On 2026-11-08 16:49 +03:00, Pascal Hambourg wrote:
> On 06/11/2026 at 13:40, Alper Nebi Yasak wrote:
>> On 2026-11-06 01:18 +03:00, Pascal Hambourg wrote:
>>>
>>>>> The MR adds recipes for armhf, are there any ChromeOS machines with this
>>>>> architecture ?
>>>>
>>>> Yes, but we can't boot d-i through this boot method on those due to size
>>>> limits (8/16/32 MiB depending on board) on the bootable image size.
> (...)
>>> IIUC your MR adds the ChromeOS subarchitecture for armhf too. So how
>>> should it be dealt with ?
>>
>> Two 32 MiB partitions maybe, in case I end up making installer images with
>> custom kernels. But not if it's a maintenance burden.
>
> Now partman-auto has a crude script which allows to build recipes from
> "templates" and a set of parameters for each partition type, so there is
> no maintenance burden for a new recipe after adding the template.
>
>>>>>>> Never clearing the flag seems a bit extreme. What about using the same
>>>>>>> logic as in partman-efi to preserve unused existing ESPs instead ?
>>>>>>
>>>>>> Maybe they chose "Do not use" on existing partitions expecting it
>>>>>> to be like the ESP case? IIRC I thought it makes no sense to mark a CrOS
>>>>>> partition as "Do not use" except to indicate it should stop being one.
>>>>>
>>>>> OK, but partman-auto marks all partitions "do not use" before starting
>>>>> automatic partitioning.
>>>>
>>>> Ah. In that case switching to the ESP logic makes sense.
> (...)
>> I guess ESP logic would make the "do not
>> use" interpretation more consistent with other partman-* packages, which
>> sounds good.
>
> Actually only partman-efi implements this logic. Setting an ESP "do not
> use" is required to prevent the installer from possibly formatting and
> mounting it, but we do not want it to clear the original ESP flag.
>
> IIUC ChromeOS is different. Partman does not really "use" (format or
> mount) a ChromeOS partition, it just sets the type flag on it and checks
> if such partition is present on the boot/root disk. And we may actually
> want manual "do not use" to clear the flag.
>
> So ESP and ChromeOS partitions may need different interpretations of "do
> not use".
The "flag" here sets the GPT Partition Type GUID for the partition (a bad
extension of a legacy abstraction), and only one such flag makes sense on
any partition, and having none means it defaults to whatever parted likes
(Linux data?, I forgot). So to clear these flags, it may arguably be better
to explicitly select to use them as something else?
(Again, no strong preference.)
>> Not marking as "do not use" in partman-auto would work, but
>> feels like a flag-specific hack side-stepping the "method-old" logic at
>> a first glance.
>
> (Note: 'method-old' is primarily intended for re-using existing
> partitions which match the recipe - typically BIOS boot, ESP and swap -
> with automatic partitioning using the largest free space)
>
> Actually this issue is not specific to ChromeOS partitions because
> partman-auto clears the method on all partitions. BIOS Boot and PReP
> flags are also cleared and this can have operational impact because they
> are used by boot loader tools and platform firmware. partman-lvm and
> partman-md have their own logic to restore the LVM and RAID methods and
> preserve their flags even though they are mostly "cosmetic" and have no
> operational impact.
>
> So I feel like this issue may deserve a more general solution and the
> expected behaviour for method and flag management should be defined first.
I guess so, but I think I'd have to dive into the partman-auto code before
having good opinions on what to do. (Sorry, I don't have much bandwidth for
that right now.)
> Concerned methods/flags: BIOS Boot, ChromeOS, PReP, ESP, LVM, RAID.
>
> Two actions may clear methods and flags:
> - Set partition to "Do not use" in manual partitioning (dont_use)
> - automatic partitioning (partman-auto)
>
> Current behaviour:
>
> BIOS Boot, ChromeOS, PReP:
> - dont_use: clear method and flag (unsure)
> - partman-auto: clear method and flag (bad)
>
> ESP:
> - dont_use: clear method, restore original ESP flag (good)
> - partman-auto: clear method, restore original ESP flag (bad)
>
> LVM, RAID:
> - dont_use: clear method and flag (unsure)
> - partman-auto: keep method and flag (good)
>
> Possible expected behaviour:
>
> BIOS Boot, ChromeOS, PReP:
> - dont_use: clear method and flag ?
> - partman-auto: keep method and flag
>
> ESP:
> - dont_use: clear method, restore original ESP flag
> - partman-auto: clear method, keep flag
>
> LVM, RAID:
> - dont_use: clear method, restore original LVM/RAID flag ?
> - partman-auto: keep method and keep flag
>
> Opinions ?
Maybe we need a "Empty partition" method that clears these flags, and
maybe zeros out the first MiB of a partition so it has no filesystem.
That would resolve the ambiguity of "Do not use" where we would always
keep the partition as-is.
Maybe codifying "Do not use" as a value in the method file ("keep/efi"
for example, where we can also have the old method info) could help?