[PATCH] Documentation/git-bisect.txt: add --no-ff to merge command

[PATCH] Documentation/git-bisect.txt: add --no-ff to merge command

[Mihail Atanassov]

The hotfix application example uses `git merge --no-commit` to apply
temporary changes to the working tree during a bisect operation. In some
situations this can be a fast-forward and `merge` will apply the hotfix
branch's commits regardless of `--no-commit` (as documented in the `git
merge` manual).

In the pathological case this will make a `git bisect
run` invocation to loop indefinitely between the first bisect step and
the fast-forwarded post-merge HEAD.

Add `--no-ff` to the merge command to avoid this issue, and make a note
of it for the reader.

Signed-off-by: Mihail Atanassov <m.atanassov92@gmail.com>
---
 Documentation/git-bisect.txt | 6 ++++--
 1 file changed, 4 insertions(+), 2 deletions(-)

diff --git a/Documentation/git-bisect.txt b/Documentation/git-bisect.txt
index 4b45d837a7..58b5585874 100644
--- a/Documentation/git-bisect.txt
+++ b/Documentation/git-bisect.txt
@@ -412,8 +412,10 @@ $ cat ~/test.sh
 #!/bin/sh
 
 # tweak the working tree by merging the hot-fix branch
-# and then attempt a build
-if	git merge --no-commit hot-fix &&
+# and then attempt a build. Note the `--no-ff`: `git merge`
+# will otherwise still apply commits if the current HEAD can be
+# fast-forwarded to the hot-fix branch.
+if	git merge --no-commit --no-ff hot-fix &&
 	make
 then
 	# run project specific test and report its status
-- 
2.16.4

Re: [PATCH] Documentation/git-bisect.txt: add --no-ff to merge command

[Jonathan Nieder]

Hi,

Mihail Atanassov wrote:

> The hotfix application example uses `git merge --no-commit` to apply
> temporary changes to the working tree during a bisect operation. In some
> situations this can be a fast-forward and `merge` will apply the hotfix
> branch's commits regardless of `--no-commit` (as documented in the `git
> merge` manual).
>
> In the pathological case this will make a `git bisect
> run` invocation to loop indefinitely between the first bisect step and
> the fast-forwarded post-merge HEAD.
>
> Add `--no-ff` to the merge command to avoid this issue, and make a note
> of it for the reader.
>
> Signed-off-by: Mihail Atanassov <m.atanassov92@gmail.com>
> ---
>  Documentation/git-bisect.txt | 6 ++++--
>  1 file changed, 4 insertions(+), 2 deletions(-)

Good catch.  Thanks for fixing it.

> diff --git a/Documentation/git-bisect.txt b/Documentation/git-bisect.txt
> index 4b45d837a7..58b5585874 100644
> --- a/Documentation/git-bisect.txt
> +++ b/Documentation/git-bisect.txt
> @@ -412,8 +412,10 @@ $ cat ~/test.sh
>  #!/bin/sh
>  
>  # tweak the working tree by merging the hot-fix branch
> -# and then attempt a build
> +# and then attempt a build. Note the `--no-ff`: `git merge`
> +# will otherwise still apply commits if the current HEAD can be
> +# fast-forwarded to the hot-fix branch.

Hmm.  I think the comment might put a bit too much emphasis on the
"how" instead of the "why".  Is it necessary to describe why --no-ff
is used at all here?  After all, a reader wondering about it is likely
to check "git help merge", which says

	Fast-forward updates do not create a merge commit and
	therefore there is no way to stop those merges with
	--no-commit.  Thus, if you want to ensure your branch is not
	changed or updated by the merge command, use --no-ff with
	--no-commit.

So I'd be tempted to leave the comment ending with "and then attempt a
build".

Alternatively: the wording says "will still apply commits", but the
reader might not think of a merge as applying patches (that's closer
to what cherry-pick does.  Is there some alternative wording that
would convey the intent more clearly?

> -if	git merge --no-commit hot-fix &&
> +if	git merge --no-commit --no-ff hot-fix &&

Good.

Thanks and hope that helps,
Jonathan

Re: [PATCH] Documentation/git-bisect.txt: add --no-ff to merge command

[Mihail Atanassov]

(Cc git@vger.kernel.org)

On Mon, 28 Oct 2019 at 21:51, Mihail Atanassov <m.atanassov92@gmail.com> wrote:
>
> Hi Jonathan,
>
> Thanks for the quick turnaround! And apologies in advance for the delayed
> and potentially mangled response, I can't get into my gmail account from
> a sensible MUA...
>
> On Sat, 26 Oct 2019 at 03:26, Jonathan Nieder <jrnieder@gmail.com> wrote:
> >
> > Hi,
> >
> > Mihail Atanassov wrote:
> >
> > > The hotfix application example uses `git merge --no-commit` to apply
> > > temporary changes to the working tree during a bisect operation. In some
> > > situations this can be a fast-forward and `merge` will apply the hotfix
> > > branch's commits regardless of `--no-commit` (as documented in the `git
> > > merge` manual).
> > >
> > > In the pathological case this will make a `git bisect
> > > run` invocation to loop indefinitely between the first bisect step and
> > > the fast-forwarded post-merge HEAD.
> > >
> > > Add `--no-ff` to the merge command to avoid this issue, and make a note
> > > of it for the reader.
> > >
> > > Signed-off-by: Mihail Atanassov <m.atanassov92@gmail.com>
> > > ---
> > >  Documentation/git-bisect.txt | 6 ++++--
> > >  1 file changed, 4 insertions(+), 2 deletions(-)
> >
> > Good catch.  Thanks for fixing it.
> >
> > > diff --git a/Documentation/git-bisect.txt b/Documentation/git-bisect.txt
> > > index 4b45d837a7..58b5585874 100644
> > > --- a/Documentation/git-bisect.txt
> > > +++ b/Documentation/git-bisect.txt
> > > @@ -412,8 +412,10 @@ $ cat ~/test.sh
> > >  #!/bin/sh
> > >
> > >  # tweak the working tree by merging the hot-fix branch
> > > -# and then attempt a build
> > > +# and then attempt a build. Note the `--no-ff`: `git merge`
> > > +# will otherwise still apply commits if the current HEAD can be
> > > +# fast-forwarded to the hot-fix branch.
> >
> > Hmm.  I think the comment might put a bit too much emphasis on the
> > "how" instead of the "why".  Is it necessary to describe why --no-ff
> > is used at all here?  After all, a reader wondering about it is likely
> > to check "git help merge", which says
> >
> >         Fast-forward updates do not create a merge commit and
> >         therefore there is no way to stop those merges with
> >         --no-commit.  Thus, if you want to ensure your branch is not
> >         changed or updated by the merge command, use --no-ff with
> >         --no-commit.
> >
> > So I'd be tempted to leave the comment ending with "and then attempt a
> > build".
>
> Fair point, I actually did spend a bit of time on the fence between your
> suggestion and what I ultimately submitted. I ended up expanding on it
> precisely because the '--no-ff' seems a bit arbitrary to the casual observer
> and requires cross-referencing other documentation (which is how I figured
> out I ought to produce this patch :)).
>
> I can't think of any wording that would be any better, so I'll push a v2 with
> no comment changes, and leave it to the reader's curiosity (or lack thereof).
>
> On a related note, if the user reads all the docs fully, they'll know to use a
> suitable merge-base for the hotfix branch and they won't get into the
> predicament in the first place. So this patch is hiding the underlying issue
> slightly. I'd still prefer to have that failsafe in there, though, for the cases
> where going into an infinite loop is costly (e.g. unattended bisect with
> long-running tests).
>
> >
> > Alternatively: the wording says "will still apply commits", but the
> > reader might not think of a merge as applying patches (that's closer
> > to what cherry-pick does.  Is there some alternative wording that
> > would convey the intent more clearly?
> >
> > > -if   git merge --no-commit hot-fix &&
> > > +if   git merge --no-commit --no-ff hot-fix &&
> >
> > Good.
> >
> > Thanks and hope that helps,
> > Jonathan
>
> --
> Mihail

-- 
Mihail

Re: [PATCH] Documentation/git-bisect.txt: add --no-ff to merge command

[Jonathan Nieder]

Hi,

Mihail Atanassov wrote:

> Thanks for the quick turnaround! And apologies in advance for the delayed
> and potentially mangled response, I can't get into my gmail account from
> a sensible MUA...

Interesting.  https://support.google.com/mail/thread/11736136 tells me
there's an issue with Kmail's oauth support.  You might want to get in
touch with the Kmail authors, or, as a fallback, use an application
specific password or other mail client.

[...]
> On Sat, 26 Oct 2019 at 03:26, Jonathan Nieder <jrnieder@gmail.com> wrote:

>> Hmm.  I think the comment might put a bit too much emphasis on the
>> "how" instead of the "why".
[...]
>> So I'd be tempted to leave the comment ending with "and then attempt a
>> build".
>
> Fair point, I actually did spend a bit of time on the fence between your
> suggestion and what I ultimately submitted. I ended up expanding on it
> precisely because the '--no-ff' seems a bit arbitrary to the casual observer
> and requires cross-referencing other documentation (which is how I figured
> out I ought to produce this patch :)).
>
> I can't think of any wording that would be any better, so I'll push a v2 with
> no comment changes, and leave it to the reader's curiosity (or lack thereof).

Thanks, that sounds good to me.

As an orthogonal point, I wonder whether we can start the multi-step
migration of making --no-commit imply --no-ff by default:

 1. Act as --ff when --no-commit is passed without --ff or --no-ff
    (the state today)

 2. Warn when performing a fast-forward merge and --no-commit was
    passed without --ff or --no-ff

 3. Error out instead of performing a fast-forward merge when
    --no-commit is passed without --ff or --no-ff

 4. Warn and refuse to perform a fast-forward merge when --no-commit
    is passed without --ff or --no-ff

 5. Refuse to perform a fast-forward merge with --no-commit is passed
    without --ff or --no-ff, just as though --no-ff were passed.

(A config setting could allow people to get the futuristic behavior
early.  And it might be possible to skip some steps. :))

[...]
>>> -if   git merge --no-commit hot-fix &&
>>> +if   git merge --no-commit --no-ff hot-fix &&
>>
>> Good.

This part still looks like a good change to me. :)

Sincerely,
Jonathan

Re: [PATCH] Documentation/git-bisect.txt: add --no-ff to merge command

[Junio C Hamano]

Jonathan Nieder <jrnieder@gmail.com> writes:

> As an orthogonal point, I wonder whether we can start the multi-step
> migration of making --no-commit imply --no-ff by default:
>
>  1. Act as --ff when --no-commit is passed without --ff or --no-ff
>     (the state today)

which means "--no-commit controls whether a new commit is created or
not and nothing else, and because --ff is the default for merge,
merging a true descendant will fast-forward".

>  5. Refuse to perform a fast-forward merge with --no-commit is passed
>     without --ff or --no-ff, just as though --no-ff were passed.

Is that a good endgame, though?  It is correct that "--no-ff" means
"do not allow the merge to be fast-forwarded and the way the option
does so is by creating an otherwise unnecessary merge commit", and
"--no-commit" means "do not allow creating any new commit", so
technically they are mutually incompatible, but would it be useful?

I'd imagine that a more useful behaviour would be for "git merge X"
with any other options to honor this basic trait: the working tree
and the index after the operation shows the result of merging X and
HEAD, if the merge can cleanly be made, and otherwise the working
tree and the index would show something close to the result of such
a merge with conflicts that would help recording the result of the
merge manually.

For that, wouldn't it make more sense ot change the semantics of the
"--no-commit" option from "no new commit gets created" to "HEAD is
not moved"?  "git merge --no-commit X" when X is a descendant of
HEAD would then become "git read-tree -m -u HEAD X" plus perhaps
storing X in .git/MERGE_HEAD file etc. to prepare for concluding
"git commit" to record the result manually.

In any case, as you said,

>>>> -if   git merge --no-commit hot-fix &&
>>>> +if   git merge --no-commit --no-ff hot-fix &&
>>>
>>> Good.
>
> This part still looks like a good change to me. :)

This looks good to me too.

Re: [PATCH] Documentation/git-bisect.txt: add --no-ff to merge command

[Junio C Hamano]

Junio C Hamano <gitster@pobox.com> writes:

> Jonathan Nieder <jrnieder@gmail.com> writes:
>
>> As an orthogonal point, I wonder whether we can start the multi-step
>> migration of making --no-commit imply --no-ff by default:
>>
>>  1. Act as --ff when --no-commit is passed without --ff or --no-ff
>>     (the state today)
>
> which means "--no-commit controls whether a new commit is created or
> not and nothing else, and because --ff is the default for merge,
> merging a true descendant will fast-forward".
>
>>  5. Refuse to perform a fast-forward merge with --no-commit is passed
>>     without --ff or --no-ff, just as though --no-ff were passed.
>
> Is that a good endgame, though?

Ah, I was confused by "refuse to perform".  You were not trying to
make the command fail outright without doing anything.  Yes, that
would be a good endgame, I would think.

I am not sure if the transition would be smooth, though.