worthy of an OFFICIAL update. The major software bug here is that the DST tables are compiled into programs. Arthur Olson of NIH has fixed this; he has posted various versions of a new flavor of "ctime" that reads the tables from a file, so if the rules change you just change the file. (No, stuffing it into the environment won't work: 1) the environment variable would have to be immensely long to hold a complete set of rules and 2) if you decided to get clever and call the environment variable "TZ", under S3/S5, at least, you'd break *all* the programs out there that don't expect all this goo in TZ and you'd have to recompile them all.)
dave@lsuc.UUCP (02/11/87)
In article <1151@ukecc.uky.csnet> edward@ukecc.UUCP (Edward C. Bennett) writes: >In article <1565@lsuc.UUCP> dave@lsuc.UUCP (David Sherman) writes: >> Below is a diff for /usr/src/libc/gen/ctime.c >>for v7 systems > >I think we've all seen this coming, but shouldn't the writers of our >respective flavors of UNIX be providing the fixes. I'm not knocking >David's good intentions, but the DST switch can be viewed as a MAJOR >software bug worthy of an OFFICIAL update. Absolutely. But no-one ever maintained v7 systems very officially, and I doubt anyone does now. By all means, if you have a UNIX which is "supported" (say by H-P, DEC or AT&T), lean on your suppliers to fix the problem. Incidentally, what happens if you can't fix it? I suppose you could simply set the clock (i.e., count of time in seconds) forward an hour on April 5 and back on April 26. Then, as long as your software is never fixed, all times reported will be apparently OK. We have a CT MiniFrame running CTIX (Sys V), by the way. I've assumed that for that system I may just hack TZ during the relevant three weeks... David Sherman -- { seismo!mnetor cbosgd!utcs watmath decvax!utcsri ihnp4!utzoo } !lsuc!dave
guy@gorodish.UUCP (02/13/87)
>Incidentally, what happens if you can't fix it? I suppose you >could simply set the clock (i.e., count of time in seconds) forward >an hour on April 5 and back on April 26. Then, as long as your >software is never fixed, all times reported will be apparently OK. Only for suitable values of "OK". The UNIX notion of time will be off by one hour, so all times recorded on the file system (either in inodes or in data files as UNIX time) will be incorrect.
brandx@ihlpl.UUCP (02/16/87)
In article <1569@lsuc.UUCP> dave@lsuc.UUCP (David Sherman) writes: >Incidentally, what happens if you can't fix it? I suppose you >could simply set the clock (i.e., count of time in seconds) forward >an hour on April 5 and back on April 26. Then, as long as your >software is never fixed, all times reported will be apparently OK. Be careful if you're going to set the clock to get around the problem. 1) Setting the clock forward an hour will not change the fact that the system will still view the time as Standard Time. 2) You'll have to wait until 4AM to move the clock back on the 26th. Remember, the hour between 2 and 3 AM doesn't exist on the day you're moving the clock back. (A minute before 3 AM is 1:59 AM, not 2:59 AM.) The software is smarter than you think here. I tried to set the clock back an hour after setting it forward to 3:15 on April 26. As makes perfect sense, the time changed to 1:15. Howard Weisberg
amos@instable.UUCP (02/17/87)
I always thought the whole idea of the BSD way of determining the dst was extermely redundant - in contrast to the sysV method, in which anyone can choose whether to have it or not any time, in run time, per process run; on BSD systems it's a big and ugly routine, compiled into every utility, and cannot be changed except by root, and then only globally. It seems all the effort is invested only to present the creating time of old files as it was at the time of creation - while programs (eg make) that use it couldnt care less! In Israel, where dst is erratic (the algorithm depends on the political constellation and the phases of the moon - no :-) !) I use a small program that changes the timezone by the settimeofday sys call, and is run twice a year by 'at'. -- Amos Shapir National Semiconductor (Israel) 6 Maskit st. P.O.B. 3007, Herzlia 46104, Israel (011-972) 52-522261 amos%nsta@nsc.com 34.48'E 32.10'N
guy@gorodish.UUCP (02/17/87)
>I always thought the whole idea of the BSD way of determining the dst was >extermely redundant Try "The V7 way of determining..."; putting it into the kernel antedates 4.2BSD. >in contrast to the sysV method, in which anyone can choose whether to have >it or not any time, in run time, per process run; on BSD systems it's a big >and ugly routine, compiled into every utility, Whereas on System V, it's *still* a big and ugly routine, compiled into every utility - or, more to the point, it's a big and ugly *table*, compiled into every utility, in most UNIX systems out there. At least Berkeley made an attempt to put tables for areas other than the US into those utilities (even though there were a number of errors in the tables). There *is* a solution; Arthur Olson came up with a version of "ctime" and friends that reads the DST rules out of a *file*, which is where they should have been in the first place. If the rules change, you just edit a file and compile it into a table; any programs run after that point will have the correct rules. The file to be read is selected by the TZ environment variable, so a user *can* override the default setting if they want. A program can also *ignore* the value of TZ; this is required by programs such as "uucico" that must know the time zone the phone lines are in, *not* the time zone the user is in. You can even arrange to give the files names like EST5EDT, so that you can set TZ to values that will work with binaries built on systems that don't have this new stuff; solutions I've seen proposed that encode the rules in the TZ environment string don't have this property, so they break binary compatibility (not to mention breaking programs that expect to convert dates not in the current year - remember, "last year" is, at certain times, less than one day ago!).
dave@lsuc.UUCP (02/19/87)
In article <13253@sun.uucp> guy@sun.UUCP (Guy Harris) writes: >>Incidentally, what happens if you can't fix it? I suppose you >>could simply set the clock (i.e., count of time in seconds) forward >>an hour on April 5 and back on April 26. Then, as long as your >>software is never fixed, all times reported will be apparently OK. > >Only for suitable values of "OK". The UNIX notion of time will be >off by one hour, so all times recorded on the file system (either in >inodes or in data files as UNIX time) will be incorrect. Note that I said, "as long as your software is NEVER fixed". The internal time may be incorrect, but it would forever (when referring to times between April 5 and April 27/87, for example) get it "right" when converting it through localtime(). David Sherman The Law Society of Upper Canada -- { seismo!mnetor cbosgd!utgpu watmath decvax!utcsri ihnp4!utzoo } !lsuc!dave
guy@gorodish.UUCP (02/20/87)
>The internal time may be incorrect, but it would forever (when referring >to times between April 5 and April 27/87, for example) get it "right" >when converting it through localtime(). This presumes that no software on the system would be bothered by the internal time being incorrect; this may be true on some systems, but I wouldn't make that assumption without studying all the software on the system (including locally-developed software).