Re: git copies of cvs modules available

[Thor Lancelot Simon <tls%panix.com@localhost>]

On Wed, Jan 13, 2010 at 11:43:59AM -0500, Arnaud Lacombe wrote:
>
> and CVS is no better, even worth as you'll only know the corruption
> months later because someone tried to build a corrupted file. And
> wait, it's distributed, you've got hundreds of backup of your repo all
> over the world for free ! All that provided that you cannot trivially
> find SHA1 collision.
> 

I'm sure that was fun to write, but the problem with it is that it's
empirically false.  We have a CVS repository, as you know, which is
over 15 years old and has millions of revisions in it.  I can assure
you the systems on which it is hosted have crashed at least dozens,
and possibly hundreds, of times in those 10 years.

In that time, for about half of which an rsync mirror of the repository
has been publically available, we have discovered only two (possibly
three) instances of corruption to the ,v files due to server crash.

It's possible we missed some, but it's spectacularly unlikely we missed
enough to put the failure rate anywhere near what we've observed with
git after a system crash, since we would have had to miss _thousands_
of such corruptions for that to be the case.

In other words, nice theory, but it doesn't square up to the facts
very well at all.

-- 
Thor Lancelot Simon                                    
tls%rek.tjls.com@localhost
  "All of my opinions are consistent, but I cannot present them all
   at once."    -Jean-Jacques Rousseau, On The Social Contract