Getting Started with Modular Boost Library Maintenance

This page briefly sketches the mechanics of maintaining a Boost library using Git and Modular Boost. The intended audience is developers getting started with Git and Modular Boost maintenance of existing Boost libraries.

This page is intended to get you started only; it does not provide in-depth coverage. See links below for that.

The Big Picture

Library maintenance occurs in the context of how Boost repositories are organized. Please study the Modular Boost Overview before continuing, since a Boost developer needs to be familiar with how Boost organizes its repositories.

The examples given on this page follow Boost recommended workflow practices, but keep workflow discussion simple for this introduction. To better understand workflow recommendations and rationale before continuing, feel free to read Modular Boost Library Workflow Overview.

Prerequisites

• A recent release of the Git command line client installed. See Getting Started with Git for the git version required.
• A (free) ​GitHub account. See Getting Started with GitHub.
• A compiler and development environment installed and working smoothly.
• Modular Boost installed as described in Getting Started with Modular Boost.
• b2 in your path. That allows the command line examples given here to work as shown on both Windows and POSIX-like systems.

Typical maintenance tasks

Getting ready to work on a library

The preferred environment library maintenance is to checkout the library's develop branch, or some other development branch, while other Boost libraries are as defined by the Boost super-project master branch. This causes local tests of your library to run against master for other Boost libraries at the point in time referenced by the Boost super-project.

This is a more realistic test environment in than testing against the possibly unstable develop branch of other Boost libraries or against the master branch of other libraries at a different point in time than that referenced by the super-project. Robert Ramey has advocated this approach to testing for years, and Git plus Modular Boost makes this approach relatively easy and fast.

cd modular-boost
git checkout master
git pull
git submodule update

The git submodule update will fail if it would result in uncommitted changes being overwritten.

The git submodule update may switch submodules back to the detached state, depending on the working copy's exact situation.

The following table, based on actual tests run using git version 1.8.4.msysgit.0, shows the effects on a submodule of the above commands.

Submodule Branch	Submodule Contents	Submodule relative to super-project	Effects on Submodule
Detached	Unmodified	Up-to-date	None
Detached	Unmodified	Behind	git pull
Detached	Uncommitted change	Up-to-date	error: Your local changes to the following files would be overwritten by checkout: ... Please, commit your changes or stash them before you can switch branches. Aborting Unable to checkout (SHA...) in submodule path '...'
Detached	Uncommitted change	Behind	error: Your local changes to the following files would be overwritten by checkout: ... Please, commit your changes or stash them before you can switch branches. Aborting Unable to checkout (SHA ...) in submodule path '...'
Detached	Committed change	Up-to-date	git checkout --detach
Detached	Committed change	Behind	git checkout --detach; git pull
master	Unmodified	Up-to-date	git checkout --detach
master	Unmodified	Behind	git pull (but not detached)
master	Uncommitted change	Up-to-date	error: Your local changes to the following files would be overwritten by checkout: ... Please, commit your changes or stash them before you can switch branches. Aborting Unable to checkout (SHA...) in submodule path '...'
master	Uncommitted change	Behind	error: Your local changes to the following files would be overwritten by checkout: ... Please, commit your changes or stash them before you can switch branches. Aborting Unable to checkout (SHA...) in submodule path '...'
master	Committed change	Up-to-date	git checkout --detach
master	Committed change	Behind	git checkout --detach; git pull
develop	Unmodified	Up-to-date	git checkout --detach
develop	Unmodified	Behind	git checkout --detach; git pull
develop	Uncommitted change	Up-to-date	error: Your local changes to the following files would be overwritten by checkout: ... Please, commit your changes or stash them before you can switch branches. Aborting Unable to checkout (SHA...) in submodule path '...'
develop	Uncommitted change	Behind	error: Your local changes to the following files would be overwritten by checkout: ... Please, commit your changes or stash them before you can switch branches. Aborting Unable to checkout (SHA...) in submodule path '...'
develop	Committed change	Up-to-date	git checkout --detach
develop	Committed change	Behind	git checkout --detach; git pull

To get more information about submodules:

cd modular-boost
git submodule summary

If for some reason you wanted to test against the current head of master for all libraries, disregarding the super-project state, the git submodule update command would be changed to:

git submodule foreach --recursive "git checkout master; git pull"

Check out the development branch of your library

You can see what branch mylib is currently on like this:

cd modular-boost/libs/mylib
git branch

Then if you need to change the branch to a development branch such as develop, do this:

cd modular-boost/libs/mylib
git checkout develop

You only have to do that once; your local repo working copy will sit on the branch until it is explicitly changed by a command you give.

Of course, you don't have to change the directory before every command, and from here on this tutorial will assume the directory has not been changed since the prior example.

If there is any possibility the branch head content in the public upstream repo has changed, you also will want to update content:

cd modular-boost/libs/mylib
git pull

From this point on, it is assumed you have already done a cd modular-boost/libs/mylib.

Testing locally

Unless you are 100% sure of the state of your library's regression tests, it is a good idea to run the regression tests before making any changes to the library:

pushd test
b2
popd

Checking status

Before making changes, it is a good idea to check status. Here is what that looks like on Windows; the message you get may vary somewhat:

>git status
# On branch develop
nothing to commit, working directory clean

Fix a simple bug directly on develop

For simple bugs, particularly in projects with a single maintainer, it is common practice to fix bugs directly in the develop branch. Creating a test case with your favorite editor, testing the test case, fixing the bug, testing the fix, and then iterating if necessary is no different than with any programming environment.

Once the fix is complete, you then commit the fix locally and push from your local repo up to your public boostorg repo on GitHub. These same commands would be used for any Git project, modular or not, so hopefully you are already somewhat familiar with them:

cd modular-boost/libs/mylib
git commit -a -m "my bug fix"
git push

There are some significant disadvantages to this simple approach:

• The fix is now made to develop but you must remember to merge it to a release branch or directly to master. It is very easy to forget to do that merge, particularly if this is a mature library you are not working with very often.
• Users who need the bug fix right away are forced to jump through hoops to retrieve the fix from develop.

Putting out a point release solves both of those problems. Read on...

Fix a bug using a hot-fix branch

Fixing a bug directly on the develop branch is fine, if that's the library's policy, but if the bug is messy, multiple maintainers are involved, interruptions are expected, or other complexities are present, then it is better practice to work on the bug in a separate branch. And doing that on a hot-fix branch solves the problems mentioned at the end of the prior section.

The operational distinction between a bug-fix branch and a hot-fix branch is that a bug-fix branch is branched from develop and then at completion merged back to develop, while a hot-fix branch is branched from master and then at completion is merged to both master and develop. With either approach, the branch is deleted after it has been merged.

git checkout master
git checkout -b hotfix/complex-boo-boo

This creates the branch hotfix/complex-boo-boo, and switches to it. Incidentally, hotfix/ is part of the name, not a directory specifier. The new branch is based on branch master because the working copy was on branch master at the time of the branch.

Since the bug is complex, it may take some time to fix and may go through several cycles of fixes, tests, and commits.

Once the bug is fixed and a final commit is done, then it is time to merge the hotfix/complex-boo-boo branch into master and develop:

git checkout master
git merge hotfix/complex-boo-boo
git push
git checkout develop
git merge hotfix/complex-boo-boo
git push
git branch -d hotfix/complex-boo-boo

Start work on a new feature

Developers are encouraged to create a (possibly private) branch to work on new features, even simple ones, since development of new features on the develop branch might leave it unstable for longer that expected. Using the Git Flow convention, the branch will be named feature/add-checksum-option.

git checkout develop
git checkout -b feature/add-checksum-option

When you create the branch, or perhaps later, you may decide the branch should be public (i.e. be present in the library's public boostorg repo) so that you can share the branch with others or just to back it up. If so, set that up by running:

git push --set-upstream origin feature/add-checksum-option

Whether or not --set-upstream origin bugfix/complex-boo-boo is actually needed depends on the branch.autosetupmerge configuration variable that isn't discussed here. If you don't supply --set-upstream origin bugfix/complex-boo-boo on your first push and it turns out to be needed, you will get an error message explaining that.

The usual cycle of coding, testing, commits, and pushes (if public) then begins. If other work needs to be done, a stash or commit may be done to save work-in-progress, and the working copy switched to another branch for awhile. If significant fixes or other enhancements have been made to develop over time, it may be useful to merge develop into the feature branch so that the eventual merge back to develop has less conflicts. Here is how the merge from develop to feature/add-checksum-option is done:

git checkout feature/add-checksum-option
git merge develop

First post-svn conversion merge to master

When you are ready to merge the develop branch to master (or better yet a release branch that's branched off master), there's a bit of housekeeping to be done the first time after the conversion from svn so that future merges proceed smoothly. We're going to create a merge point between the develop and master branches so that Git knows the last point the two branches were in synch. Once we've done that Git will perform a merge by replaying the commits on develop on top of master, starting from the last known merge: in other words Git will perform the tricky stuff of figuring out what to merge for us.

• Navigate to the history for your library on GitHub, starting with branch master. For example the Config library ​can be seen here.
• Look down through the commit history and make a note of the last merge in svn land. In our example, it was on October 25th 2013.
• Use the dropdown box to change the history to point to the develop branch. ​Config library can be seen here.
• Find the last commit before the date noted above. Click on the commit message for that commit to go to the actual diff for that change, ​in our example here. The SHA1 for that commit is shown below and to the right of the commit message, in this case it's 67f6b934f161dc5da2039004986a14d9217afae4. Copy that SHA1 to your clipboard.

Create a merge to that specific commit. Begin by changing your library to the master branch:

git checkout master

Now create a merge to the specific commit above; since we don't really want to actually change the master branch we'll use the -s ours option to avoid any conflicts:

git merge --no-ff --no-commit -s ours 67f6b934f161dc5da2039004986a14d9217afae4

You can now use git status and git diff to check for modifications, in this case there are none. Now make the commit and push to GitHub:

git commit -am "Create first merge point for Git"
git push

Then navigate to your libraries history again, and check that the merge shows up, ​our config example is here. You're now ready for "routine" merges to proceed as per Git Flow (or whatever other strategy you wish to use).

Lightweight library release

Small, simple libraries and simple releases just merge develop into master, tag master, and declare victory.

git checkout master
git merge --no-ff develop
git tag -a 1.56.1

Heavyweight library release

Large, complex libraries, particularly those with multiple developers working in parallel, need to use a release procedure that scales up better than the lightweight procedure. The Git Flow approach is recommended. Find out more at Modular Boost Library Workflow Overview and be sure to study the examples given in ​Vincent Driessen's original blog posting.

Acknowledgements

Beman Dawes created and maintains this page. The content has been revised many times based on comments and list postings from Andrey Semashev, John Maddock, Daniel James, Michael Cox, Pete Dimov, Edward Diener, Bjørn Roald, Klaim - Joël Lamotte, Peter A. Bigot, and others.