Re: CVS Re: [Simulavr-devel] (No subject header)

[Joerg Wunsch]

(Joel: are you around and listening?)

As address@hidden wrote:

> >I do hope some of you do have a CVS backup, do you?
> 
> My most recent intact copy is from March.

Is that a CVS repository backup (through rsync), or just a checked out
local copy?

In case nobody has ever taken CVS repository backups (which I'm afraid
of), I think Joel is likely to have the most recent checked out
working copy.  At least, he's been the last one who committed
something.  If we have to fall back to that copy, this means to lose
(some) CVS history though.

The 20090529 partial backup has proven to be severely incomplete for
the avr-libc and avrdude projects.  For simulavr, I can immediately
see it's lacking all the data in the simulavr/ part of the repository
(i.e., the historic code base), while simulavrxx/ shows at least some
data.  The historic simulavr code is not a very big deal though since
this part of the CVS repository hasn't been touched for a while, so
it's fine to use the (known to be complete) 20050429 backup for it.

I'd like to suggest the following recovery strategy:

. fetch rsync://cvs.sv.gnu.org/sources/backup-20090429/simulavr into a
  local directory, say /tmp/simulavr-cvs

. fetch rsync://cvs.sv.gnu.org/sources/backup-20090529-incomplete/simulavr
  on top of it

. the resulting /tmp/simulavr-cvs can now be used as a local CVS
  repository to work against

. copy the existing latest checked out working directory into a safe
  place, say /tmp/simulavr-recovery

. cd /tmp/simulavr-recovery

. cvs -d /tmp/simulavr-cvs update -Pd

If this doesn't make CVS yell about versions lacking in the
repository, this likely means that /tmp/simulavr-cvs is indeed a valid
(and up to date) repository for the project, so it can be filed as a
support request on savannah for recovery.  See

https://savannah.gnu.org/forum/forum.php?forum_id=5828

for the procedure.

For a cross check, go to a different directory, run a cvs checkout
against /tmp/simulavr-cvs, and compare the result against the old
working copy using diff -ur.  Except for the files under CVS/, there
shouldn't be any difference.

(In my case, the working copy was pretty old, so the above did
correctly update everything, and did not complain about missing
versions.)

If the above procedure yields any complaint from CVS about lacking
versions, I think it's safer to proceed with the 20090429 backup on
savannah then (again, we'd have to file a support request for that).
Once that repository backup is in place, the developer with the most
recent working copy (Joel?) could use it, and check out a pristine
working copy out of that one.  Then, copy over the most recent working
copy modulo the contents of its CVS subdirectories.  Check using "cvs
diff -u", the differences ought to be everything that has been
committed in 2009-05.  Go ahead, and commit that one back into the
recovered savannah repository as a single large commit.  For the
commit messages, I could provide the person who's going to perform
that commit a complete listing of all CVS commit messages that have
been sent to the simulavr-commit mailing list in May, 2009.

-- 
cheers, J"org               .-.-.   --... ...--   -.. .  DL8DTL

http://www.sax.de/~joerg/                        NIC: JW11-RIPE
Never trust an operating system you don't have sources for. ;-)