Automatic Git Repository Generator
This is a simple shell script for automatically generating Git repositories for uses such as the testing of Git tools. The files it creates are very simple at the moment and don't make a very realistic repository; however, it can generate large numbers of changes with branches, tags and merges - already enough to trouble repository browsers and visualisation tools.
It's a shell script called in the following fashion:
./stress.bash -n 1000 my_repo
That would create a repository called my_repo and perform a thousand
operations on it.
The script is called with the following syntax:
./stress.bash [<options>] <repo_name>
Where repo_name is a mandatory parameter and everything else is optional.
The script takes a small number of command-line options at present:
-h or -? Display usage information
-n <number> Number of operations to perform (default 30)
-d <path> Path to a dictionary file (default /usr/share/dict/words)
The script first creates a repository and adds itself to that repository in the initial commit. It then creates three randomly-named files to begin with, each containing a random word. After that, it performs the number of operations specified before exiting.
Each operation begins with the generation of a random number from 0 to 99. This number is used to weight the type of operations taking place - content change commits are naturally more frequent than branch creations and merges.
Create File (frequency around 2%) adds a new file to the repository in the current branch. The file name is picked at random and a random word is inserted into the file.
Create Tag (frequency around 5%) tags the head of the current branch. An extra random choice is made - around 50% of the tags created will be annotated tags.
Create Branch (frequency around 10%) creates a new branch from master. The branch name is chosen at random.
Switch Branch (frequency around 15%) checks out a randomly-selected branch other than the current one. This is to try to make the repository history a bit more interesting.
Merge Branch (frequency around 5%) merges the current branch into master. It always creates a merge commit to keep the history looking nice (it will not do a fast-forward merge). Any conflicts arising are resolved by removing the conflict markers and committing the unique lines from the resulting files. This isn't entirely realistic but it means there'll be enough merges to make the history interesting.
Modify Content (the remainder, around 63%) selects a random subset of the files in the current branch for modification. It then adds a random word to each of the selected files and commits the results. The average size of the commits will naturally grow as the repository grows, which is probably a reasonable reflection of real repositories.
There are undoubtedly portability bugs as I've only been using it locally (Mac OS X). The probability weightings of the different operations should be configurable. The content modification operations should also delete lines and should sometimes insert rather than append. Files should sometimes be removed. Basically, it should better reflect real repository characteristics.
I'd like to add other operations (cherry-picks, notes, forced commits, signed commits) because they're also useful test data when playing with visualisation tools. Handling of binary files would possibly be useful too.
Suggestions, bug reports and patches all welcome: git-stress
git-stress is released under the MIT license. See LICENSE.md