Re: Patchwork review workflow: archival rules

[Siddhesh Poyarekar <siddhesh@gotplt.org>]

On 11/30/21 03:50, Carlos O'Donell wrote:
> On 11/15/21 23:22, Siddhesh Poyarekar wrote:
>> Hello,
>>
>> What do people think about adding the following to the patch review
>> workflow[1]?  This should help keep the queue under control.
>>
>> ~~~~~~
>> 3. Maintaining your patch queue
>>
>> 3.1 (existing content)
>>
>> 3.2. Outdated Patches
>>
>> To keep tha backlog in patchwork manageable, outdated patches may be archived at regular intervals.
>>
>>   * Patches in Changes Requested status will be archived 1 year after they have been posted
> 
> 
> Agreed. And we can decrease the "1 year after" if we find it helps keep the queue cleaner.
>   
>>   * Patches older than 3 months that fail to apply to the current main branch will be set to Rejected
> 
> This will be an important step in keeping the queue clean.
> 
> I agree with this move.
> 
>>   * Unresolved patches older than 2 years will be archived.
> 
> Agreed.
> 
>> ~~~~~~
>>
>> Thanks,
>> Siddhesh
>>
>> [1] https://sourceware.org/glibc/wiki/Patch%20Review%20Workflow
> 
> All of this looks good to me.
> 
> We can change it any time we want IMO, so long as we set expectations and
> have a public discussion with our reason for making the change.

Thanks, I'll make the change in the wiki and run some commands over the 
weekend to implement this.

> I'm curious if you have data about the above?

What kind of data are you looking for?  I came up with arbitrary time 
limits that I believe are conservative enough.  Rebasing a patchset that 
doesn't apply to current mainline for example is a necessary condition 
and there's no way we'd accept it unless it is rebased (and hence a new 
version posted) but I still proposed waiting 3 months before changing 
status as a conservative measure.

Similarly for archiving Changes Requested patches; there's no way it's 
going to come back to New or Accepted/Committed unless the consensus on 
the exact same patch changes, which I have never seen happen in the past 
decade.  That is, there's always an updated version that drives 
consensus.  I still kept the timeout at 1 year as a conservative start.

Siddhesh