Re: b4 review workflow issue with partially accepted serieses

On Tue, 05 May 2026 09:26:42 +0900, Mark Brown wrote:
> Hi Konstantin,
> 
> b4 review tracks the state of a series at the top level.  This gets
> confused if you partially accept a series (eg, taking fixes at the start
> of a bigger series that needs more work) then the state gets confused.
> The bits that are accepted trigger the applied->thanks->archive flow
> but that means that new versions of the series can't be ingested even
> after the series is archived.  This is a pretty severe workflow problem,
> there should be a way of splitting serieses and/or retrieving them from
> the thanked/archived states to reviewing.

Okay, I tried to figure out a good way to approach this. Splitting
series was taking me down to a very bad rabbit hole, so instead I added
a special "partial" state that will hopefully do what you're looking for
-- but please provide feedback if it's not.

Basically:

When you take a series using cherry-pick and only select a subset of its
patches, b4 now checks coverage across all patches and sets the status
to "partial" (shown as ◐ in the listing) instead of "accepted". The
review branch stays alive and in-tree exactly as it was, and the series
remains in the active working set — so when a v2 arrives, b4 picks it up
and the update_revision action lights up just as it would for a series
you're still reviewing.

You can then review it again to take more patches or do anything else
with a series that is being reviewed. If you take a subset of patches
again, the series will go back to the "partial" state. If you take them
all, it will go to "accepted".

From the action menu a "partial" series offers everything a "reviewing"
series does: take remaining patches, rebase, range-diff, upgrade to a
newer revision, snooze, mark as waiting, abandon, archive, thank, etc.

I also added a smaller bonus fix while I was in there: a "thanked"
series can now be reopened back to "reviewing" via the action menu
("Return to reviewing"), in case something was thanked prematurely. The
review branch is still in-tree for a thanked series so no reconstruction
is needed.

This is *lightly* tested and hopefully won't blow up too many times.
Modeling complex series workflows is a challenge, but hopefully any
explosions will be contained.

Thanks!
-- 
Konstantin Ryabitsev <[email protected]>