From: Johannes Schindelin <johannes.schindelin@gmx.de>
Date: Mon, 08 Jun 2015 17:08:09 GMT
Subject: Re: [PATCH 2/2] fsck: report errors if reflog entries point at invalid objects
Message-ID: <9adde72489de0b1ad66bd2c4730b18e9@www.dscho.org>
In-Reply-To: <20150608165643.GA6863@peff.net>

Hi Peff,

On 2015-06-08 18:56, Jeff King wrote:
> On Mon, Jun 08, 2015 at 06:00:09PM +0200, Johannes Schindelin wrote:
> 
>> >> I like the idea, but I am a bit uncertain whether it would constitute
>> >> "too backwards-incompatible" a change to make this an error. I think
>> >> it could be argued both ways: it *is* an improvement, but it could
>> >> also possibly disrupt scripts that work pretty nicely at the moment.
>> >
>> > What kind of script are you worried about?
>>
>> I was concerned about scripts that work on repositories whose reflogs
>> become inconsistent for whatever reason (that happened a lot to me in
>> the past, IIRC it had something to do with bare repositories and/or
>> shared object databases).
> 
> I think these repositories are already broken. You cannot run `git gc`
> in such a repository, as it will barf when trying to walk the reflog
> tips during `git repack`.
> 
> We run into this exact situation at GitHub because of our shared object
> databases. Our per-fork repack code basically has to do:
> 
>   if ! git repack ...; then
>     git reflog expire --expire-unreachable=all --all &&
>     git repack ... ||
>     die "ok, it really is broken"
>   fi

Good point. So if I needed any more convincing that Michael's patch is a bug fix (as opposed to a backwards-incompatible change), this did it.

Ciao,
Dscho