possible fuzziness for b4 trailers -u?

possible fuzziness for b4 trailers -u?

[Vlastimil Babka (SUSE)]

Hi,

I sometimes run b4 trailers -u to collect tags on patches that have been
meanwhile slightly modified in git since the posting. Sometimes it's on the
b4/$cover_topic branch as a submitter, sometimes it's already applied (but
not stable as in no history rewrite) on a maintainer tree branch (I do use
--since-commit and doublecheck the result of the update closely since some
past surprises...).

AFAICS the trailers update relies only patch-id (?) so it will miss the tags
when patch-id changes due to these local modifications. Could it try some
fuzzy matching in that case? Some ideas:

- in b4/$cover_topic branch the metadata allows finding the posting thread
and matching the patch subject
  - ok, subject can sometimes change too, but not so often as the code

- in maintainer tree branch, there's usually a patch.msgid.link Link: that
can lead to the given patch rather deterministically.

Thanks for consideration!
Vlastimil

Re: possible fuzziness for b4 trailers -u?

[Konstantin Ryabitsev]

On Thu, Jun 18, 2026 at 10:50:52AM +0200, Vlastimil Babka (SUSE) wrote:
> I sometimes run b4 trailers -u to collect tags on patches that have been
> meanwhile slightly modified in git since the posting. Sometimes it's on the
> b4/$cover_topic branch as a submitter, sometimes it's already applied (but
> not stable as in no history rewrite) on a maintainer tree branch (I do use
> --since-commit and doublecheck the result of the update closely since some
> past surprises...).
> 
> AFAICS the trailers update relies only patch-id (?) so it will miss the tags
> when patch-id changes due to these local modifications. Could it try some
> fuzzy matching in that case? Some ideas:
> 
> - in b4/$cover_topic branch the metadata allows finding the posting thread
> and matching the patch subject
>   - ok, subject can sometimes change too, but not so often as the code
> 
> - in maintainer tree branch, there's usually a patch.msgid.link Link: that
> can lead to the given patch rather deterministically.

This was actually easier than I'd thought, because we already did the bulk of
this logic for "b4 dig". The latest master has a new --fuzzy flag to `b4
trailers`, which you can also couple with the new --interactive flag, which
should let you review the trailers before you actually apply them.

Try it out and let me know if this is what you were looking for.

Cheers,
-K

Re: possible fuzziness for b4 trailers -u?

[Vlastimil Babka (SUSE)]

On 6/18/26 20:30, Konstantin Ryabitsev wrote:
> On Thu, Jun 18, 2026 at 10:50:52AM +0200, Vlastimil Babka (SUSE) wrote:
>> I sometimes run b4 trailers -u to collect tags on patches that have been
>> meanwhile slightly modified in git since the posting. Sometimes it's on the
>> b4/$cover_topic branch as a submitter, sometimes it's already applied (but
>> not stable as in no history rewrite) on a maintainer tree branch (I do use
>> --since-commit and doublecheck the result of the update closely since some
>> past surprises...).
>> 
>> AFAICS the trailers update relies only patch-id (?) so it will miss the tags
>> when patch-id changes due to these local modifications. Could it try some
>> fuzzy matching in that case? Some ideas:
>> 
>> - in b4/$cover_topic branch the metadata allows finding the posting thread
>> and matching the patch subject
>>   - ok, subject can sometimes change too, but not so often as the code
>> 
>> - in maintainer tree branch, there's usually a patch.msgid.link Link: that
>> can lead to the given patch rather deterministically.
> 
> This was actually easier than I'd thought, because we already did the bulk of
> this logic for "b4 dig". The latest master has a new --fuzzy flag to `b4
> trailers`, which you can also couple with the new --interactive flag, which
> should let you review the trailers before you actually apply them.
> 
> Try it out and let me know if this is what you were looking for.

Just used it successfully, thanks a lot! The --interactive flag will also be
very useful.

> Cheers,
> -K