Re: [PATCH v3 2/5] reset: introduce --[no-]refresh option to --mixed

From: Junio C Hamano <gitster@pobox.com>
To: Phillip Wood <phillip.wood123@gmail.com>
Cc: Victoria Dye via GitGitGadget <gitgitgadget@gmail.com>,
	git@vger.kernel.org, derrickstolee@github.com,
	Victoria Dye <vdye@github.com>
Subject: Re: [PATCH v3 2/5] reset: introduce --[no-]refresh option to --mixed
Date: Fri, 18 Mar 2022 10:17:59 -0700	[thread overview]
Message-ID: <xmqqy2179o3c.fsf@gitster.g> (raw)
In-Reply-To: <80a5f411-2ddb-177f-4139-1dfff436deda@gmail.com> (Phillip Wood's message of "Fri, 18 Mar 2022 11:08:31 +0000")

Phillip Wood <phillip.wood123@gmail.com> writes:

> This keeps the current behavior of not refreshing when --quiet is
> given. I wonder how disruptive it would be to take the opportunity to
> get rid of that hack and go back the the original behavior of
> refreshing when --quiet is given. There are a couple of assumptions
> that make me think it might be acceptable
>
> 1 - anyone using a sparse index wont notice as refreshing the index
>     should be fast for them
>
> 2 - the large repositories that are affected exist in managed
>     environments where an admin who reads the release notes could set
>     reset.refresh in a central config so individual users are not
>     inconvenienced.

I would very much prefer to see "--quiet" not making contribution to
the decision to refresh or not in the longer term.  Many plumbing
commands expect that the calling scripts refresh the index with an
explicit use of "update-index --refresh" and leave the index not
refreshed, but working on unrefreshed index is a trade-off between
performance and correctness.

 * Turning "--quiet" not to refresh may incur performance regression
   for shorter term.  It will not hurt correctness.

 * Introducing "--no-refresh" to mark "reset --quiet" invocations,
   where the freshness of the index does not matter for correctness,
   would help regain performance without breaking scripts.  All
   "reset --quiet" invocations in scripts written before this series
   are supposed to be safe (as they lived with their "reset --quiet"
   that does not refresh), but newly written scripts may start
   expecting that "reset --quiet" would refresh for correctness.

 * If we allow reset.refresh to be set to "no", however, that would
   affect _all_ uses of "reset --quiet", including the ones in newly
   written scripts that expect "reset --quiet" to refresh.  They
   would be forced to say "reset --quiet --refresh", just in case
   the user has such a configuration; otherwise these scripts will
   be declared "buggy" for not explicitly saying "--refresh".

I do not think reset.refresh is a good idea, but I very much like
the idea to making "reset" (regardless of "--quiet") to refresh by
default.

Thanks.