Making the world safer for idiots like me [was Re: Add a cgi script that support multiple repositories]

[Bryan O'Sullivan]

On Tue, 2005-06-28 at 00:49 -0700, Matt Mackall wrote:

> - we want to store the repo name (and description) in .hg/hgrc

Could we expand this to "repo name", "description", and "uuid"?

There is currently no prohibition in Mercurial against merging two
completely unrelated trees, and I think this is a problem.

Consider:

        hg clone http://www.selenic.com/hg my-hg
        hg clone http://www.kernel.org/hg my-kernel
        
        cd my-kernel
        hg pull ../my-hg    -> succeeds!  aiieeee!

There are butterfingers types out there who will accidentally perform
pulls like this.  I know this because I am the existence proof.

A certain other SCM package has saved my bacon several times by imposing
a notion of "package", which is "the logical project in this repo and
any clones of it".

Here's a scheme I propose to make hg a little safer, while still being
more flexible than the other SCM.

      * "hg init" creates a UUID, and puts it into a set of UUIDs
        in .hg/hgrc:
        [uuids]
        original = ae31-f015-49ae-130c
      * "hg clone" copies the uuids section from the cloned repo, and
        puts it into .hg/hgrc.
      * By default, "hg pull" and "hg push" honour the uuids section of
        hgrc.  If the source repo's "original" UUID isn't in the list of
        UUIDs on the destination side, the operation is aborted, unless
        you use the "-f" flag.
      * If you repeatedly want use an "unrelated" repo as a source, just
        add its "original" UUID to the uuids section of .hg/hgrc in each
        repo you want to pull/push into.

This adds a safety check, so that you can't pull from an "unrelated"
repo without deliberately deciding to do so.

Thoughts?

	<b