Re: [PATCH 0/2] rebase --recreate-merges --keep-empty: don't prune empty

From: Phillip Wood <phillip.wood@talktalk.net>
Date: Mar 22, 2018, 11:03 UTC
Message-ID: <1f7194bb-f48b-e229-94ae-44c58636172f@talktalk.net>
In-Reply-To: <xmqqr2oer19v.fsf@gitster-ct.c.googlers.com>

On 20/03/18 19:32, Junio C Hamano wrote:

Show 21 quoted lines

> Phillip Wood <phillip.wood@talktalk.net> writes:
> 
>> On 20/03/18 15:42, Johannes Schindelin wrote:
>> ...
>>> As indicated in another reply, I'd rather rebase the --recreate-merges
>>> patches on top of your --keep-empty patch series. This obviously means
>>> that I would fold essentially all of your 2/2 changes into my
>>> "rebase-helper --make-script: introduce a flag to recreate merges"
>>>
>>> The 1/2 (with s/failure/success/g) would then be added to the
>>> --recreate-merges patch series at the end.
>>>
>>> Would that be okay with you?
>>
>> Yes, that's fine, it would give a clearer history
> 
> With or without the above plan, what we saw from you were a bit
> messy to queue.  The --keep-empty fix series is based on 'maint',
> while the --signoff series depends on changes that happened to
> sequencer between 'maint' and 'master', but yet depends on the
> former.

Yes, that is awkward and unfortunate but the idea behind splitting them into two separate series was to have a single set of bug fixes in the history. The feature needed to be based on master, so if I'd had the bug fixes in the same series you'd of had to cherry-pick them to maint which would break branch/tag --contains. I'm not sure if that is a better option.

Best Wishes

Phillip

Show 16 quoted lines

> In what I'll be pushing out at the end of today's integration run,
> I'll have two topics organized this way:
> 
>  - pw/rebase-keep-empty-fixes: built by applying the three
>    '--keep-empty' patches on top of 'maint'.
> 
>  - pw/rebase-signoff: built by first merging the above to 0f57f731
>    ("Merge branch 'pw/sequencer-in-process-commit'", 2018-02-13) and
>    then applying "rebase --signoff" series.
> 
> Also, I'll revert merge of Dscho's recreate-merges topic to 'next';
> doing so would probably have to invalidate a few evil merges I've
> been carrying to resolve conflicts between it and bc/object-id
> topic, so today's integration cycle may take a bit longer than
> usual.
>