From: Maciej Ciemborowicz <maciej.ciemborowicz@gmail.com>
Date: Wed, 23 Sep 2026 12:49:31 GMT
Subject: Re: [PATCH v4 0/3] refs: report old OIDs for batched deletions
Message-ID: <20260923124931.39424-1-maciej.ciemborowicz@gmail.com>
In-Reply-To: <CAOLa=ZTN1TU2A1sgEhiw=ymMYr6Ge11cMEubSaeKqr4WNU=2EQ@mail.gmail.com>

Hi Junio,

> Can we avoid REF_TRANSACTION_FLAG_SKIP_HOOK by changing the call chain
> instead of exposing a general mechanism for skipping hooks?

I looked into that, but I do not think it removes the distinction
cleanly. The packed-refs transaction is an internal, physical part of
one logical files-backend update: it needs the normal transaction
machinery, while the reference-transaction hook must observe only the
outer logical update.

Restructuring the call chain would either duplicate the
prepare/finish/abort lifecycle or hide the same hook-suppression
decision in a less explicit helper. I do agree that this should not be
a public escape hatch, though. I will keep it internal to the refs
implementation and limit its use to these internal transactions.

> Should the int copy parameter simply be bool copy, given that the
> function has only two modes?

Yes, agreed. It is a two-state mode, so bool copy expresses the
contract more clearly. I will change it in the next reroll.

Thanks,
Maciej