Re: [PATCH 3/4] unpack-trees: Fix same() for submodules

From: Junio C Hamano <gitster@pobox.com>
Date: Dec 22, 2017, 19:57 UTC
Message-ID: <xmqq3742sfhv.fsf@gitster.mtv.corp.google.com>
In-Reply-To: <20171220221725.129162-4-sbeller@google.com>

Stefan Beller <sbeller@google.com> writes:

Show 9 quoted lines

> The function same(a, b) is used to check if two entries a and b are the
> same.  As the index contains the staged files the same() function can be
> used to check if files between a given revision and the index are the same.
>
> In case of submodules, the gitlink entry is showing up as not modified
> despite potential changes inside the submodule.
>
> Fix the function to examine submodules by looking inside the submodule.
> This patch alone doesn't affect any correctness garantuees, but in

guarantees?  I somehow misread this as orangutan ;-)

Show 21 quoted lines

> combination with the next patch this fixes the new test introduced
> earlier in this series.
>
> This may be a slight performance regression as now we have to
> inspect any submodule thouroughly.
>
> Signed-off-by: Stefan Beller <sbeller@google.com>
> ---
>  unpack-trees.c | 2 ++
>  1 file changed, 2 insertions(+)
>
> diff --git a/unpack-trees.c b/unpack-trees.c
> index bf8b602901..4d839e8edb 100644
> --- a/unpack-trees.c
> +++ b/unpack-trees.c
> @@ -1427,6 +1427,8 @@ static int same(const struct cache_entry *a, const struct cache_entry *b)
>  		return 1;
>  	if ((a->ce_flags | b->ce_flags) & CE_CONFLICTED)
>  		return 0;
> +	if (S_ISGITLINK(b->ce_mode) && should_update_submodules())
> +		return !oidcmp(&a->oid, &b->oid) && !is_submodule_modified(b->name, 0);

Isn't this is working at the wrong level?

It is technically correct to say "the function is used to check if two entries a and be are the same", but that statement misses a more important aspect of the function. It is asked if these two things are the same and does not get involved in seeing if the working tree is dirty with respect to (one of) the entries it was passed.

For example, a virtual merge in a recursive merge would not (and should not) care if local changes exist in the working tree, and this function does not get any hint if it is to look at the working tree to tell such a case and the outermost merge apart.

Somebody has to be careful not to stomp on local changes in the outmost merge in a recursive merge, but I do not think this function is designed to be the place to do so. Isn't a submodule with "potential changes" the same way?

>  	return a->ce_mode == b->ce_mode &&
>  	       !oidcmp(&a->oid, &b->oid);
>  }