Found in testing of the 12.7.0 live images.

Doing a live installation using calamares from either of these images
fails if no network is available.

Running /usr/sbin/bootloader-config fails to install the appropriate
grub packages and times out after 600 seconds.

Sequence of issues:

1. /etc/apt/sources.list for the image has deb.debian.org listed above
   the on-media packages. Has that changed in live-build?
2. Connman is running a DNS server on 127.0.0.1, which apt uses to
   look up deb.debian.org
3. That DNS server does not fail if no network is available. Instead,
   it hangs.

The bootloader-config script eventually times out and the installation
fails.

The workaround is to configure networking and then things work, but
offline installation is an important thing for some people.

Testing shows that there is no sign of this issue in the equivalent
11.11 live images.

Hello Steve, list,

That is an openQA-scenario that I have been working on, but did not
complete yet.
a network device (UEFI-secure boot)
With the 12.7.0 lxde image: I've started Calamares from the console with
the --debug commandline option, otherwise there would not be a log
visible (the missing log issue has been fixed in newer versions of
Calamares)
- sources-media uses 'apt-get update' and takes 1.5 minute
- bootloader-config calls 'apt-get install grub-efi-amd64' and
'install-grub'. 'apt-get install' would have taken about 13 minutes with
the timeouts, where it would have failed much quicker in 12.6.0

So what changed in 12.7 that makes the network time-out so slow?

The 11.4 image (built with live-wrapper) only has deb.debian.org in
/etc/apt/sources.list
There is no difference in content of sources.list between the 12.6.0 and
12.7.0 live images.

Interestingly, the order in sources.list appears to matter.
I've recently manually tested a sid gnome image with Calamares without
network and all lines in sources.list are attempted, until one works.
Here in the 12.7.0 lxde image, if the local repository is mentioned
first, the installation of grub-efi-amd64 happens in a few seconds, no
network access is being attempted.

Changing the order doesn't work for the Calamares installer, Calamares
creates its own variant of this file when installing.

This is a new regression. It should have failed quickly.
cannot be upgraded to 12.7 with a live image, but needs to stay at 12.6.

Proposed fixes:
* Could you add on the down page a note for the 12.7.0 live images,
mentioning this regression?
* Fix connman, to have a shorter timeout
* Fix Calamares, to benefit from a more suitable order of sources.list

Future prevention:
* How could this be prevented? Can a live image which will be 12.8.0 be
prepared and tested by openQA, before the next point-release weekend?

With kind regards,
Roland Clobus

Hello Steve, list,

That scenario is now active in openQA.
Trixie and sid appear not to be affected.

[snip]

It turns out that 12.6 was slow as well but just made it within the 600
seconds limit, so this is not a regression as I first thought.

In that case, there is no need to rush things, and we can try to fix it
for the 12.8 release.
In the mean time, openQA is prepared and will regularly report issues.
(Bookworm is built by Jenkins and then fed to openQA on the 1st of each
month)

With kind regards,
Roland Clobus

ACK.

Yes, I would expect that to be the case.

Bah.

Hmmm. I don't remember seeing this at all when we tested 12.6. I'd
have expected testers to mention a long delay here.

Sure, that's all we can do at this point. There's at least an obvious
workaround here - just connect to a network.

I didn't encounter a particularly long delay (certainly not 10 minutes)
while testing lxde on 12.9 release day, for whatever reason. Did something
change in connman? (I don't see a stable update having happened...)

    smcv