From: Phillip Wood <phillip.wood123@gmail.com>
Date: Thu, 21 Mar 2019 14:35:12 GMT
Subject: Re: [RFC PATCH 01/11] sequencer: always discard index after checkout
Message-ID: <8c663552-b792-b885-ea59-abe3137efe26@gmail.com>
In-Reply-To: <CACsJy8CRN1NUD1DdKXkOoaYqCU_RJU56ph=eBAGSS-EXMAcmww@mail.gmail.com>

On 20/03/2019 01:50, Duy Nguyen wrote:
> On Wed, Mar 20, 2019 at 2:04 AM Phillip Wood <phillip.wood123@gmail.com> wrote:
>>      It would perhaps be better to pass around the_index rather than
>>      the_repository
> 
> Not by a large margin. For sequencer.c most operations require more
> than just the index and passing 'struct repository *' around has been
> the norm. And as soon as you need to load the index back (not sure if
> you should do it here btw, after discard_index, since we have the
> index loaded before) you need 'struct repository' not 'struct
> index_state'.

Thanks, I think I'll leave it as it is then. After we checkout the new 
base we reload the index in the loop that picks the commits. For 'rebase 
<upstream> <branch>' after we checkout <branch> we create the todo-list 
which involves a revision walk and then checkout the new base. I'm not 
entirely sure if it needs reloading before we create the todo list but I 
think it probably not as I don't think rebase--interactive.c loads the 
index (which would explain why this only becomes an issue when we stop 
forking rebase--interactive from rebase because then rebase.c has loaded 
the index to check there are no uncommitted changes)

Best Wishes

Phillip



> 
>>   builtin/rebase--interactive.c |  2 +-
>>   sequencer.c                   | 27 +++++++++++++++++----------
>>   sequencer.h                   |  3 ++-
>>   3 files changed, 20 insertions(+), 12 deletions(-)