What is a branch?

[Jesse Glick]

Mark A. Flacy wrote:
> Perhaps it would be more useful to mention those cases where 
> manipulating a "set of changesets in a named branch in a repository
> containing other named branches" is superior to manipulating a "set
> of changesets in a special repository."  I cannot think of any, but
> there are others in the list who should be able to do so.

1. Using a separate repository forces you to have a separate working 
directory, which is a big deal if your checkout is a gigabyte and your 
build process takes an hour and you have environment variables and 
various tools expecting your sources at a particular file path. Could 
also be solved using the experimental local branches extension, I guess. 
Using a separate repository does have the advantage that you can leave 
uncommitted local modifications in one repository while working on 
another, without using the often-wished-for "shelve" feature or its 
equivalent.

2. For long-lived branches that get merged back, committing to a named 
branch records a branch name in history, making it perhaps easier to 
understand where a set of changes came from. Otherwise the branch is 
anonymous after it has been merged and its old repository is discarded.