Re: Stable release process proposal (Was: Linux 5.10.109)

[Alexey Khoroshilov]

On 30.03.2022 07:36, Greg Kroah-Hartman wrote:
> On Wed, Mar 30, 2022 at 02:49:00AM +0300, Alexey Khoroshilov wrote:
>> Dear Greg,
>>
>> First of all, thank you very much for keeping stable maintenance so well.
>>
>> We (Linux Verification Center of ISPRAS (linuxtesting.org)) are going to
>> join a team of regular testers for releases in 5.10 stable branch (and
>> other branches later). We are deploying some test automation for that
>> and have met an oddity that would to discuss.
>>
>> Sometimes, like in 5.10.109 release, we have a situation when a
>> released version (5.10.109) differs from the release candidate
>> (5.10.109-rс1). In this case there was a patch "llc: only change
>> llc->dev when bind()succeeds" added to fix a bug in another llc fix.
>> Unfortunately, as Pavel noted, this patch does not fix a bug, but
>> introduces a new one, because another commit b37a46683739 ("netdevice:
>> add the case if dev is NULL") was missed in 5.10 branch.
> This happens quite frequently due to issues found in testing.  It's not
> a new thing.
>
>> The problem will be fixed in 5.10.110, but we still have a couple oddities:
>> - we have a release that should not be recommended for use
>> - we have a commit message misleading users when says:
>>
>>     Tested-by: Pavel Machek (CIP) <pavel@xxxxxxx>
>>     Tested-by: Fox Chen <foxhlchen@xxxxxxxxx>
>>     Tested-by: Florian Fainelli <f.fainelli@xxxxxxxxx>
>>     Tested-by: Shuah Khan <skhan@xxxxxxxxxxxxxxxxxxx>
>>     Tested-by: Bagas Sanjaya <bagasdotme@xxxxxxxxx>
>>     Tested-by: Salvatore Bonaccorso <carnil@xxxxxxxxxx>
>>     Tested-by: Linux Kernel Functional Testing <lkft@xxxxxxxxxx>
>>     Tested-by: Sudip Mukherjee <sudip.mukherjee@xxxxxxxxxxxxxxx>
>>     Tested-by: Guenter Roeck <linux@xxxxxxxxxxxx>
>>
>> but actually nobody tested that version.
>>
>> There are potential modifications in stable release process that can
>> prevent such problems:
>>
>> (1) to always release rс2 when there are changes in rc1 introduced
>>
>> (2) to avoid Tested-by: section from release commits in such situations.
>>
>> Or may be it is overkill and it too complicates maintenance work to be
>> worth. What do you think?
> I think it's not worth the extra work on my side for this given the
> already large workload.  What would benifit from this to justify it?
I see, thank you.

I believed the goal is to provide some minimal quality guarantees for a
particular version of the code. But if the process of updates is quite
intensive, it may make sense to transfer responsibility for particular
release verification downstream.

Best regards,
Alexey