On 2026-10-31 01:57, Paul Eggert via tz wrote:
On 2026-10-30 17:01, Brian Inglis via tz wrote:
Many distros have build and check policies that they *MUST* work without any network access

To do that, use "make CURL=: check" as documented in the Makefile. That should suffice when building from a release tarball. Although building from Git typically does need network access to fetch the Git repository and leap- seconds.list, after that CURL=":" should suffice there, too.

Thanks Paul,

Never noticed those note or thought to look there
That should make life better running package builds in the CI, eliminate the need for `make -i check` to ignore those failures, and enable auto-deploy to ship the updates!

make DATAFORM=rearguard EXPIRES_LINE=1 PACKRATDATA=backzone PACKRATLIST= REDO=posix_right VERSION_DEPS= -ik check

That command uses make's -i option, which does not work with TZDB's Makefile. Generally speaking, "-i" is not a good idea unless a Makefile is specifically written with "-i" in mind.

Omitting "-i" should fix the problem with "rm -fr" that you mentioned, because when the previous command in that recipe fails, "make" will not go on to execute "rm -fr".

Hopefully, but it would be better never to clear the directory contents, except at the start or end of the whole build, rather than within that stage.

or depending on some change of the upstream sources, or moved into the main build process, or to a maintainer rule, otherwise everything is redone if there is any issue during or after that stage

Tests are redone if dependencies change, and by default all tests depend on pretty much every source file. You can work around some of this by using "make VERSION=unknown VERSION_DEPS="; although this may cause the version info to be wrong, it should avoid redoing some tests if you change just a few files and don't care about version mismatches.

I am talking about building packages for distros, where we are building straight from a previously cached downloaded tarball, and not changing anything. Most of us know if we do patch anything distributed, it's a good idea to do a maintainer build, to run those extra dependency checks.

taking 15 minutes just for the zdumps.

It shouldn't take that long even on a slow CPU. Are you compiling without optimization? If so, try using -O2 or something like that. If that doesn't help, please use profiling to see why zdump is so slow, as it shouldn't be. Even the oldest, slowest machine I have ready access to (a circa 2001 750 MHz UltraSPARC III and with slow access to files via NFS) is several times faster than what you describe.

I have an oldish AMD APU running Cygwin builds on rusty drives.
It looks like the package or distro options use -O2, although I thought I saw
-O3 somewhere?
Even in the GH CI, that part of that stage takes 5 minutes: I would try to look at building the zone list and `zdump` all once thru `sed` to cleanup directly into the consolidated output file.

PS. In release 2026f, EXPIRES_LINE=1 was deprecated, and we plan to remove it in a future release because leap second tables are now frozen indefinitely.

Understood, but Cygwin is currently using that file to provide TIME_TAI, and I am also looking to drop posixrules, so I need to see those impacts on Cygwin POSIX time functions, not to mention my own use on Classic NTP/ntpd/utils including Meinberg's for MS Windows, and if that is being changed to handle lack of an expiry date.

--
Take care. Thanks, Brian Inglis              Calgary, Alberta, Canada

La perfection est atteinte                   Perfection is achieved
non pas lorsqu'il n'y a plus rien à ajouter  not when there is no more to add
mais lorsqu'il n'y a plus rien à retrancher  but when there is no more to cut
                                -- Antoine de Saint-Exupéry

Reply via email to