Re: [PATCH] pull: abort if --ff-only is given and fast-forwarding is impossible

[Elijah Newren <newren@gmail.com>]

On Mon, Jul 12, 2021 at 11:20 AM Alex Henrie <alexhenrie24@gmail.com> wrote:
>
> On Mon, Jul 12, 2021 at 11:51 AM Elijah Newren <newren@gmail.com> wrote:
> >
> > On Mon, Jul 12, 2021 at 10:08 AM Junio C Hamano <gitster@pobox.com> wrote:
> > >
> > > Phillip Wood <phillip.wood123@gmail.com> writes:
> > >
> > > > Thanks for revising this patch, I like this approach much better. I do
> > > > however have some concerns about the interaction of pull.ff with the
> > > > rebase config and command line options. I'd naively expect the
> > > > following behavior (where rebase can fast-forward if possible)
> > > >
> > > >   pull.ff  pull.rebase  commandline  action
> > > >    only     not false                rebase
> > > >    only     not false   --no-rebase  fast-forward only
> > > >     *       not false    --ff-only   fast-forward only
> > > >    only     not false    --ff        merge --ff
> > > >    only     not false    --no-ff     merge --no-ff
> > > >    only       false                  fast-forward only
> > > >    only       false      --rebase    rebase
> > > >    only       false      --ff        merge --ff
> > > >    only       false      --no-ff     merge --no-ff
> > >
> > > Do you mean by "not false" something other than "true"?  Are you
> > > trying to capture what should happen when these configuration
> > > options are unspecified as well (and your "not false" is "either set
> > > to true or unspecified")?  I ask because the first row does not make
> > > any sense to me.  It seems to say
> > >
> > >     "If pull.ff is set to 'only', pull.rebase is not set to 'false',
> > >     and the command line does not say anything, we will rebase".
> >
> > I think Phillip is trying to answer what to do when pull.ff and
> > pull.rebase conflict.  If I read his "not false" means "is set to
> > something other than false", then I agree with his table, but I think
> > he missed covering some cases.
> >
> > I think his table says that pull.rebase=false cannot conflict with
> > pull.ff settings, but any other value for pull.rebase can.  That makes
> > sense to me.
> >
> > I'd similarly say that pull.ff=true cannot conflict with any
> > pull.rebase settings...but that both pull.ff=only AND pull.ff=false
> > conflict with pull.rebase={true,merges}.
> >
> > My opinion would be:
> >   * conflicting command line flags results in the last one winning.
> >   * --no-rebase makes pull.ff determine the action.
> >   * --ff makes pull.rebase determine the action.
> >   * any other command line flag (-r|--rebase|--no-ff|--ff-only)
> > overrides both pull.ff and pull.rebase
> >   * If no command line option is given, and pull.ff and pull.rebase
> > conflict, then error out.
> >
> > I believe my recommendation above is consistent with every entry in
> > Phillip's table except the first line (where I suggest erroring out
> > instead).
>
> I'm not sure that --no-ff should imply --no-rebase because `git
> rebase` actually has a --no-ff option to rewrite commits even when
> fast-forwarding is possible. And it's not really necessary to make
> --ff-only imply --no-rebase because we're going to make `git pull`
> handle --ff-only itself without invoking `git merge`. However, the
> rest of this proposal could be implemented in a straightforward manner
> by making --rebase on the command line imply --ff, and I think that
> would be a fine solution.

git rebase has a --no-ff but it doesn't do anything like what git
pull's --no-ff has long been documented to do.  git pull's --no-ff
very clearly states that a merge commit will be created, and thus is
tied to merging instead of rebasing.

Also, I don't think pull needs to provide a union of all merge and
rebase options; so there's no need to add a way to invoke a 'rebase
--no-ff' from pull.  (Folks can run individual fetch & merge or rebase
steps, after all.)  I think we should concentrate on making sure that
we provide reasonable behavior when conflicting options/command-lines
from the already provided set are specified.