Re: ps/receive-use-only-advertised

From: Junio C Hamano <gitster@pobox.com>
Date: Nov 18, 2022, 23:27 UTC
Message-ID: <xmqqmt8nkfmx.fsf@gitster.g>
In-Reply-To: <Y3PMLFuvR6hlIzbq@coredump.intra.peff.net>

Jeff King <peff@peff.net> writes:

Show 19 quoted lines

>> > >  "git receive-pack" used to use all the local refs as the boundary
>> > >  for checking connectivity of the data "git push" sent, but now it
>> > >  uses only the refs that it advertised to the pusher.  In a
>> > >  repository with the .hideRefs configuration, this reduces the
>> > >  resource needed to perform the check, and also forces the pusher to
>> > >  prove they have all objects that are necessary to complete the
>> > >  history on top of only the history available to them.
>> >
>> > We have at a later point established that this is not true: the pusher
>> > does not have to prove they have all objects. As the `--not --all` part
>> > in git-rev-list(1) is merely an optimization the semantics aren't
>> > changed at all
>> 
>> Thanks; this is Junio's description from when he picked up the topic
>> back in WC #08 from October. I'll need to update before graduating it
>> down.
>
> Yeah, I think you can probably just remove everything from "and also
> forces..." on.

Hmph, --not --all is a way to make the rev-list error out if the proposed new tips are not connected with existing objects to the "existing" refs, and I thought that removing "hidden" ones from "--all" means that the pusher is now forced to push objects all needed. But you are right. Excluding the hidden refs from "--all" while doing the "rev-list new-tip --not --all" traversal does not exclude pre-existing objects that are only reachable from these hidden refs in the object store, so it does not help ensure that the pusher has to prove they have these objects.