Re: Host refs to push are chosen (documentation error)?

From: Junio C Hamano <gitster@pobox.com>
Date: Feb 18, 2010, 02:04 UTC
Message-ID: <7vaav7wb8v.fsf@alter.siamese.dyndns.org>
In-Reply-To: <32541b131002171751n454081a9w90f2957d7a40d87@mail.gmail.com>

Avery Pennarun <apenwarr@gmail.com> writes:

Show 9 quoted lines

>> ++
>> +For any of the `$GIT_DIR/refs` cases above, packed refs from
>> +`$GIT_DIR/packed-refs` may be substituted (e.g., a line with
>> +`refs/heads/master` in the packed-refs file would have the same
>> +precedence as the file `$GIT_DIR/refs/heads/master`).
>
> Are they really equal precedence?  How does git decide, when there's
> one of each?  I'd guess packed-refs might actually be strictly higher
> precedence, but I'm not sure.

This is not about "I have packed refs/heads/master and then a loose refs/heads/master also exists" case. After packing, update ref simply creates a new loose ref, and we read loose ones and give precedence to it over packed one. This is so that you once gc and most of the unchanging refs will stay in packed file, but some active branches can be quickly flipped without having to rewrite the whole packed refs file (or removing the ref from that file).

But the statement is not about that.  The statement is about this:

    "refs/tags/master" gets higher precedence over "refs/heads/master",
    and this order is indifferent to where these two refs come from,
    either packed refs or loose in the filesystem.