Mercurial authentication info in history

Question

I have a "central" Mercurial repository, which configured to use HTTPS and requires authentication to clone-pull-push changes. Developers has their own repositories on their computers. They configure their local settings freely, and for example add section like

[ui]
username = anyname

to their local mercurial.ini file.

When a user try to push his changes to the "central" repository, he authenticates, but authentication info is not stored in Mercurial. Mercurial store locally configured username as revisions author in central repository. So I cannot find who really made changes in central repository, but I strongly wish to do it. Mercurial developers does not care about it and consider this behavior to be correct.

But I want to keep authentication info near changesets. I think the best way to do it is add one more additional field in revision description, like "pusher id" and store there authentication data.

Extensions I found do not implement similar functionality. Can you give me info about some third-party extensions, hooks, or just code templates or ideas how to do it? (I'm absolutly new in Python)

Answer 1

The fundamental problem that makes Mercurial developers (like myself) reject this is that changesets are immutable. It is impossible for a server to add extra information to the changesets when they are pushed.

More concretely: a changeset is identified by it's changeset hash. This hash is computed based on all the information the changeset contains, such as username, date, commit message, and the change itself. You cannot change any part of this, without also changing the changset hash — otherwise the integrity of the repository is destroyed.

This gives you security against accidental (or malicious!) changes made on the server: if Alice and Bob talk about "changeset X", then they can be sure they really mean the same thing. If the server (or someone else) could change the content of a changeset without affecting the ID, then Alice and Bob would not be guaranteed that "X" really means the same thing in both their repositories. This property is of course also fundamental to the way Mercurial works when synchronizing repositories.

You have two options here:

1. You can let the server reject a push if Alice tries to push a changeset with Bob's name in it. This is can be done with a pretxnchangegroup hook on the server. It will inspect the HG_SOURCE environment variable and verify that the user listed there is also the committer of all pushed changesets between HG_NODE and tip.

2. You can let the server log the pusher. This is called a "pushlog". The Mozilla project uses one and the source appears to be here. There you make your server store information about who pushed what. This is done in a changegroup hook that logs the necessary information in a small database.

   If you want a push log, then take a look at Kallithea, which has this functionality built in. Kallithea is in general a great way to host Mercurial repositories! It has much more functionality than the normal hgweb CGI script.