sychronize tracking repositories
2012 by Simon Thum
This scrips intends to sync near-automatically via git in "tracking" repositories where a nice history is not as crucial as having one.
Licensed under CC0
Suppose you have a set of text files you care about, multiple machines to work on, and a central git repository (a.k.a. bare reporitory) at your disposal. You do not care about atomic commits, but coarse versioning and backup is grave. For example, server configuration or org-mode files.
In that case, git-sync will help you keep things in sync.
Unlike the myriad of scripts to do just that already available, it follows the KISS principle: It is safe, small, requires nothing but git and bash, but does not even try to shield you from git. It is non-interactive, but will probably exit with a useful hint or error.
It is ultimately intended for git-savy people. As a rule of thumb, if you know how to complete a failed rebase, you're fine.
Tested on msysgit and a real bash. In case you know bash scripting, it will probably make your eyes bleed, but for some reason it works.
What does it do?
It will likely get from you from a dull normal git repo with some changes to an updated dull normal git repo equal to origin. It does this by commiting, pulling & pushing as appropriate.
Care has been taken that any kind of problem, pre-existing or not, results in clear error messages and non-zero return code, but of course no guarantee can be given.
The intent is to do everything that's needed to sync
automatically, and resort to manual intervention as soon
as something non-trivial occurs. It is designed to be safe
git-sync will likely refuse to do anything not known to
You can invoke git-sync in "check" mode, in which git-sync will not do anything except return zero if syncing may start, and non-zero if manual intervention is required.
How am I supposed to use it?
Mode can be empty, sync, or check.
In "check" mode, it will indicate if syncing may start. This is useful to see if manual intervention is required (indicated by text and non-zero exit code).
In sync mode (the default), just calling
git-sync inside your
repository will sync with the current branches remote, if that
branch is whitelisted. The repository must not be in the middle of a
rebase, git-am, merge or whatever, not detached, and untracked files
may also be a problem (see Options). Likely, sync will just
work. Else, a clear error message should appear. If you don't sync in
an intertwined manner (from multiple repositories/machines),
git-sync is virtually guaranteed to succeed. Otherwise it will try
to rebase, which may fail. This is where you'll need your git skills.
How does it work?
The flow is roughly:
- sanity checks. You don't want to do this in the middle of a rebase.
- Check for new files; exit if there are, unless allowed (see Options). In check mode, exit with 0 at this point.
- Check for auto-commitable changes.
- perform auto-commit if there are (see Options)
- one more check for leftover changes / general tidyness
- fetch upstream
- Relate upstream to ours. If ahead, push. If behind, fast-forward. If diverged, rebase, then push.
- At exit, assert sync state once more just to be safe.
On the first invocation,
git-sync will ask you to whitelist the
current branch for sync using git config. This has to be done once for
every repository (and branch, for completeness).
Because git-sync rebases, the order of commits does not always reflect the order of changes. However auto-commit records originating machine name and time by default.
There are three
git config-based options for tailoring your sync:
Tells git-sync to invoke auto-commit even if new (untracked) files are present. Normally you have to commit those yourself to prevent accidential additions. git-sync will exit at stage 3 with an explanation in that case.
A string which will be used in place of the default commit message (as shown below).
A string which is being eval'ed by this script to perform an auto-commit. Here you can run a commit script which should not leave any uncommited state. The default will commit modified or all files with a more or less useful message.
By default, commit is done using:
git add -u ; git commit -m "changes from $(uname -n) on $(date)"
Or if you enable
git add -A ; git commit -m \"changes from $(uname -n) on $(date)\";"
I declare this work to be useable under the provisions of the CC0 license.
Attribution is appreciated, but not required.
Thanks go to all the people behind git.