From: Jonathan Nieder <jrnieder@gmail.com>
Date: Mon, 06 Dec 2010 07:32:14 GMT
Subject: Re: [PATCH] describe: Don’t look up commits with --exact-match
Message-ID: <20101206073214.GA3745@burratino>
In-Reply-To: <alpine.DEB.2.02.1012060149550.23348@dr-wily.mit.edu>

Anders Kaseorg wrote:
> On Fri, 3 Dec 2010, Jonathan Nieder wrote:

>> Instead of looking up the commit to be matched exactly in the commits
>> hash table, this makes a linear search.
[...]
> I don’t think it’s ever a problem: in my repository with 1800 tags on a 
> warm cache, that loop accounts for about 0.1% of even the fastest 
> non-exact-match query (a commit right after a tag).

Thanks for checking.  Makes sense.

>>> -	for_each_ref(get_name, NULL);
>>> +	for_each_rawref(get_name, NULL);
>>
>> Orthogonal change snuck in?
>
> This does fall under the category of “Don’t lookup commits,” and is 
> necessary to get the speedup (otherwise for_each_ref has already looked up 
> the commits that the rest of the patch is trying to avoid looking up).  

Depends on what "Don't lookup commits" means, I suppose.  I
think the difference between _ref and _rawref is

      if (!(flags & DO_FOR_EACH_INCLUDE_BROKEN)) {
            if (entry->flag & REF_BROKEN)
                  return 0; /* ignore dangling symref */
            if (!has_sha1_file(entry->sha1)) {
                  error("%s does not point to a valid object!", entry->name);
                  return 0;
            }
      }

so if I understand correctly, for_each_ref would still allow one to
get away without unpacking the objects.  Is that correct?