/pkg-ssh/openssh/trunk

Debian OpenSSH source package handling

======================================

The Debian package of OpenSSH is maintained in Bazaar

(http://bazaar-vcs.org/, or the 'bzr' package in Debian). You will need at

least version 1.16.1; the version in Debian testing as of the time of

writing (2009-12-21) is fine, or you can use the version in lenny-backports.

URLs are as follows:

  Anonymous branch: http://anonscm.debian.org/bzr/pkg-ssh/openssh/trunk

  Web browsing:     http://anonscm.debian.org/loggerhead/pkg-ssh/openssh/trunk

  Authenticated, for developers with commit access only:

                    bzr+ssh://bzr.debian.org/bzr/pkg-ssh/openssh/trunk

Although it's possible that I may use something like bzr-loom in the future

to better manage things like the Kerberos/GSSAPI patch, right now there's no

funny business and all that developers need to do is:

  # To check out:

  bzr co bzr+ssh://bzr.debian.org/bzr/pkg-ssh/openssh/trunk openssh

  # To update:

  bzr up

  # To edit:

  # hack hack hack, and 'bzr add' any new files

  debcommit # or bzr commit

  # note that this pushes automatically; you can use 'bzr unbind' to

  # temporarily prevent this, or 'bzr branch' to create a local branch which

  # you can merge later

  # To release:

  dch -r && debcommit -r

If you have lots of branches, you'll probably want to use a shared

repository to save space. Run 'bzr init-repo .' in an ancestor directory of

all your OpenSSH working directories. For example, I have a shared

repository in ~/src/debian/openssh/, upstream checkouts in

~/src/debian/openssh/upstream/, and my own working trees in

~/src/debian/openssh/trunk/.

Patch handling

--------------

This package uses quilt to manage all modifications to the upstream source.

Changes are stored in the source package as diffs in debian/patches and

applied automatically by dpkg-source when the source package is extracted.

To configure quilt to use debian/patches instead of patches, you want either

to export QUILT_PATCHES=debian/patches in your environment or use this

snippet in your ~/.quiltrc:

    for where in ./ ../ ../../ ../../../ ../../../../ ../../../../../; do

        if [ -e ${where}debian/rules -a -d ${where}debian/patches ]; then

                export QUILT_PATCHES=debian/patches

                break

        fi

    done

After unpacking the source package, all patches will be applied, and you can

use quilt normally.

If you check out the source code from bzr, then all patches will be applied,

but you will need to inform quilt of this manually.  Do this by running:

    debian/rules quilt-setup

To add a new set of changes, first run quilt push -a, and then run:

    quilt new <patch>

where <patch> is a descriptive name for the patch, used as the filename in

debian/patches.  Then, for every file that will be modified by this patch,

run:

    quilt add <file>

before editing those files.  You must tell quilt with quilt add what files

will be part of the patch before making changes or quilt will not work

properly.  After editing the files, run:

    quilt refresh

to save the results as a patch.

Alternately, if you already have an external patch and you just want to add

it to the build system, run quilt push -a and then:

    quilt import -P <patch> /path/to/patch

    quilt push -a

(add -p 0 to quilt import if needed). <patch> as above is the filename to

use in debian/patches.  The last quilt push -a will apply the patch to make

sure it works properly.

To remove an existing patch from the list of patches that will be applied,

run:

    quilt delete <patch>

You may need to run quilt pop -a to unapply patches first before running

this command.

You should only commit changes to bzr with all patches applied, i.e. after

'quilt push -a'.

Merging new upstream releases

-----------------------------

(Most developers will not need to read this section.)

Thanks to the import from Portable OpenSSH CVS provided by Launchpad

(https://code.launchpad.net/~vcs-imports/openssh/main, accessible by the

shortcut 'lp:openssh' from the bzr client), the Debian branch is a true DVCS

branch from upstream. This is a worthwhile property, but preserving it does

take a little bit of work.

Launchpad only imports CVS HEAD, but upstream sometimes produces releases

from a branch. We use the same software used by Launchpad to import the

branch as well, but a few small hacks are necessary to do good branch

imports. In Bazaar, it's important that the same file in different branches

should have the same file-id, otherwise merge attempts will try to delete

and re-add the file which usually doesn't work out very well. Occasionally a

file is added to CVS HEAD and then also added to a branch, and cscvs isn't

quite smart enough to spot this and copy over the file-id. We need to help

it out.

To fetch the necessary code:

  bzr branch lp:~cjwatson/launchpad-cscvs/openssh-branch-imports

  # or 'bzr pull' in the appropriate directory to update this, if you

  # already have a copy

To import a branch, V_5_3 in this example:

  export PATH="/path/to/cscvs/openssh-branch-imports:$PATH"

  export PYTHONPATH=/path/to/cscvs/openssh-branch-imports/modules:/path/to/cscvs/openssh-branch-imports

  # in a CVS checkout of :ext:anoncvs@anoncvs.mindrot.org:/cvs module

  # openssh:

  cscvs cache -b

  # or 'cscvs cache -u' if you've done this before and want to update

  cvs up -rV_5_3

  # Now we need to get a few bits of information from cscvs' cache.

  sqlite CVS/Catalog.sqlite

  sqlite> select csnum,log from changeset where branch = 'V_5_3' order by startdate;

  # There will be a solid block of "Creation of branch V_5_3" changesets at

  # the start; look for the first revision *after* this. Substitute this in

  # the following wherever you see "CSX".

  sqlite> select revision,filename from revision where branch = 'V_5_3' and csnum >= CSX and revision not like '%.%.%' order by filename;

  # Anything listed here will need to be added to the openssh_ids dictionary

  # in modules/CVS/StorageLayer.py in cscvs. Please send Colin Watson a

  # patch if you do this.

  # Next, look up the branchpoint revision in the main bzr import (bzr

  # branch lp:openssh). It's usually easiest to just look it up by commit

  # message and double-check the timestamp. Substitute this revision number

  # for "BPR" in the following. /path/to/openssh/main is wherever you've

  # checked out lp:openssh.

  bzr branch -rBPR /path/to/openssh/main /path/to/openssh/5.3

  # If you're using Bazaar signed commits with a GPG agent, make sure that

  # your agent has seen your passphrase recently. Now you can start the

  # actual import!

  cscvs -D4 totla -SC V_5_3.CSX: /path/to/openssh/5.3

  # If this fails at the end with a "directories differ" message, you may

  # have forgotten to switch your CVS checkout to the appropriate branch

  # with 'cvs up -r...' above. Otherwise you'll have to debug this for

  # yourself. It's also worth double-checking that any files added to the

  # branch have file-ids matching those on the trunk, using 'bzr ls -R

  # --show-ids'.

Now we have a Bazaar branch corresponding to what's in CVS. Previous such

branches are available from Launchpad, for reference purposes:

  https://code.launchpad.net/openssh

However, upstream releases involve a 'make distprep' step as well to

construct the tarball, and we need to import the results of this as well to

get a clean package.

Start by unpacking the upstream tarball (remember to check its GPG signature

first!). Copy the .bzr directory from the upstream branch you prepared

earlier. Now we have another branch, but with a working tree corresponding

to the upstream tarball. Modifications and deletions are handled

automatically, but we need to handle additions explicitly to make sure

file-ids are correct (see above). Run:

  bzr add --file-ids-from=/path/to/openssh/debian/trunk

  bzr st --show-ids

  # compare this with 'bzr ls --show-ids' in the Debian trunk to make sure

  # the result will be mergeable

  bzr ci -m 'Import 5.3p1 tarball'

Add a parent revision for the previous tarball branch, to make it easier for

bzr to compute accurate merges.

  bzr log -n0 /path/to/openssh/debian/trunk | less

  # find revision number for previous tarball import, hence 'PREVIOUS'

  bzr merge -rPREVIOUS /path/to/openssh/debian/trunk

  # merge history only, no file changes

  bzr revert .

  bzr ci -m 'add 5.2p1 tarball parent revision'

  bzr tag upstream-5.3p1

Next, merge this into the gssapi branch

(bzr+ssh://bzr.debian.org/bzr/pkg-ssh/openssh/gssapi/). For this branch, we

want to ignore the normal results of merging and take only the patch from

http://www.sxw.org.uk/computing/patches/openssh.html; of course such a patch

needs to exist first! To do this, run this in the gssapi branch:

  bzr merge /path/to/openssh/tarball/branch

  bzr revert -rrevno:-1:/path/to/openssh/tarball/branch .

  patch -p1 </path/to/openssh/gssapi/patch

  bzr add --file-ids-from=/path/to/openssh/debian/trunk

  # you may need to deal with applying configure.ac changes to configure

  # here

  bzr ci -m 'import openssh-5.3p1-gsskex-all-20100124.patch'

You should now be able to 'bzr merge' from the gssapi branch into the Debian

trunk, resolve conflicts, and commit. If you see lots of "Contents conflict"

messages, you may have got the file-ids wrong. Once you've committed the

merge, you can throw away the tarball branch, as all its history will have

been incorporated.