Re: [PATCH v2] fetch: Protect branches checked out in all worktrees

[Junio C Hamano <gitster@pobox.com>]

Jeff King <peff@peff.net> writes:

> But all of that means we could actually drop check_not_current_branch()
> in favor of the update_local_ref() check. Doing this:
>
> diff --git a/builtin/fetch.c b/builtin/fetch.c
> index f7abbc31ff..c52c44684a 100644
> --- a/builtin/fetch.c
> +++ b/builtin/fetch.c
> @@ -869,7 +869,7 @@ static int update_local_ref(struct ref *ref,
>  	}
>  
>  	if (current_branch &&
> -	    !strcmp(ref->name, current_branch->name) &&
> +	    !strcmp(ref->name, current_branch->refname) &&
>  	    !(update_head_ok || is_bare_repository()) &&
>  	    !is_null_oid(&ref->old_oid)) {
>  		/*
> @@ -1385,20 +1385,6 @@ static int prune_refs(struct refspec *rs, struct ref *ref_map,
>  	return result;
>  }
>  
> -static void check_not_current_branch(struct ref *ref_map)
> -{
> -	struct branch *current_branch = branch_get(NULL);
> -
> -	if (is_bare_repository() || !current_branch)
> -		return;
> -
> -	for (; ref_map; ref_map = ref_map->next)
> -		if (ref_map->peer_ref && !strcmp(current_branch->refname,
> -					ref_map->peer_ref->name))
> -			die(_("Refusing to fetch into current branch %s "
> -			    "of non-bare repository"), current_branch->refname);
> -}
> -
>  static int truncate_fetch_head(void)
>  {
>  	const char *filename = git_path_fetch_head(the_repository);
> @@ -1587,8 +1573,6 @@ static int do_fetch(struct transport *transport,
>  
>  	ref_map = get_ref_map(transport->remote, remote_refs, rs,
>  			      tags, &autotags);
> -	if (!update_head_ok)
> -		check_not_current_branch(ref_map);
>  
>  	if (tags == TAGS_DEFAULT && autotags)
>  		transport_set_option(transport, TRANS_OPT_FOLLOWTAGS, "1");
>
> passes the entire test suite except for one test which expects:
>
>   git fetch . side:main
>
> to fail. But that is only because "side" and "main" point to the same
> commit, and thus the fetch is a noop. The code in update_local_ref()
> covers that case before checking the HEAD case (which I would argue is
> a completely reasonable outcome).

So we can lose redundant code that was added there only because the
original safety was broken by actually fixing the original safety?

That is very nice.

> The reason I bring this up is that I think doing the check in
> update_local_ref() makes much more sense. We don't abort the whole
> fetch, but just treat it as a normal per-ref failure. That gives us the
> usual status-table output (I thought it might also avoid wasting some
> work of actually fetching objects, but I think the current check kicks
> in before we actually fetch anything).

Yes I agree with that reasoning.  

Thanks for digging this to the bottom, both of you.