On 30/05/2025 at 13:01, Colin Watson wrote:
On Fri, May 30, 2025 at 12:11:44PM +0200, Pascal Hambourg wrote:
On 17/04/2025 at 21:51, Colin Watson wrote:
libparted wrongly detects ext4 without journal as ext2 instead of ext4.
(...)
This bug affects the debian installer partitioning tool, partman, which
relies on libparted to detect partition filesystem types. partman
accepts to use such ext4 partition (without formatting it) only as
ext2, then tries and fails to mount it as ext2 because of incompatible
features.
Note: this is not a theoretical bug, it was reported by an affected D-
I user in #debian-boot IRC channel.
If it's a non-trivial bug, why did you file it as Severity: minor?
https://www.debian.org/Bugs/Developer#severities - "a problem which
doesn't affect the package's usefulness, and is presumably trivial to
fix". It's usually for things like typos.
I am no expert in bug severity assessment, and it seems to me that the
bug matches the above definition quite well: it does not affect parted's
overall usefulness, affects only uncommon use cases (d-i + ext4 without
journal) and is rather trivial to fix (reorder filesystem feature checks).
Please can you summarize the real-world situation in which this came up?
To me this seems like a very niche sort of situation, but if you could
actually explain it then that might help.
If you have IRC logs, the discussion took place on April 17th. Quoting
the OP "Salv"'s report:
"I need to use existing partition and filesystem because I have some fs
format switches to keep. My root is a ext4 fs formatted with "-m 0 -O
^has_journal,^resize_inode,^quota -E stride=64,stripe-width=64 -T news",
but latest testing debian installer (weekly buid) recognizes that
partition as ext2. Try to workaround: if I try to select ext2 with
no-format option, in the next step it will fail to mount and fail to
continue install Debian. If I select ext4 fs, it will just format
without keeping data."
Upstream reviewed the patch and intends to merge it soon [1] but I
guess it is already too late for trixie release.
[1] <https://alioth-lists.debian.net/pipermail/parted-devel/2025-
May/005930.html>
The patch is now merged upstream with commit e61584dd.
A plan B for trixie release is to add a workaround in partman-base to
double-check the filesystem type with blkid or other available tool
when the filesystem detected by parted_server is ext2.
I don't think it makes sense to resort to plan B. Either it should be
fixed in parted or it should just be put off until forky. My question
is whether it's worth fixing in trixie at all (never mind a bookworm
point release) given what so far has been very sparse information about
its actual real-world impact.
This is why I started this discussion.
