Re: [PATCH 6/7] read-cache: refuse to create index referring to external objects

From: Junio C Hamano <gitster@pobox.com>
Date: Jan 28, 2013, 06:06 UTC
Message-ID: <7vtxq1yh6t.fsf@alter.siamese.dyndns.org>
In-Reply-To: <CACsJy8CXC=poDBenBwo6t8=Qv-_zvGbvHYo-cDdGA8_fpw4Cyg@mail.gmail.com>

Duy Nguyen <pclouds@gmail.com> writes:

Show 6 quoted lines

>>> Another thing needs to be done for this to work. The current reading
>>
>> For *what* to work???
>
> The "forbid making our repository depend on objects we do not have but
> we know about afterwe peek submodule odb"

With your "when our object database is contaminated, check objects we base our new object on are available in local or our alternates" together with the "when we try write_object(), do not bypass it with has_sha1_file() check because that may find the object in submodule odb we should *not* have access to; instead check with the same 'local or our alternates' test" I brought up in the message you were responding to, I do not think object read order does not make a difference to our effort to prevent the object database breakage due to temporary contamination by submodule objects.

>>> Regardless the submodule odb issue, I think we should prefer
>>> reading local loose objects over alternate packed ones.

I suspect you are alluding to make write_object() check with has_sha1_file_locally() so that we can wean our repository from existing alternates, but I do not think it is a right approach (instead, you just fully repack locally if you want to dissociate yourself from your alternates). What I was suggesting was to change it to check with has_sha1_file_proper(), to make sure we do not omit writing an object we need to access to, when we know it will vanish once we stop temporarily borrowing from the submodule object store.