Pull Request Manager

Description

A program that continually checks and potentially merges GitHub pull requests.

Protocol

The protocol used to choose a pull request to process is the following:

1. Find all open pull requests associated with the specified repositories;
2. From those, select only pull requests made by the administrators of these repositories;
3. For each pull request:
   • read its comments, and determine whether the bot's last attempt has succeeded, and whether either the repository's branch to which the pull request was made, or the pull request itself, has changed since the last attempt;
   • if an administrator of the repository has posted a comment that starts with @<bot_name>, followed by a regular expression specified on a line in positive.txt (case ignored), followed by . or !, followed by any text, and the bot has previously succeeded or the refs changed, prefer processing this pull request;
   • otherwise, process a pull request only if no pull request satisfies the previous step, and the corresponding refs of the pull request have changed.

Once a pull request has been chosen, it is processed as follows:

1. Try building the underlying system with the changesets from the pull request:

    make <component-for-pull-request's-repository>-build
    make api-build
   

   If these steps fail, report the problem as a comment on the pull request, and stop processing this pull request.

   While currently not enabled, it is easy to configure the bot to only re-build the system if refs have not changed. This is disabled, since it is safer to re-build the system before a merge even if the refs have not changed, because the bot is not checking refs of all dependent repositories (many of which are not on GitHub).

2. If a merge is requested (through administrator's positive comment; see above), verify that refs have not changed since the pull request started being processed:

   • if refs have changed, report the problem as a comment on the pull request, and stop processing this pull request;
   • otherwise, push the changesets of the pull request into the requested branch of the main repository, comment about this on the pull request, and close the pull request.

The above process is repeated continuously. After every processed pull request, or if there is no pull request to process, the program waits for a while (1min). If a connection error occurs, the program waits for a longer period of time (10min) before retrying. Privileges are refreshed every 5 runs.

In a pull request's description, one can express its dependencies on other pull request. For example, one can write the following:

Dependencies: 23@my-repo

The above implies that this pull request will not be processed before pull request 23 in my-repo (of the same GitHub organisation) has been merged.

Dependencies on other libraries

The program currently depends on:

• python-github2, a Python implementation of GitHub's API.

Setup

Clone the module and its submodule(s) with:

git clone git://github.com/xen-org/pull-request-manager.git
cd pull-request-manager
git submodule init
git submodule update

The following steps are required to run the program:

• Make sure the dependencies are satisfied;

• Create settings.py, and within define bot_email, bot_api_token, and builds_path variables with appropriate values; and,

• Create a github-xen-git target in your .ssh/config file, which points to the bot's private key:

      Host github-xen-git
        HostName github.com
        User git
        IdentityFile /home/roks/.ssh/id_rsa_xen_git
  

Running

To start the program, execute the following command:

python main.py

Extras

The tool also features support for JIRA through the jiralib project. To use it, define jira_url, jira_username, and jira_password variables in settings.py. See the closeTicket function within main.py for more information.

Feedback and Contributions

Feedback and contributions are welcome. Please submit contributions via GitHub pull requests.