Re: [PATCH] migration: Fix writing mapped_ram + ignore_shared snapshots

[Fabiano Rosas <farosas@suse.de>]

Peter Xu <peterx@redhat.com> writes:

> On Thu, Nov 27, 2025 at 11:35:06AM +0000, Paweł Zmarzły wrote:
>> On Wed, 26 Nov 2025 at 21:42, Peter Xu <peterx@redhat.com> wrote:
>> > Now I start to question whether I should have that other fix of yours to be
>> > for this release or next.
>> >
>> > If this use case is completely broken, we shouldn't need to rush -rc
>> > window, now I plan to merge all these fixes later when 11.0 dev window
>> > opens.  Let me know if you, or Fabiano, has any comments.
>> 
>> It is broken if you set ignore-shared and actually have any shared
>> block, but what could work today is if you just toggle the
>> ignore-shared flag on without setting up any shared blocks. In that
>> case, writing will work fine, but reading will crash. That's how I
>
> Yep, I suppose either side of reliable failure means it's completely
> broken. :(
>
> That's IMHO an important evaluation because we could modify the image
> layout without worrying breaking others only if it's completely broken..
>
>> stumbled upon this rabbithole in the first place: I forgot to unset
>> the flag and was surprised by parsing error. Whether it is worth
>> fixing now - I don't know, setting ignore-shared when there are no
>> shared blocks doesn't really make sense, so most likely nobody does it
>> on purpose.
>> 
>> In either case, I need to stop working on this for now, I thought this
>> will be a tiny side project that'll help me get my first patches in
>> (and get used to collaborating over email), but now it's growing in
>> complexity and I have other things that I need to prioritize.
>> Hopefully I'll come back to this within 11.x window. Thanks for all
>> the help so far, Peter, Fabiano!
>
> Don't worry, thanks for all the contributions even so far!
>
> Your patch actually looks pretty good already and mergeable, I just
> nitpicked things here and there as I want to double check on things I
> stated, and make it slow to get thoroughly discussed.
>
> Personally, I think it's ok we queue this one already into -next together
> with the other one, then we clean things on top.
>
> Fabiano, sounds good to you?  PS: take your time reading, as long as you
> agree we put it in -next only, then there's no rush. :)

Yep, that's fine.

>
> Thanks,