From: Junio C Hamano <gitster@pobox.com>
Date: Tue, 19 Apr 2016 23:02:22 GMT
Subject: Re: [PATCH 4/6] Convert struct name_entry to use struct object_id.
Message-ID: <xmqqshyhxkw1.fsf@gitster.mtv.corp.google.com>
In-Reply-To: <1460934641-435791-5-git-send-email-sandals@crustytoothpaste.net>

"brian m. carlson" <sandals@crustytoothpaste.net> writes:

> @@ -314,7 +314,7 @@ static int threeway_callback(int n, unsigned long mask, unsigned long dirmask, s
>  	}
>  
>  	if (same_entry(entry+0, entry+1)) {
> -		if (entry[2].sha1 && !S_ISDIR(entry[2].mode)) {
> +		if (entry[2].oid->hash && !S_ISDIR(entry[2].mode)) {

Thanks for a warning in the cover letter.

"if (entry[2].oid && !S_ISDIR(entry[2].mode)" would be a faithful
conversion, wouldn't it?

threeway-callback is called from the unpack-trees codepath and
tree-walk.c::traverse_trees() is what prepared these entries.

An entry[] which is an N-element array represents the traversal
state of walking N trees in parallel, and the callback function like
this one gets an entry that has been tree-walk.c::entry_clear()ed
(i.e. memset to full of '\0's) for a tree that lacks the corresponding
path.

Try to do merge-tree of two trees, where there is a path that we did
not change since common ancestor, while the other side removed it,
and you would hit this if () statement, I would think.