Re: [Monotone-devel] Re: Poll: another possible problem migrating to rosters

[Nathaniel Smith]

Thanks for the feedback.

On Sat, Nov 19, 2005 at 02:05:11PM +0100, Tom Koelman wrote:
> We have used merging descendends from multiple graph roots for two
> different scenarios in the wild:
> 
>  - Adding a new file that has been developed independently to an
>    existing project, thus preserving both their histories.

Nod, that makes sense.  This is exactly the use case for which moving
tree roots is useful; ideally you want to represent this as a merge of
two trees, where you kept everything from the LEFT tree, and deleting
everything from the RIGHT tree except for the new file.  Where
"everything" here, in both cases, includes the project root directory.

But playing such games with tree roots adds new and exciting failure
cases to merging (subtle things like, merging "add_dir foo/MT" with
"rename foo <root>" needs to raise a special sort of conflict, because
while neither side has a versioned file named "MT", the result of the
merge does...).  So while the current roster code is written with this
functionality in mind, we weren't planning to turn it on until we'd
had a chance to get the basic stuff shaken out.

>  - Because currently there is no sane way to declare a certain branch
>    dead, we created a new branch from scratch called "dead" to which
>    we propagate branches that we want to get rid of.

Hrm, I can think of lots of ways to put some sort of mark on a branch;
make up a magic cert you stick on the branch head, for instance.
Which doesn't give you any particular built-in tool support, but
AFAICT neither does merging to a special dead branch...  Why did you
choose to do it this way instead of some other?

-- Nathaniel

-- 
Details are all that matters; God dwells there, and you never get to
see Him if you don't struggle to get them right. -- Stephen Jay Gould