A brief history of (and agenda for) time zone libraries
Summary: The GNU C library is derived from Arthur David Olson's 1989 time zone code, with extensive changes. The good news is that these changes improve the modularity of Olson's code; the bad news is that it is now hard to merge some of Olson's more recent improvements into the GNU C library. Despite the bad news, Olson's fixes and improvements should be merged into the GNU C library somehow. While we're on the subject, the Linux time zone code also comes from Olson's 1989 code, but has only few minor changes; it should be updated to the current edition of Olson's code. __________ Olson's 1989 code. The original timezone code was written by Arthur David Olson. This public domain code has a good reputation. Even the Posix standard comes as close as a standard can to saying ``This is good stuff -- check it out!''; see Posix 1003.1-1990 section B.8.1.1, lines 4195-4201. The most famous publication of Olson's code was in comp.sources.unix 18, 111-117 (19 April 1989) -- see ftp.uu.net:/usenet/comp.sources.unix/volume18/localtime3/*. GNU C Library. The GNU C Library (see prep.ai.mit.edu:pub/gnu/glibc-1.05.tar.z) is derived from Olson's 1989 code. Here are its changes: 1. Rewrite the Makefile to use GNU make and GNU libc standards. 2. Split localtime.c into several smaller, more modular files. These files are: __tzset.c gmtime.c mktime.c offtime.c tzfile.c tzset.c. Completely rewrite these files in the GNU style. 3. Add some time functions not in Olson's package, e.g. adjtime. 4. Update some of the New Zealand time zone rules. Linux. Linux's code for timezones (see tsx-11.mit.edu:pub/linux/sources/sbin/timesrc.tar.Z) is also derived from Olson's 1989 code, but it has only the following minor changes: 1. Rewrite the Makefile to meet Linux standards. 2. Add a few #includes, or change them from "" form to <> form. Olson's current edition. Arthur David Olson is still maintaining both the code and the corresponding timezone database (see elsie.nci.nih.gov:pub/tz*). The differences between Olson's 1989 code and his current edition are as follows: 1. Fix some minor Makefile portability problems. 2. Fix some bugs in the C code, and avoid some possible core dumps. 3. Bring the leap second database up-to-date through June 1993. 4. Update the time zone database for Israel, Australia (which is particularly wild and woolly -- they hold voter referendums to change the timezone rules!), New Zealand, Britain, Portugal, Starke County Indiana, Saskatchewan. Remove the nonexistent US Pacific Presidential Election Time, which caused some problems late last year on some West Coast Suns. The New Zealand updates are a superset of the GNU C library's updates. 5. Add a -y option to zic to specify a command that tests for arbitrary year types; this is used in some of the new time zone rules. Proposal. Olson's changes should be propagated into both Linux and the GNU C library. Linux should update to Olson's current edition; this should be a simple merge. By the way, the Linux code should also give credit to Olson instead of silently removing his README file; this will help prevent the current situation from recurring. The GNU C library changes fall into two categories: (A) changes to Olson's localtime.c, and (B) everything else. Merging (B) is easy, almost as easy as merging the Linux timezone code, and I've already submitted patches to the GNU C library developers to do this. Merging (A) is harder, since the GNU C library has split up localtime.c into lots of little pieces. The GNU C library changes generally improve Olson's code -- they make the module interfaces cleaner, and the code is easier to follow. Unfortunately, it is now hard to tell how to propagate Olson's localtime.c changes and improvements into the GNU C library. The best person to do this propagation is the person who split up localtime.c in the first place. Another possibility is for the GNU C library to switch to a version of localtime.c that is as close as possible to Olson's latest version; that might simplify future maintenance.
Olson's 1989 code.
The original timezone code was written by Arthur David Olson. This public domain code has a good reputation. Even the Posix standard comes as close as a standard can to saying ``This is good stuff -- check it out!''; see Posix 1003.1-1990 section B.8.1.1, lines 4195-4201. The most famous publication of Olson's code was in comp.sources.unix 18, 111-117 (19 April 1989) -- see ftp.uu.net:/usenet/comp.sources.unix/volume18/localtime3/*.
I wish I had known that this code was being maintained, and where to get it from, before now.
Proposal.
Olson's changes should be propagated into both Linux and the GNU C library.
Indeed. Linux is not my concern, but I plan to update glibc today.
The GNU C library changes fall into two categories: (A) changes to Olson's localtime.c, and (B) everything else. Merging (B) is easy, almost as easy as merging the Linux timezone code, and I've already submitted patches to the GNU C library developers to do this. Merging (A) is harder, since the GNU C library has split up localtime.c into lots of little pieces. The GNU C library changes generally improve Olson's code -- they make the module interfaces cleaner, and the code is easier to follow.
These issues are very important to the GNU project, and to me.
Unfortunately, it is now hard to tell how to propagate Olson's localtime.c changes and improvements into the GNU C library. The best person to do this propagation is the person who split up localtime.c in the first place.
I am that person, and I plan to do the work.
Another possibility is for the GNU C library to switch to a version of localtime.c that is as close as possible to Olson's latest version; that might simplify future maintenance.
I really don't want to do this. It took me quite a while to untangle the spaghetti in the first place. This would certainly not simplify maintenance of the glibc code; it would return it to the obfuscated spaghetti I started with. Having the code be modular and clean simplifies maintenance for me. Using code close Olsen's code only simplifies future merging with Olsen's code.
Date: Sat, 15 May 93 18:14:01 -0400 From: roland@gnu.ai.mit.edu (Roland McGrath) I wish I had known that this code was being maintained, and where to get it from, before now. Sorry, I didn't know about the current timezone mess until yesterday, or I'd have written earlier. To keep up to date with Olson's changes, write him <ado@elsie.nci.nih.gov> and ask to be put onto the tz mailing list -- he regularly informs the list when he releases updates. The GNU C library patches that I sent yesterday, plus Brendan Kehoe's, should have merged almost all of Olson's fixes; all that remains is to merge Olson's localtime.c fixes. It took me quite a while to untangle the spaghetti in the first place. This would certainly not simplify maintenance of the glibc code; it would return it to the obfuscated spaghetti I started with. Unfortunately, that ``spaghetti'' is well-tested and has fewer bugs than the GNU C timezone code that is trying to replace it. Relatively few people have tried the GNU C timezone code, whereas hundreds of thousands of people use Olson's code every day. (E.g. it's part of SunOS.) Even now, I suspect that reverting to a simple variant of Olson's code will reduce the FSF's maintenance costs in the long run -- perhaps Olson's code is not as elegant, but it works and it's maintained well. I know that this will require giving up cleaner code that you invested a lot of time on, but if giving up that code _saves_ you time, then you should do it. Here's a wishful thought: perhaps if you contacted Olson, the two of you could try to merge the code into a single version that would satisfy the FSF's needs for easy library maintenance without having to rewrite it all from scratch. I'll volunteer to help with this (assuming my involvement would help :-).
participants (2)
-
eggert＠twinsun.com -
roland＠gnu.ai.mit.edu