Pacman recently upgraded python from 3.8 to 3.9, which breaks the gdb package:

% gdb
gdb: error while loading shared libraries: libpython3.8.so.1.0: cannot open shared object file: No such file or directory

Over in  FS#119 , @abaumann mentioned that the libicu upgrade from 67 to 68 was part of what broke gdb. I missed the libicu upgrade when looking at gdb, but now that that one is resolved, I noticed a number of other packages that still link against the old and now nonexistant /usr/lib/libicu*.so.67 files. On my particular system, the following packages are installed, link to icu 67, and need rebuilt as well:

* harfbuzz-icu (2.7.0-1.0)
* mongo-c-driver (1.17.3-1.0)
* postfix (3.5.6-2.0)
* postgresql (12.4-1.0)
* postgresql-old-upgrade (12.5-1.0)
* samba (4.12.3-1.2)
* smbclient (4.12.3-1.2)
* syslog-ng (3.28.1-3.0)
* texlive-bin (2020.54586-4.0)
* xfsprogs (5.8.0-1.0)

… but I’m sure you have a way of determining all the affected packages?

As a workaround, in case others have the same problem, I manually extracted the old versioned .so files from the previous package to get things working again:

```
tar -C / -xf /var/cache/pacman/pkg/icu-67.1-1.0-i686.pkg.tar.zst –wildcards ‘usr/lib/lib*.so.*’ ```

- I’ll just have to remember to clean them up again later.

The packages for syslog-ng since version 3.27.1-1.0 appears to have been built without systemd support. Although their PKGBUILD runs configure with –enable-systemd, running syslog-ng –version shows Enable-Systemd: off. The official Arch x86_64 package (at version 3.28.1-1) doesn’t exhibit this problem.

I tried building the package locally on i686 using the official Arch PKGBUILD and could not reproduce the problem. My locally built version correctly finds libsystemd during configure, and gets built with systemd support. This leads me to believe the problem happens somewhere in the Archlinux32 build process.

The practical upshot of the problem is that the systemd unit file supplied with syslog-ng lists it as a Type=notify service, but when it fails to notify systemd after startup, systemd eventually kills it and restarts it repeatedly:

Jul 24 07:32:31 wolfie systemd[1]: Starting System Logger Daemon "default" instance...
Jul 24 07:32:32 wolfie syslog-ng[18977]: syslog-ng starting up; version='3.28.1'
Jul 24 07:34:01 wolfie systemd[1]: syslog-ng@default.service: start operation timed out. Terminating.
Jul 24 07:34:01 wolfie syslog-ng[18977]: syslog-ng shutting down; version='3.28.1'
Jul 24 07:34:01 wolfie systemd[1]: syslog-ng@default.service: Failed with result 'timeout'.
Jul 24 07:34:01 wolfie systemd[1]: Failed to start System Logger Daemon "default" instance.
Jul 24 07:34:02 wolfie systemd[1]: syslog-ng@default.service: Scheduled restart job, restart counter is at 1.
Jul 24 07:34:02 wolfie systemd[1]: Stopped System Logger Daemon "default" instance.
Jul 24 07:34:02 wolfie systemd[1]: Starting System Logger Daemon "default" instance...

The ArchLinux32 mirror at http[s]://mirror.math.princeton.edu/pub/archlinux32/ appears to have stopped syncing properly in October. I contacted web@math.princeton.edu on January 1 to alert them to the problem, and received the following response from a Princeton sysadmin:

Here is the configuration I am using:

archlinux32:
   remotesrc: “rsync://mirror.archlinux32.org/archlinux32/”
   localdest: “/var/www/html/pub/archlinux32”
   extraopts: “–port=22873”

Looks like at some point this stopped working from upstream:

Raw standard error:
rsync: failed to connect to mirror.archlinux32.org (85.10.198.216): Connection refused (111)
rsync: failed to connect to mirror.archlinux32.org (2a01:4f8:a0:5264::2): Network is unreachable (101)
rsync error: error in socket IO (code 10) at clientserver.c(125) [Receiver=3.1.2]

Can you recommend a better upstream target I can use? And if there are additional constraints on checkin time or frequency.

I informed him that I am not an official member of the ArchLinux32 team, but suggested that using the regular rsync port of 873 might work as mirror.archlinux32.org appears to accept connections to that port from the Internet. As of the time of this writing, it appears that the mirror is now again up to date (although it has other issues returning HTTP error 403 for some of its files), but I promised to bring his additional questions to the official team.

So, to iterate, is this the appropriate target to mirror, or should they use some other parameters?

Thank you.

In my quest to find a working mirror close to me, I’ve come across another issue with at least a couple of mirrors. Some mirrors respond with a HTTP 403 Forbidden for certain files, notably many of the Pacman database files. At the time of this writing, I’m aware of this problem happening with the princeton.edu and the clarkson.edu mirrors. I’ve been in contact with the admin of one of the mirrors, and they’ve pointed out that the permission bits are set to 0600 at the source. If the primary mirror could have their files be world readable, it would likely fix this problem automatically across more than one mirror.

Is that something we can do?

Steps to verify:

% rsync -av rsync://mirror.archlinux32.org/archlinux32/i686/extra/extra.db.tar.gz .
receiving incremental file list
extra.db.tar.gz

sent 43 bytes  received 2,044,453 bytes  371,726.55 bytes/sec
total size is 2,043,851  speedup is 1.00

% ls -l extra.db.tar.gz
-rw------- 1 user user 2043851 Jan  8 11:51 extra.db.tar.gz

After my most recent sysupgrade, postgresql refuses to start. The main postgres binary appears to be linked to multiple versions of libicu:

$  ldd /usr/bin/postgres | grep icu
	libicui18n.so.69 => /usr/lib/libicui18n.so.69 (0xb6c00000)
	libicuuc.so.69 => /usr/lib/libicuuc.so.69 (0xb6a0c000)
	libicuuc.so.70 => /usr/lib/libicuuc.so.70 (0xb6582000)
	libicudata.so.69 => /usr/lib/libicudata.so.69 (0xb4601000)
	libicudata.so.70 => /usr/lib/libicudata.so.70 

(0xb27a4000)

Installing the icu69 compatibility package made it work again, but it doesn’t look right that it should link against multiple versions.