org-mode/README_maintainer

# -*- mode:org -*-

#+title: Maintainer tasks
#+startup: noindent

This document describes the tasks the Org-mode maintainer has to do
and how they are performed.

* Working with patchwork

John Wiegley is running a patchwork server that looks at the
emacs-orgmode mailing list and extracts patches.  The maintainer and
his helpers should work through such patches, give feedback on them
and apply the ones which are good and done.  A task for the maintainer
is to every now and then try to get old stuff out of that list, by
asking some helpers to investigate the patch, by rejecting or
accepting it.

I have found that the best workflow for this is using the pw script by
Nate Case, with the modifications for Org-mode made by John Wiegley
and Carsten Dominik.  The correct version of this script that should
be used with Org mode is distributed in the =mk/= directory of the
Org mode distribution.  Here is the basic workflow for this.

** Access to the patchwork server

If you want to work on patchwork patches, you need write access at the
patchwork server.  You need to contact John Wiegley to get this
access.

There is a web interface to look at the patches and to change the
status of patches.  This interface is self-explanatory.  There is also
a command line script which can be very convenient to use.

** Testing patches

To start testing a patch, first assign it to yourself

: pw update -s "Under Review" -d DELEGATE-NAME NNN

where =NNN= is a patch number and =DELEGATE-NAME= is your user name on
the patchwork server.

The get the patch into a branch:

: pw branch NNN

This will create a local topic branch in your git repository with the
name =t/patchNNN=.  You will also be switched to the branch so that
you can immediately start testing it.  Quite often small amends need
to be made, or documentation has to be added.  Also, many contributors
do not yet provide the proper ChangeLog-like entries in the commit
message for the patch.  As a maintainer, you have two options here.
Either ask the contributor to make the changes and resubmit the patch,
or fix it yourself.  In principle, asking to contributor to change the
patch until it is complete is the best route, because it will educate
the contributor and minimize the work for the maintainer.  However,
sometimes it can be less hassle to fix things directly and commit the
changes to the same branch =t/patchNNN=.

If you ask the contributor to make the changes, the patch should be
marked on the patchwork server as "changes requested".

: pw update -s "Changes Requested" -m "What to change" NNN

This will send an email to the contributor and the mailing list with a
request for changes.  The =-m= message should not be more than one
sentence and describe the requested changes.  If you need to explain
in more detail, write a separate email to the contributor.

When a new version of the patch arrives, you mark the old one as
superseded

: pw update -s "Superseded" NNN

and start working at the new one.

** Merging a final patch

Once the patch has been iterated and is final (including the
ChangeLog-like entries in the commit message), it should be merged.
The assumption here is that the final version of the patch is given by
the HEAD state in the branch =t/patchNNN=.  To merge, do this:

: pw merge -m "maintainer comment" NNN

This will merge the patch into master, switch back to master and send
an email to both contributor and mailing list stating that this change
has been accepted, along with the comment given in the =-m= message.

At some point you might then want to remove the topic branch

: git branch -d t/patchNNN

* Releases

** Main releases

The release number for main releases look like this:  =7.13=

Main releases are made whenever Org is in a state where the feature
set is consistent and we feel that the features that are implemented
is something we want to support in the future.

A major release turns the current state of the master branch into a
release.  The release process is a single make command:

: make release TAG=7.13

Before issuing this command, you should make sure that everything
during the process will work right, you can do so by running

: make testrelease TAG=7.13

When this fails, make sure to clean up.  =git reset --hard= if
necessary, and check if there are unwanted files, directories, or
branches left over from the testing.

** Minor releases

The release number for minor releases look like this:  =7.13.01=

Minor releases are small amends to main releases.  Usually they fix
critical bugs discovered in a main release.  Minor bugs are not
fixed - they will be adressed in the next main release.  Only the fix
to the bug is bundled into a release, without the main development
work going on in the master branch.  Since the bug fix will also be
needed in the master branch, usually the fix is made in master and
then cherry-picked into maint.  When this is done, a release is made
from maint with this command:

: make fixrelease TAG=7.13.01

** Updating release files on orgmode.org server

As of 2011-01-15, these directives of the Makefile are meant to be
used /from orgmode.org server/ and will copy the release files to the
webserver directory.

- ~$ make makerelease :: creates a =RELEASE/= directory containing
     manuals and release files (=org.tar.gz=, =org.zip=, etc.)

- ~$ make sync_release :: copy the content of =RELEASE/= to the right
     location on the server

- ~$ make sync_manuals :: copy the manuals from =doc/= to the right
     location on the server

- ~$ make relup :: call the three directives described above.

** Between releases

While working on master between releases, I used to use something like
7.02trans as the version string.  I no longer do this.  =M-x
org-version= will spit ut complete version infor related to git, with
the nearest commit and tag.  I you ever need to set a special version
number, use this:

: mk/set_version 7.02trans

and commit the result.  Note that the above command does not change
the version string in the file from which Org's homepage is generated.
To change that as well, you would use a =--all= flag.  To change only
this file, use =--only=.

* Synchonization with Emacs

This is still a significant headache.  Some hand work is needed here.

Emacs uses bzr.  A useful introduction to bzr for Emacs developers can
be found [[http://www.emacswiki.org/emacs/BzrForEmacsDevs][here]].  While I see all the advantages this would have, I
cannot bring myself to switch away from git for my day-to-day work,
because I know git so well, and because git seems to me as being much
more powerful, conceptionally simple (once you have [[http://newartisans.com/2008/04/git-from-the-bottom-up/][bent your head
around it]]), and so much faster.

So the way I have been doing things with Emacs is this:

1. I do not update the version in Emacs too often.  Just once every
   few months - this is frequently enough for the Emacs release cycle.
   Care must be taken to get in a *new and stable* version shortly
   before Emacs goes into feature freeze and pretest, because that
   version is going to be in the wild for a long time.

2. I watch the Emacs diffs for changes made by the maintainers of
   Emacs in the org-mode files in Emacs.  Any changes that come up
   there, I merge into the development version of Org-mode.
   Occasionally I do not do this, if I do not agree with a change.
   The changes go into Org /without/ a ChangeLog-like entry in the
   commit message.  The reason for this is that we will later generate
   a ChangeLog file from our commit messages, and I do not want double
   ChangeLog entries in the Emacs ChangeLog file.

3. When I have made a release (usually I wait for the minor releases
   to stabilize), I *copy* org files into the Emacs repository.  Yes,
   I do not merge, I copy.  This has been the source of some problems
   in the past - Emacs developers are not happy when I accidentally
   overwrite changes they made.  But I have not had the patience to
   work out a better mechanism, and I really dislike the idea that the
   version in Emacs starts diverging from my own.

   Careful: Copy /org.texi/ and /orgcard.tex/ into the right places,
   and also copy the lisp files with *two exceptions*: Do *not* copy
   /org-colview-xemacs.el/ and /org-install.el/.  The former does not
   belong in Emacs.  And the latter would actually be harmful because
   Emacs generates its own autoloads.  The Emacs distribution contains
   an empty /org-install.el/, so that users can have =(require
   'org-install)= in .emacs with no ill effects.  So if you were to
   copy /org-install.el/, you would overwrite that empty placeholder
   file.

4. Generate the ChangeLog entries

   For this, I do in the org-mode git repository

   : mk/make_emacs_changelog release_7.02.05..release_7.03.02

   This will spit out ChangeLog entries (for the given commit range)
   that need to go into the ChangeLog files in Emacs.  Org-mode
   contributes to 3 different ChangeLog files in Emacs:

   : lisp/org/ChangeLog    (for lisp changes)
   : doc/misc/ChangeLog    (for org.texi changes)
   : etc/ChangeLog         (for refcard changes)

   When you run the =make_emacs_changelog= program, you will be
   prompted for a date in ISO format YYYY-MM-DD, this date will be
   used in the ChangeLog entries - Emacs developers want these dates
   to be the time when the change has been installed into Emacs, not
   the time when we made the change in our own repository.  So all the
   ChangeLog entries will get the same date.  You will also be
   prompted for the kind of ChangeLog you want to make, possible
   answers are =lisp=, =texi=, and =card=.  The program will then
   select the correct entries for the specified ChangeLog file.  If
   you don't like being prompted, you can give the date and type as
   second and third command line arguments to =make_emacs_changelog=,
   for example

   : mk/make_emacs_changelog release_7.02.05..release_7.03.02 2010-12-11 lisp

   These entries need to be added to the ChangeLog files in Emacs.
   You should, in the ChangeLog file, select the inserted region of
   new entries and do =M-x fill-region=, so that the entries are
   formatted correctly.  I then do look through the entries quickly to
   make sure they are formatted properly, that the email addresses
   look right etc.

5. Commit the changes into the bzr repository and you are done.  Emacs
   developers often look throught the commit and make minor changes -
   these need to be merged back into our own repo.

* Updating the list of hooks on Worg

  The file /org-configs/org-hooks.org/ contains a list of all hooks in
  Org.  This list has to be updated after hooks have been added or
  removed.  The perl script =mk/list-hooks.pl= creates the
  entire section "Hooks and Function variables", including its
  level-one headline.  I guess babel code could be used to update this
  automatically, but I have not implemented this - I have been doing
  it by hand every few months.

* Copyright assignments

  The maintainer needs to keep track of copyright assignments.  Even
  better, find a volunteer to do this.  

  The list of all contributors from who we have the papers is kept on
  Worg at http://orgmode.org/worg/org-contribute.php, so that
  committers can check if a patch can go into the core.

  The assignment process does not allways go smoothly, and it has
  happened several times that it gets stuck or forgotten at the FSF.
  The contact at the FSF for this is: copyright-clerk@fsf.org 

  Emails from the paper submitter have been ignored in the past, but
  an email from me (Carsten) as the maintainer of Org mode has usually
  fixed such cases within a few days.