Re: netfilter moving to git

[Patrick McHardy <kaber@trash.net>]

Jan Engelhardt wrote:
> On Tuesday 2008-05-13 16:17, Patrick McHardy wrote:
>>>
>>> All of that I have long done in my copy of the git repository,
>>> announced already around mid-February. - git://dev.medozas.de/xtables
>>> (Never mind the repository's name, it's the official SVN-top-off-the-head)
 >>
>> Feel free to send me instructions for repeating these steps,
> 
> while (!done && !looks_good) {
> 	git-filter-branch
> }
> 
> Seriously, there is no automated way to properly populate the Author,
> Committer, Date and Email fields, especially if the practice how it
> is effectively stored in the logs has changed over time.
> http://dev.medozas.de/files/iptables.aux/ is the base scripts
> (running do_rewrite), plus hand fixups (do_rewrite2) where the
> automated tools fail to catch every case.

I don't think its particulary important to fix up the commiter
names (and strip the certificate-mess), it you really need to
know, its easy to find out.

>> but I certainly won't replace our repository by an external one.
> 
> This is the one from Yasuyuki, kept in-sync with the SVN.
> 
> There is no difference in the blobs:
> diff -dpru -x .svn -x .git iptables.svncheckout iptables.git;
> (account for $Id$ expansion)
> 
> or the directory structure:
> cd iptables.svncheckout/;
> for i in `find [a-z]* -type d`; do
> 	diff -dpru <(git ls-tree HEAD $i/) \
> 		   <(cd ../xtables/; git ls-tree HEAD $i/);
> done | less -S

I took the last one from Yasuyuki and resynced it, might have
missed a more current version though.

> Only changes so far are logs to fix obvious spelling mistakes or
> metadata fixups (separation of author--committer), or 
> perhaps just give a better message for the older commits like "argh".
> 
> [something like
> 	for ((n = 0; n < 1244; ++n); do
> 	diff -dpru <(cd iptables.gitified; git log HEAD~"$n"!) \
> 	<(cd xtables; git log HEAD~"$n"!)
> 	done;
> should make it visible]
> 
> Fact is there has been going a lot of work into keeping it in 
> git-style shape. It's alright if you do not trust it, but then
> again, it would be a waste to run with a crippled-metadata repo
> or to do the conversion again.

True, but as a matter of principle, I can't use an untrusted repository.

And anyways, I also don't see anything wrong with simply starting with
a clean history by only importing the latest version, In that case I'd
have no problem using your fixed-up repository as historical reference.