CVS-like Working Practice

Mercurial allows multiple working practices. In a CVS-like world, you will typically have one central repository; let's call it the "main line". This corresponds to CVS's notion of the "trunk".

Repositories tend to be long-lived, and the "authoritative branches" are clones of the central repository.

Working environment

Developers may have access to the central repositories on a server via SSH (whether that is through "normal" SSH access with each developer having their own login, or through shared SSH logins, but note that they do not actually log in to the server to get a shell prompt), or they may have access via HTTP (including push privileges), and push their changes directly from their local clones into the appropriate repositories.


With CVS, the practice would be as follows:

# Check for work done...
cvs update
# Do work, consider changes...
cvs diff
# Commit changes...
cvs commit

This doesn't always work out...

cvs commit: Examining .
cvs commit: Up-to-date check failed for `test.txt'
cvs [commit aborted]: correct above errors first!

...and so the following workflow is typically required if cvs commit complains:

# Obtain new changes...
cvs update
# CVS may complain about "conflicts during merge".
# Edit conflicts, then commit...
cvs commit

Note that CVS won't let you commit anything new until you deal with any conflicts.

The basic workflow in Mercurial

The corresponding practice would be as follows with Mercurial:

# Check for work done (part of the update operation in CVS)...
hg pull
# Start working with the latest changes (part of the update operation in CVS)...
hg update
# Do work, consider changes...
hg diff
# Commit changes (part of the commit operation in CVS)...
hg commit
# Push changes (part of the commit operation in CVS)...
hg push

The single command equivalent of cvs update is really...

hg pull -u

When you invoke a pull operation, Mercurial will invite you to update to the pulled revision. If you have uncommitted changes to some files and someone starting from the same revision has already committed changes to those files, Mercurial will try and get you to merge both sets of changes. This is more or less what CVS does in this situation.

You don't usually want to do hg pull -u or hg update unless you want to work with other people's changes. Mercurial gives you the choice of integrating your own work with other people's or just obtaining that other work for separate perusal. See Working independently for more details.

Conflicts and merging in Mercurial

When pulling and pushing, hg pull and hg push may respectively complain about multiple heads being created:

abort: push creates new remote heads!
(did you forget to merge? use push -f to force)

This means that the content of a repository has been edited in different ways, committed to various repository clones, and these different changes have not yet been reconciled. In the diagram below, an attempt is being made to push changes to a remote repository where some "competing" changes have been committed:

The warning about multiple heads revolves around the following situation:

Multiple heads can exist in a repository, either after hg pull in a local clone or after hg push -f to a remote clone, but this leaves any work for reconciling the competing changes for later, which may not be desirable. Nevertheless, Mercurial provides the option of deferring such work, whereas CVS will confront you with the conflict situation immediately. If the remote clone is being used as a CVS-style "central repository" inside which nobody works directly, having multiple heads can be inconvenient because they cannot be merged without cloning or pulling from that repository, doing the necessary merges, and then pushing the work back.

To reconcile competing changes, the following workflow becomes necessary:

# Get the remote changes...
hg pull
# Heads were created, so merge them...
hg merge
# Edit any conflicts, then commit...
hg commit
# Push merged changes...
hg push

Here, we number the operations to make their order a bit more obvious:

Again, you can defer pushing changes until later. The principal advantage of Mercurial here is that you do not need to be confronted with merging others' changes on every commit.

A note about file-specific commits

(See also the notes on the "unit of work" in CVS and Mercurial.)

In CVS, it is possible to commit specific files, and if your work on these files does not conflict with other people's work - they have perhaps been working on other files instead - then your changes will be accepted by the repository without any further action:

# Change a specific file, then commit only that file...
cvs commit test.txt

This will succeed, despite the repository having updates available on other files.

However, Mercurial does not version files independently: the repository as a whole is versioned. Consequently, if someone has changed a file in the repository, even if such changes are independent of changes you have made to another file, a conflict situation can arise.

# Change a specific file, then commit only that file...
hg commit test.txt
# Now try and push the commit...
hg push

This will cause a complaint about multiple heads.

In order to have the changed files co-exist in a single repository revision, a pull and merge is necessary, as described above.

Working independently

With CVS it is not possible to perform various version control operations without getting the remote repository involved, but with Mercurial you can keep working and versioning your own work without needing to pay any attention to what others might be doing.

If you perform an hg pull but not an hg update you can check what other people have been doing, but your own work will not be affected by their changes.

Ultimately you are likely to want to combine your own work with theirs, so an update will need to happen at some stage.

If you do not perform an hg push after committing changes, you remain free to keep committing as many changesets as you like without being affected by people working elsewhere.

Branches and merging

To set up a branch, you can clone a repository and treat one of the clones as the main line and the other as a branch:

# Outside the project directory.
hg clone project project_maintenance
# Or inside the project directory.
hg clone . ../project_maintenance

Here, the project clone might act as the main line and the project_maintenance clone as a branch.

Someone may be responsible for "backporting" changes from a branch to the main line. They do this by pulling changes from the branch and the main line into a local repository, merging appropriately, then pushing back to the main line.

When the main line reaches a release point, someone creates a clone on the server at the appropriate revision, and people who need to work on that branch clone it, then start pushing their changes back.

See also

CvsLikePractice (last edited 2012-06-20 16:49:32 by PaulBoddie)