From: Junio C Hamano <junkio@cox.net>
Date: Tue, 22 Nov 2005 17:46:18 GMT
Subject: Re: [PATCH 4/6] Add check_repo_format check for all major operations.
Message-ID: <7vd5ksefth.fsf@assigned-by-dhcp.cox.net>
In-Reply-To: <200511221555.24572.matlads@dsmagic.com>

Martin Atukunda <matlads@dsmagic.com> writes:

>> that call in the current series yet.  Even if you had, this does
>> not feel quite right to me.
>
> would something like the following apply in this case: (totally untested :)

Yes, although that is exactly what I said "this does not quite
feel right" ;-).  Is it hard to arrange things so that a process
does exactly one check_repo_format() during its lifetime?

Two more issues I've been thinking about:

1. The core.repositoryformatversion scheme assumes and relies on
   that at least .git/config would stay forward compatible.  But
   if you cover unconditionally the three main entry points, I
   suspect "git-var" or "git-config-set --get" would stop
   working in a wrong repository.  I think the scripts and
   Porcelains need a way to check if we are on a repository from
   the correct vintage before running other low-level commands,
   so one of these commands probably needs to be made to work
   without check_repo_format() dying. Or we could introduce a
   new command 'git-check-repo-format' for this specific
   purpose, and special case only that one.

2. Some commands are read-only and are handy for problem
   diagnosis (e.g. cat-file), so it _might_ make sense to allow
   them to attempt running in a newer repository (which may well
   fail due to repository format difference).  I am not sure
   about the merit of doing that outweighs the complexity,
   though.  What you did covers _everything_ uniformly, and
   certainly is simpler, easier to explain, and nicer.