Why not hard link for hg copy?

[Peter Arrenbrecht]

On Tue, Jun 9, 2009 at 4:46 PM, Neal Becker<ndbecker2 at gmail.com> wrote:
> Matt Mackall wrote:
>
>> On Tue, 2009-06-09 at 10:21 -0400, Neal Becker wrote:
>>> Subject says it all.
>>
>> Other programs cannot be trusted to do the right thing with hardlinks.
>> Otherwise, normal cp would use hardlinks by default as well.
>>
>
> I suppose this actually brings up the question of just what are the
> semantics of hg copy supposed to be?  I was assuming that this marks the
> file as _forever_ being a duplicate.  In this model, you would not be
> allowed to check in a change to file A that is different than file B.  You
> would have to tell hg they are no longer copies.
>
> If this is not the semantics, then it's not obvious to me what the purpose
> of hg copy is.

The purpose is to help merge be clever:

  hg init foo; cd foo
  echo A >a
  hg ci -Am orig-a
  hg cp a b
  hg ci -Am copied-to-b
  hg up 0 # revert to before we copied a->b
  echo B >a # overwrite contents of a
  hg ci -Am modified-a
  hg up -C 1 # back to where we copied a->b

When we merge, we get:

  $ hg merge tip # merge change to a
  merging b and a to b
  1 files updated, 1 files merged, 0 files removed, 0 files unresolved
  (branch merge, don't forget to commit)
  $ ls a
  B
  $ ls b
  B

So merge recognized a fix to a that should also go into b (it reckons).
-parren