PartialClone feature -- how hard would it be to do?

[Matt Mackall]

On Fri, Dec 14, 2007 at 10:40:08AM -0800, Chris Quenelle wrote:
> 
> There is a wiki page here:
>   http://www.selenic.com/mercurial/wiki/index.cgi/PartialClone
> and an issue record here:
>   http://www.selenic.com/mercurial/bts/issue105
> 
> that describe a proposed feature "Partial Clone" which would allow you
> to create a clone with only some subtrees of the your sources.  Not
> just in your working directory, but also in the history DB itself.
> 
> I've always assumed this would difficult because of the way the revision
> checksums are created and used globally.  But I can't find a more in-depth 
> discussion of how hard this would be.   Was this addressed in a previous
> email someplace?  Maybe I could copy/paste that discussion onto the wiki 
> page?
> 
> How hard would this be to implement?

On a scale of 1 to 10, I'd rate it about a 7.

The simplest approach is probably "allow cloning of subdirectories".
Here's how that would work:

- add a file called "subdir" to .hg/store containing the
  subdirectory we're cloning
- do a clone as usual, but don't write out revlogs that aren't in subdir
- adjust the functions dealing with paths (localrepo.wjoin, etc.)
  appropriately
- teach various things (checkout, status, etc.) to skip files outside
  the subdirectory
- teach merge not allow merging when files outside the subdirectory
  conflict

Merging is the tricky part, and I expect there's a gotcha or two
buried in there.

-- 
Mathematics is the supreme nostalgia of our time.