We've now completed the switchover to mercurial. It took a little longer
than planned, but everything seems to be working OK.
The Xen mercurial repositories live at
http://xenbits.xensource.com/xen-unstable.hg (and similarly for
xen-2.0-testing.hg and xen-2.0.hg). You can also point a browser at this
location to view the mercurial revision history.
The mercurial repository retains pretty much all the revision history
and structure of the old bk repository. The only serious missing feature
in hg is support for file renames. This means that the revision history
for the various linux sparse tree versions has become segmented, which
makes follow changes awkward. We've retained all the rename information
in changeset comment metadata, and when mercurial finally gets rename
support, we'll be able to create a new repository that reconstructs this
information (unfortunately, the new repository will not have a common
ancestor with the old one, so this will require another 'flag day').
A mercurial patchbot has been created that will email out full patches
to the xen-changelog@xxxxxxxxxxxxxxxxxxx mailing list within a few
minutes of changesets being committed.
I've attempted to integrate mercurial into the nightly build, but this
is completely untested. It's almost certain to fail, but I'll sort it
out in the morning.
Appended is a mercurial 'cheat sheet' that should help get people up to
The switch-over to mercurial has required a lot of effort. Particular
thanks are due to Michael Fetterman and James Bulpin, but also Andrew
Warfield, Arun Sharma and others.
Bk is dead. Long live hg!
Mercurial(hg) Cheatsheet for Xen
Written by Andrew Warfield, extended by Michael Fetterman and Ian Pratt
June 29, 2005
The Xen project has moved from BitKeeper to Mercurial for source
control. This note aims to provide a quick guide to getting up and
running with the new tools as quickly as possible, and is written from
the perspective of someone who has been using BK.
For a more detailed exposition, see the mecurial tutorial:
The Hg manpage is available at:
There's also a very useful FAQ that explains the terminology:
There's also a good README:
Mercurial is available at:
You will also need a Python version >= 2.3
How Mercurial is different from BK
There are several pertinent differences between hg and bk. This section
aims to give an overview of the conceptual differences between the two
SCMs -- if you just want examples to get going, skip ahead to "Getting
Xen". The key differences are:
- No explicit per-file locking. You do not need to explicitly
check a file out before editing it.
- No notion (currently) of file renames.
- A repository can have multiple active heads.
- Automatic merge support is currently inferior to BK's.
- No graphical tools.
- No per-file revision history, only per-changeset (we never really
used this anyhow)
- Hg repositories tend to be rather bigger than Bk ones, but Hg does
Mercurial is based on the notion of changesets as complete, immutable,
versions of the repository. You make changes to a working version of
the repository that is based on a particular changeset. When you
commit, you will generate a new child changeset with whatever changes
you choose to apply.
A major difference between Hg and BK is that you aren't forced to
resolve conflicts immediately: BK forced you to resolve conflicts
immediately on any merge, and it then immediately created a changeset
with those conflicts' resolutions. Frequently, you then had to add yet
another changeset to fixup the things for which the automatic merge
yielded bad results. Hg puts the results of the merge into your work
directory, and remembers what you merged with (so that it can later
record both of the merge parents, if you decide to make a changeset),
but it doesn't immediately create a changeset.
A further feature of Hg is that it allows a repository to have multiple
heads. This means that you can have changesets with no common descendent
in one repository -- something BK won't allow. This is actually pretty
neat. For example, it would enable you to have both the 2.0-testing and
unstable trees in a single repository.
To get familiar with Hg, it's best to start playing with an existing
repository (see "Getting Xen", next), and understand the "history",
"co", and "manifest commands in Hg. These are discussed below in the
context of working with a repository, but I'll mention them here in an
attempt to demonstrate how mercurial structures things:
Assuming you have an existing repository, you can do the following:
hg history | head
Looking at the first few lines of output, you can see the changeset at
the head of the current branch, known as the 'tip' (and given a special
| changeset: 5599:6cbf9ec05cd9e05c0c46a85df7fc00262633cd3d
| tag: tip
| user: kaf24@xxxxxxxxxxxxxxxxxxxx
| date: Tue Jun 28 18:47:14 2005
| summary: bitkeeper revision 1.1768
By default, Hg just shows the first line of the changset comments. You
can find further information with "hg -v history".
The changeset identifier has two parts, a _local_ monotonically
increasing changeset id, 4792 above, and a _global_ hash, which follows
the colon on the changeset line. The hash uniquely identifies the
changeset and its lineage back to the root of the changeset tree
-- it is useful for distributed management and so on. However, as it is
a bit unruly, the local id will allow you to work easily with the local
repo. Additionally, a tags mechanism (described below) lets you give
common names to specific changesets.
You should always use the global hash when referring to versions of the
mainline Xen respoitory. With Bk you could often get away with using the
shortform version, but with Hg the local hashes are pretty much
guaranteed to be different.
The co command lets you change the working version of the repository to
a different changeset. You can use either the local or global hash name,
hg co 4791
This would rewind my working repository to the previous changeset.
There are various options to this, mentioned below and in the
documentation ("hg help co"). In general, it does the right thing,
although it doesn't currently seem to clean up unused directories that
have been created by other checked-out versions. This can confuse the
Xen build system. Hg will probably get fixed soon, but in the meantime
you can cleanup with "find -depth -type d -print | xargs -r rmdir".
You can return to the tip with:
The manifest command lets you see the contents of the repository for the
This will print a bunch of records of the form:
| 98856c45c35a504bc6da06a62b7787ddfdfd1c8b 644 COPYING
| f28971eedc5b54e7a9b26dd18d52992955354981 644 Config.mk
| a3575cc4db59e50bbac8a039a0c74f081a8dfc4f 644 Makefile
| 7fc869aae2945a9f4626fad96552db3103e61cb9 644 README
This lists the hash of each file, its 1-bit 'executable' atribute
(either file permission mode 644 or 755), and the file name. So, to
determine the files that change across two changesets, you would dump
the respective manifests to files, and use diff.
One slightly confusing aspect of Hg is that many of the commands have
aliases, and hence when looking things up in the man page its not always
obvious what the underlying command is. For example 'co' is actually an
alias for the 'update' command, but 'co' seems to make more sense, at
least to RCS refugees like me.
To avoid taxing the Mercurial server with a complete pull of the Xen
repository, it is best to download a tarball of a seed repository
Untar the repository on your disk, cd into it, and then pull the most
hg pull -u
The repository parent is stored in a repository configuration file,
.hg/hgrc, from the repository root. If you look at this file, you will
| default = http://xenbits.xensource.com/xen-unstable.hg
"default" specifies the appropriate parent repository for hg to pull
from. Hg allows you to pull additional repositories, for instance if
you want to work between unstable and testing concurrently.
The command "hg pull" simply adds changesets to your repository, without
any merging of any kind. "hg pull -u" implies merging with the current
state of your working directory. If you weren't already "updated" to
your local repository's tip, you might be surprised to find yourself
merging the results of the pull with a non-tip node in your local
Creating a child repository from an existing repository
If you wanted to create additional local child repositories,
hg init [path or url]
is effectively equivalent to bk clone. The major difference is that it
should be run from the root of your new repository. So:
bk clone /foo/bar
would be replaced with:
hg init /foo/bar
NB: newer version of hg support a 'clone' command that works in the same
manner as bk.
Normal edits may be made in place. File creation needs explicit
marking, though deletes should be picked up automatically
touch a (or otherwise created a file)
hg add a
You can see what has changed using:
| C foo/foo.c
| R foo/bar.c
| ? a.txt
Shows that in the current repo, foo.c has been changed, bar.c has been
deleted, and a.txt is new, but has not been added. '?' changes to 'A'
after "hg add a.txt".
will pop up an editor with a list of files to be committed to the
repository. It will look vaguely like this:
| HG: manifest hash 6397b04bd5c2a992482d973b685a7e5e498788e7
| HG: changed doc/thesis/new.tex
| HG: removed doc/2005-hotdep-protection/paper.tex
Your comments can go anywhere in this file. The first line is the most
important, as it will show as the changeset description in
non-verbose-mode history listings.
You can do commits without the editor and of partial sets of files using
command-line switches. See:
hg help commit
You can use the -A (--addremove) flag to commit e.g. "hg -A commit" to
ash mercurial to scan the tree looking for newly created files to add in
to the changeset. This avoids having to explicitly use "hg add".
Generating a patch
Generating a patch is easy,
hg export [changeset]
will generate a patch describing the diff between that changeset and its
Pushing changesets to a parent repository
Pushes changes up to a parent. You can't push if you pulled the
repository off the web interface. In fact, you can currently only push
to an ssh target -- filesystem drectory targets don't work, but this
will be fixed soon.
Here are a collection of common commands to get you started:
hg history | less
shows the history of changesets, starting from the most recent. You
want to pipe it to some sort of pager. For more complete details,
hg -v history | less
will include files modified and full (not just first-line) comments.
Additionally, you can see just the tip (head of the current
branch) of the repository by typing:
hg [-v] tip
Moving to a specific changeset
hg co [changeset]
will update the working directory to reflect the specified changeset. By
default, it will refuse to overwrite any modified files.
You can use the -m option to cause the checked out version to be merged
into the current tree. The "-C" (--clean) option will additionally
overwrite any locally modified files.
Any commits that are made to non-head changesets will obviously fork the
tree, creating a new head. You can see all the heads in a tree with "hg
Managing changeset tags
To create a tag to the current changeset:
hg tag tagname
This will _immediately_ generate a changeset with a change to the file
.hgtags in the repository root. The new tag in this file will look
| 35159ed4b30538e7a52c60ad0a63f7e9af156e4c tagname
and may be used to identify that changeset throughout the repo.
Storing tags in this file and generating changesets immediately forces
people to merge and keep tags up to date across the repository.
Note that tags are resolved by searching .hgtags in each of the
repository heads, sequentially, and using the first match. "hg heads"
lists the current heads.
The "hg tags" command, will lists all the currently valid tags.
Hg server and source browser
hg serve -p port
Launches a web server on the specified port, serving a source browser
for the repository. This browser may be used to examine the changeset
history, view annotated source files, generate diffs.
Additionally "hg pull" may be run against it.
Additional useful commands
(that probably only need one-line descriptions)
(Slightly) more detail on all of these is available with
hg help [command]
View an annotated version of a source file:
Get a historical version of a file:
NB: Most commands accepting a version number want the changeset's
version number. "hg cat" is different in that it wants the *file*'s
Unadd a file to the current commit:
List all heads in the current repository:
Undo exactly one (and ONLY one) changeset:
Shows the differences between whatever changeset you most recently
checked out, and your current working directory:
Show the parents of a changeset:
NB: Changesets have either one or two parent changesets. If your
working directory contains the uncommitted results of a merge, then you
have two parents. Otherwise, the single parent is the changeset which
you most recently checked out.
Xen-devel mailing list