From: Junio C Hamano <gitster@pobox.com>
Date: Sat, 14 Mar 2015 22:19:03 GMT
Subject: Re: [PATCH v2 01/10] Define a structure for object IDs.
Message-ID: <xmqqbnjvw7c8.fsf@gitster.dls.corp.google.com>
In-Reply-To: <CACsJy8BeFO3RQJ6pU5+q6E4n3A36X9PgYoi7xB_NQ4C_JYaP5w@mail.gmail.com>

Duy Nguyen <pclouds@gmail.com> writes:

>> Anyway, wouldn't this be all academic?  I do not see how you would
>> keep the object name in the <pack, nth> format in-core, as the
>> obj_hash[] is a hashtable keyed by <sha-1>, and even when we switch
>> to a different hash, I cannot see how such a table to ensure the
>> singleton-ness of in-core objects can be keyed sometimes by <hash>
>> and by <pack, nth> in some other time.
>
> I'm implementing something to see how much we gain by avoiding object
> lookup. The current approach is having "struct object ** obj" in
> "struct packed_git", indexed by "nth". So when you have <pack, nth>
> and pack->obj[nth] is valid, you'll get to "struct object *" without
> hashing.

But do you realize that the hashtable serves two purposes?  Grab the
object from its name is one thing, and the other one I am not seeing
how you will make it work with "sometimes <sha-1> sometimes <pack,nth>"
is to ensure that we will have only one in-core copy for the same object.
We even walk the hashtable when we want to drop the flag bits from
all in-core objects, so even if you instanciated an in-core object
without going through the object name layer, the hashtable needs to
have a pointer to such a pointer, no?