Re: git push mirror mode V4 (replacement stack)
- From
Andy Whitcroft <apw@shadowen.org>
- Date
- Nov 12, 2007, 11:00 UTC
- Message-ID
- <20071112110016.GG301@shadowen.org>
- In-Reply-To
- <20071109233041.GC301@shadowen.org>
On Fri, Nov 09, 2007 at 11:30:41PM +0000, Andy Whitcroft wrote:
> Following this mail is a complete replacement git push mirror mode > stack (V4). It folds down all the various patches into a logical > sequence (thanks Dscho). This stack passes the entire test suite, > and I have been using the same code for real work here.
Ok, I have spotted one oddity with this feature. The symbolic refs are getting converted to real refs in the mirror. Generally speaking this is the <remote>/HEAD refs but I guess it may be possible to have others.
I have had a looked about and I am actually confused as to whether we maintain remote symbolic refs at all? Cirtainly git-clone.sh seems to do some hoop jumping, comparing the sha1's of all of the fetched branches and replacing the HEAD reference with a symbolic reference should it find a match.
I am unsure if this a huge problem or not. Its not preventing me using it as an effective mirror, but if we assume one is making the mirror as a backup, then there would be slightly more than an rsync to convert the the repo back into your original, though I guess there already is as you would want to insert your config into the remote also.
Perhaps someone with a better understanding could point me to where we we maintain these refs, if we indeed do?
-apw