[darcs-users] ANNOUNCE: darcs-2.4.3 (update and changes)

Eric Kow kowey at darcs.net
Wed May 12 09:33:33 UTC 2010


On Sun, May 09, 2010 at 22:46:57 +0100, Eric Kow wrote:
> The darcs team would like to announce the immediate availability of darcs
> 2.4.3

> The two most important issues resolved were
> that Darcs would fail on Windows with an "mmap of _darcs/index", and
> that darcs get --partial would produce inconsistent repositories.

Just a recap and status update.

Windows users: I'm really sorry, but it looks like you'll have to
continue using Darcs 2.3.1 for now.

1. We've found a bug in Darcs (when built with GHC 6.12) where trying to
   push or put patches over via SSH will fail due to unexpected CRLF to
   LF conversions.
   
   The good news here is that darcs apply's insistence on checking patch
   bundle hashes prevents these from being wrongly applied.  We're
   working to resolve the issue.

2. Salvatore reports a build failure with GHC 6.10 (if I understand
   correctly).

Now our release process has been tied to the regression testing suite
for a while (we have a release.sh script which automates the tarball
creation and which aborts of the tests fail), but we haven't done a good
enough job at improving coverage in terms of platform.

We're taking the lessons from these recent pains to heart.

* Coordination between Issue and Release Managers: The 2.4.1 release
  has taught us to improve the coordination between our Issue and
  Release Managers.  Things are slightly more formal now, in that we
  need a green light from the Issue Manager to let us know that all
  bugs marked for the release have been accounted for.

* Coordination between Release Manager and packagers: The 2.4.3
  release has taught us to improve the coordination between the
  Release Managers and packagers.  This release was too hurried on my
  part.  I was too eager to get something out the door to replace the
  broken 2.4.1 and I didn't give volunteers nearly enough time to test
  the release.  The lesson is that the more we rush, the slower we get
  things done.  So we're going to make a point of slowing down.

  For future releases, we'll quietly put out a *final* tarball for
  packagers one week before the announcing the release.  This will be
  an official release in the sense that if anything should go wrong,
  we'll have to go through the process of bumping the versions.  I
  think that giving the packagers some extra time will buy us a bit
  of insurance and will also allow us to deliver binaries at the same
  time as source packages.

* More build slave coverage: as you saw in Petr's last message, we've
  now improved our buildbot coverage so that we're systematically
  covering the last two GHC releases [according to our GHC support
  policy] for each of the three major platforms (2 x 3 = 6 build
  slaves).

* Automated testing of release branch: We've been using buildbot to
  automatically test the main darcs repository for a while now.
  Thanks to Petr's hard work, we've extended our automated testing
  to the release branch as well.  This ensures that for each release
  we will have run the regression testing suite on the 6 key platforms
  above.

Sorry for this, everybody!
We're still learning, unfortunately, the hard way sometimes.


-- 
Eric Kow <http://www.nltg.brighton.ac.uk/home/Eric.Kow>
PGP Key ID: 08AC04F9
-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: application/pgp-signature
Size: 198 bytes
Desc: Digital signature
URL: <http://lists.osuosl.org/pipermail/darcs-users/attachments/20100512/fd5d88bc/attachment.pgp>


More information about the darcs-users mailing list