Control: block -1 by 987587
Hi,
Holger Wansing <hwansing@mailbox.org> (2021-04-24):
> While trying to debug another problem yesterday, I found that the RC1 graphical
> installer fails to go forward, when selecting one of these languages:
> - Kannada
> - Marathi
> - Persian
> - Sinhala
>
> (tested with rc1-netinst and -netboot image, on an amd64 qemu machine
> with 1G RAM)
>
> Installer hangs, with only displaying a headline and nothing more,
> eating up all CPU time.
Alright, finally, I'm able to reproduce the issue with locally built
netboot-gtk mini.iso but also a repacked RC 1 netinst with just the
cdrom/gtk/initrd.img file updated! (I decided to give up spending time
trying to get debian-cd to build an image really similar to the released
one, that's a nice to have for me but not needed to debug this issue,
and followed <https://wiki.debian.org/ManipulatingISOs>.)
I've been stuck for a while but diffoscope delivered the truth!
> Interestingly, yesterday's daily does not have this problem, while RC1
> has!!! And sadly it has not been discovered, that alpha3 is also
> affected, while alpha2 is fine!
Most of my testing (no pun intended) usually comes after one of those:
make -C build/ build_netboot-gtk USE_UDEBS_FROM=testing
make -C build/ build_netboot-gtk USE_UDEBS_FROM=unstable
For a netinst, one would need to use the build_cdrom_gtk target instead,
and deploy build/dest/cdrom/gtk under install.amd64/ in the ISO.
**BUT** (and that's a big but and I cannot lie ♫)
This minimal `make` call is lacking a variable that makes a huge
difference for the languages above. Their translations are incomplete
and trigger a warning after language selection, but that can only happen
if there's a translation-status file!
Quoting `debian/rules` selectively:
# Daily builds vs. uploads to unstable:
ifeq (${SUITE},UNRELEASED)
TRANSSTATUS=
else
TRANSSTATUS=translation-status
endif
Once I figured that out, reproducing the issue is just a matter of
adding this on the `make` line:
make -C build/ build_netboot-gtk USE_UDEBS_FROM=testing TRANSSTATUS=translation-status
> Yet another problem which started to appear with alpha3, just like
> "Bug#987377: rescue-mode: when in graphical mode, locks up one prompt before the shell"
> !!!
Yes, that looks like the very same issue, as reverting to the old
libpango udeb means we no longer get stuck with those languages. For
more details about the revert, see this message:
https://bugs.debian.org/987377#91
In case things change in the archive in the meanwhile (we should get at
least a new linux upload soon-ish, one can still stick to archive
contents from when debian-installer 20210415 (for RC 1) was built:
make -C build/ build_netboot-gtk USE_UDEBS_FROM=testing TRANSSTATUS=translation-status SNAPSHOT_TS=20210415T151642Z
> The commit
> https://salsa.debian.org/installer-team/debian-installer/-/commit/95fdc73ca8cde8d7a360fd3d742fc947a045ec0f
> "Drop fontconfig tweaks introduced in version 20170828 (See: #873462)"
> comes to my mind, also included in alpha3 ... ???
Irrelevant in the end.
> The text installer seems to be not affected (well, only Persian from
> the langs above is available in text installer, and Persian works
> there).
Not surprising in hindsight since that's libpango. :)
> We already have
> "Bug#983897: installation-reports: Installation in Uyghur hangs"
> where the installer hangs at some later steps - noticed for Uyghur and
> Kannada so far...
Yes, this is likely the kind of “bad luck somewhen” similar to what
happens with rescue mode for some languages (e.g. English) but not all
(e.g. neither French nor German).
> Uyghur is however not affected by this language chooser fail bug.
This part is likely explained by the fact it's at `2 M` while affected
languages are at `2 P` in `build/translation-status`. Note that all
languages at `2 P` don't trigger this bug though (e.g. Telugu doesn't).
Full list at `2 P`:
am bn bo bs dz ka km ku lo mk ne nn si tl
Back to the “bad luck somewhen” aspect for Uyghur, it would be nice to
check with an unpatched and with a patched image. I might try and do
that at some point, but since both #987377 and this bug are being
confirmed as originating from the same issue (#987587), it looks to me
like I should move on to trying to find a solution (sticking to an old
version of the udeb isn't really a suitable one…).
> As already mentioned by Samuel, this may be another resurgence - in a
> tightened form - of
> "#929877: installation-reports: Buster installer hangs at hard disk step with
> arabic language"
> from 2019, where the real reason has not been found...
From a quick read, I would think that's a rather different issue, due to
difficulties with some specific glyphs or glyph combinations.
Cheers,
--
Cyril Brulebois (kibi@debian.org) <https://debamax.com/>
D-I release manager -- Release team member -- Freelance Consultant
Attachment:
signature.asc
Description: PGP signature