From: Jonathan Tan <jonathantanmy@google.com>
Date: Thu, 18 Jan 2018 22:31:09 GMT
Subject: Re: What's cooking in git.git (Jan 2018, #02; Tue, 9)
Message-ID: <20180118143109.bc0eb10b036411e0a5a3694f@google.com>
In-Reply-To: <CAP8UFD0P7kVo2NP4Wq7OaSV4H1+sqHapuzW5AQef+enNS0S5hw@mail.gmail.com>

On Thu, 18 Jan 2018 09:56:50 +0100
Christian Couder <christian.couder@gmail.com> wrote:

> I am still not very happy with fetch_object() not returning anything.
> I wonder what happens when that function is used to fetch from a repo
> that cannot provide the requested object.

My idea was to save a verification step - the caller of fetch_object()
needs to reattempt the object load anyway (which includes a verification
that the object exists), so I didn't see the need to have fetch_object()
do it too.

> Also I think the "extensions.partialclone = <remote>" config option is
> not very future proof. If people start using partial clone, it is
> likely that at some point they will need their repo to talk to more
> than one remote that is partial clone enabled and I don't see how such
> a config option can scale in this case.

In the case that they want to talk to more than one
partial-clone-enabled repo, I think that there still needs to be one
"default" remote from which missing objects are fetched. I can think of
a few reasons for that - for example, (a) we need to support commands
that give a nonexistent-in-repo SHA-1 directly, and (b) existing Git
code relies on the ability to fetch an object given only its SHA-1. (a)
can be overcome by forbidding that (?) and (b) can be overcome by an
overhaul of the object-fetching and object-using code, but I don't think
both (a) and (b) will occur anytime soon.