[petsc-dev] Branch 'next' has been rewound

[Jed Brown]

== Users ==

I have rewound the branch 'next' after the release, so it will
not fast-forward (just this once).  If you have a local 'next' branch,
you should abandon it and check out a fresh one.  You can do this with

  $ git checkout master
  $ git branch -D next   # delete the old branch
  $ git pull
  $ git checkout next    # revisit 'next' if you want

If you are tracking 'master', 'git pull' will be sufficient.


== Developers ==

With v3.4 tagged, we can decide what to do with the branches left over
from 'next' (now in a temporary 'next-pre-3.4' that will be deleted).
If they should be rebased, it's best to do it now.

  $ git checkout my/branch
  $ git rebase v3.4
  $ git push -f --dry-run # should show force-update to your branch only, then remove '--dry-run'

If you like everything in the branch, but there are conflicts, you can
sync with v3.4 now [1].

  $ git checkout my/branch
  $ git merge v3.4
  $ git push


Here is a status summary for the branches that used to be in 'next':


e8577e4 2013-04-10 (karlrupp/feature-viennacl) ViennaCL: fix LOCDIR

  This has a simple conflict with 'knepley/remove-sieve' (below) in
  src/vec/vec/interface/vecregall.c

b61a8fb 2013-03-15 (klaij/fieldsplit-simple-ex70) SNES ex70.c: free array returned by PCFieldSplitGetSubKSP()

  This example highlights a bug/feature request in fieldsplit.  If it
  can be handled without interface changes, it should go to 'maint'.

a05fad5 2013-05-12 (knepley/remove-sieve) DMPlex: remove bad "DMMESH" documentation

  This branch is a bit chaotic and could reasonably be rebased and have
  the incomplete commits squashed together.

1cf896f 2013-03-22 (knepley/solkx) SNES ex75: Added Maple verification test, and added preliminary FEM test

  This branch contains one commit (a9ffcdab03f) that is unrelated and
  made it into v3.4 in a different form (3e298deff27), so rebasing would
  make it cleaner.

aee7f6d 2013-05-08 (prbrune/snes-npcjacobianlagging) Tests for persistent lagging of the Jacobian or Preconditioner

  This made changes to src/snes/examples/tutorials ex15_4 that conflict
  with 'prbrune/snes-npcrhsfix' (merged 2013-05-03).  It should either
  be rebased or it should merge from v3.4.


The remaining commits in 'next' have either been rebased (Karl
accidentally rebased 'karlrupp/feature-viennacl' after merging it to
'next') or are commits that were made in an inappropriate place and were
later cherry-picked back to the proper branch.

ee5caa0 Revert "MPIUNI: Now it returns error codes correctly"
5b3707c Revert "DMPlex: fix initialization of variables in degenerate case"
607f2c0 DMPlex: fix initialization of variables in degenerate case
2786d98 Repair bad merge of 'barry/rm-xxxregisterdynamic'
2233159 ViennaCL: Updated to new AMS viewer model
ec89c54 ViennaCL: Fixed includedir in viennacl.py package handler
3cdc909 ViennaCL: Applied const-correctness to functions.
9d766b4 ViennaCL: Eliminated whitespace violations, removed/fixed erroneous comments.
c21fe5b ViennaCL: Device type can now be specified via command line.
67575d3 ViennaCL: Now catching the correct type of exceptions (std::exception)
19900a9 ViennaCL: Eliminated dead/unused code
c2cfb21 ViennaCL: Fixed issues if size 0 is passed.
bb99cb5 ViennaCL: Setting VIENNACL_WITH_OPENCL flag to enable OpenCL backend as default.
4b3d4ac ViennaCL: Added package detector for OpenCL, added OpenCL-dependency for ViennaCL.
9329d19 ViennaCL: Added MPI-implementation for AIJ matrices.
19b9568 ViennaCL: Removed seqviennacl/aijassemble.cxx (would trigger error, now uses existing fallback)
dbcb10b ViennaCL: Renamed files to fulfill conventions (.cpp -> .cxx, all lowercase), removed restriction of C++ everywhere.
c65bdabe ViennaCL: Added sequential AIJ matrix.
8bf5113 ViennaCL: Added MPI vectors.
bff62fe ViennaCL: Removed unnecessary call to resize() at vector construction.
1737cda ViennaCL: Download of package now working.
8034546 ViennaCL: First draft, including all operations for sequential vectors.
8de993d Bug fix for MatLoad_MPIBAIJ() buffer size for rank zero read was wrong size; simplified code as well MatLoad_MPIAIJ() also This fix can also be ported back to 3.3



[1] You may remember that we normally ask branches to avoid merging from
upstream.  A release is an exception to this: the release should be at
least as stable as the arbitrary place that you started working from,
and it's a well-defined point, so it's perfectly acceptable to merge
v3.4.