[install] Pacman segfaults with a single server in the mirrorlist

[LukeShortCloud]

Pacman will always report a segmentation fault when trying to download database files when the XferCommand is configured to use wget.

Installing arch...
+ pacstrap -i /winesapos base base-devel wget --noconfirm
==> Creating install root at /winesapos
==> Installing packages to /winesapos
:: Synchronizing package databases...
--2024-03-30 05:12:17--  http://repo.ialab.dsu.edu/archlinux/core/os/x86_64/core.db
Resolving repo.ialab.dsu.edu (repo.ialab.dsu.edu)... 138.247.115.248
Connecting to repo.ialab.dsu.edu (repo.ialab.dsu.edu)|138.247.115.248|:80... connected.
HTTP request sent, awaiting response... 200 OK
Length: 131629 (129K) [application/octet-stream]
Saving to: ‘/winesapos/var/lib/pacman/sync/core.db.part’

     0K .......... .......... .......... .......... .......... 38%  539K 0s
    50K .......... .......... .......... .......... .......... 77% 1.05M 0s
   100K .......... .......... ........                        100% 62.9M=0.1s

2024-03-30 05:12:18 (921 KB/s) - ‘/winesapos/var/lib/pacman/sync/core.db.part’ saved [131629/131629]


error: segmentation fault
Please submit a full bug report with --debug if appropriate.
25h==> ERROR: Failed to install packages to new root

[LukeShortCloud]

It seems like the issue is with Pacman itself. There is a known bug that Pacman will result in a segfault when (1) there is only one server in a mirrorlist and (2) GPG key checking is enabled.

This was fixed 1 week ago. The latest release of Pacman is 6.1.0 from 3 weeks ago. We will have to wait for a new release of Pacman.

https://bbs.archlinux.org/viewtopic.php?id=293911
https://gitlab.archlinux.org/pacman/pacman/-/issues/108

[LukeShortCloud]

This issue only appears while using the new WINESAPOS_SINGLE_MIRROR=true environment variable.

[LukeShortCloud]

A bugfix Pacman 6.1.1 release with this fix is being planned.

https://gitlab.archlinux.org/pacman/pacman/-/milestones/2#tab-issues

[LukeShortCloud]

I can confirm that (1) not using a single mirror and (2) commenting out the sed commands to enable the XferCommand allows local builds to work again with pacman 6.1.0-3.

[GuestSneezeOSDev]

if pacman is not working we can use this command

pacman -Syu haveged
systemctl start haveged
systemctl enable haveged

rm -fr /etc/pacman.d/gnupg
pacman-key --init
pacman-key --populate archlinux

[JerryWJackson]

I can verify that adding a duplicate Server line immediately beneath the existing one for only the winesapos repo fixes this segfault and allows package installation and db sync to complete as expected.

So it is the same bug from upstream.

[LukeShortCloud]

@GuestSneezeOSDev Thanks for the suggestion! I gave it a try but the service checks to see if the Linux kernel version is 5.6 or older before starting. Newer Linux kernels appear to have patches that make the random entropy on-par with what haveged provides. As part of our builds, we already do the second part of importing the latest Arch Linux GPG keys.

@JerryWJackson Thanks for confirming! I'm holding out for Pacman 6.1.1 to be released instead of relying on the mentioned workaround of adding an extra Server = line. Hopefully that will be soon.

Oddly, our GitHub Actions pipelines used to fail with this issue on the Chaotic AUR repository but that is no longer the case. I only notice it on local builds. We use container builds (with cache removed) so everything is supposed to be reproducible but apparently it is not.