This page explains what you need to know if you want to start hacking on VIFF. In addition to these instructions you will want to read up on the coding style used by VIFF (it is the normal Python style, nothing fancy there).
VIFF is developed using Mercurial (also known as hg after its command line program). This is a distributed revision control system which allows you to participate fully in the development even if you do not have what is traditionally known as “commit access”. You can also work offline and take advantage of the many fast operations offered by Mercurial.
After installing Mercurial you can checkout a copy of the source using this command line:
hg clone http://hg.viff.dk/viff/
This will create a directory called viff/ where you will find the source code. Please test that VIFF works as expected on your computer by following the instructions on unit testing.
When you have created a new feature or fixed a bug, then you need to send your changes to one of the VIFF developers. If you share a file system with one of the developers, then the easiest way to get your changes back into VIFF is to ensure that one of the developers has read access on the repository files. He can then simply pull the changesets over and eventually push them out to the VIFF repository.
Alternatively, you can setup a Mercurial repository where one of the developers can hg pull from. You can do this by uploading your clone on some public web server (any webserver works for this since the developers can pull using hg pull http://...) or by running:
hg serve -p 8000
which creates a web server on port 8000 just like the one running at http://hg.viff.dk/viff/. The default port number is 8000, so you can leave that out.
A final option is the patchbomb extension for Mercurial, which will allow you to use:
hg email -t email@example.com -o
to send the changesets not present in the VIFF repository (-o) to the VIFF patches list (-t firstname.lastname@example.org). You will probably want to test using a -n flag or by sending the patches to your own address first to make sure everything looks okay. You can get the full list of options using hg help email.
The advantage of using patchbomb is that allows everybody to go over the code and comment on it before the changesets are pulled into the repository. The mails sent out by patchbomb contain all the metadata needed (name of committer, date, parent changeset, etc.) for importing the changes into the repository, just as if the changesets had been pulled using hg pull.
Please format your commit messages as follows:
topic: terse one-line summary (60 characters max) After the summary line, you're encouraged to provide a bigger description of your changes. If your change is small or obvious from the diff, you can leave this out. Please wrap your paragraphs at ~70 characters or so. That ensures that they are readily readable on a standard terminal.
The topic in the summary line describes which part of the code the changeset touches. There’s no fixed list of topics, but a change to viff/X.py normally gets the “X” topic.
When developing your changes you will probably make many commits representing contained steps. Even though you have made a commit, you can still change it as long as you have not shared it with anybody. The idea is that you are allowed to rewrite history as you see fit in your own private repository, but if your changes have been pulled to the outside, then you can no longer change them.
Also, you can only change commits in a linear history back from your repository tip. This means that if you pull in changes from the main VIFF repository and merge them periodically, then you can no longer edit changesets past the last merge. We therefore recommend that you develop your feature until you are satisfied with it and only merges with whatever new changesets there might be in the VIFF repository when the feature is done and debugged.
Now, to change a past commit you use the Mercurial Queues extension also known as MQ. It gives you a powerful set of tools to work with the past history. The basic concept is that changesets can be converted into patches, which depend on each other and form a stack. Like any good stack, you can push and pop elements from it. In this case you push and pop patches.
To get started you will want to import the normal changesets into MQ. Let us suppose you found an error in revision 430 (use hg view or hg log to find the revision numbers). You then want to import revision 430 and the following changesets into MQ with this command:
hg qimport -r 430:tip
Nothing much happened — your working directory is left unchanged. To see that the command did something you can check the current patch series with hg qseries. When importing changesets revision N is called N.diff in the patch series.
What we want is to pop off the other patches so that 430.diff is the topmost patch. This is done with:
hg qgoto 430.diff
This updates your working directory to look exactly like it did when you originally committed revision 430! You can now edit the files to correct the error you found, and when you are done you run:
to refresh the patch in 430.diff. You can use hg qrefresh -e to edit the commit message too. Now comes the fun part — you must now push the the other patches back on the stack with:
hg qpush -a
If this goes well with no complaints, then you can finish the patches again with:
hg qfinish -a
The end result is a completely normal repository with no sign of this surgery. You can repeat this as many times as needed to slowly develop your patches until you are satisfied with the results.
If the changes you made to the patch are conflicting with other patches in your stack, then the pushing will stop where the error was encountered, and the conflicting patch hunk is stored in a .rej file. There is no need to panic if this happens — all you need to do is to determine how the hunk in the .rej file(s) should be applied (if at all) and then run hg qrefresh to indicate that the current patch is okay. Now continue applying patches with hg qpush -a and fix any remaining conflicts.