From: Jeff King <peff@peff.net>
Date: Sun, 22 Mar 2020 19:22:34 GMT
Subject: Re: [PATCH 0/2] Slightly simplify partial clone user experience
Message-ID: <20200322192234.GC671552@coredump.intra.peff.net>
In-Reply-To: <5baf9f4c-9ce0-0384-1113-431038c2a646@gmail.com>

On Sun, Mar 22, 2020 at 12:45:29PM -0400, Derrick Stolee wrote:

> >> I have mixed feelings on this. I do like making things less arcane for
> >> users. But are we locking in a behavior for --partial that we might not
> >> want to live with forever? I.e., the current thinking for partial clones
> >> is to fetch no blobs at all, get all commits and trees, apply sparse
> >> filters, and then fault in the blobs we need. But imagine we later grow
> >> the ability to easily avoid fetching all of the trees. Would we regret
> >> having the simple name "--partial" taken?
> > 
> > I agree with that. Something like "--filter-blobs" for
> > "--filter=blob:none" and perhaps "--filter-blobs=<size>" for
> > "--filter=blob:limit=<size>" might be worth it though.
> 
> Thanks for the perspective on this. The --filter-blobs[=<size>] should
> be less likely to collide with an alternative definition of "partial".

Yes, though I wonder if it's really that big an improvement over
"--filter=blob:none".

> While we are thinking in this space, what if we had a "partial-clone"
> builtin? It could be a light wrapper around "git clone" where
> 
>   git partial-clone [--limit=<size>] [options] <url> [<dir>]
> 
> would do the same thing as
> 
>   git clone --filter=blob:[none|limit=<size>] [options] <url> [<dir>]
> 
> Just spit-balling here.

I think that just shifts my concern out to the other command; i.e.,
would we later want the partial-clone command to behave differently.
And introducing a wrapper creates a weird non-orthogonality where it's
not necessary. I _do_ think a wrapper could make sense when it's doing
substantially different things, or multiple steps that don't belong in
the original program (e.g., the way Scalar works).

> In the meantime, I'll work to adjust my patches to only be the
> documentation of the --filter option.

Thanks, that seems like an obvious and easy improvement.

-Peff