Re: [Monotone-devel] another newbie question - nonlinear ordering

[Jon Bright]

Hi,

I've reordered your mail slightly to put your ending question at the start.

K. Richard Pixley wrote:
> Is this history really lost in monotone?  Or how is this handled?

It was, once upon a time.  Since then, though, changesets were added.  A 
changeset is (roughly speaking), the SHA1 over a textual representation 
of the changeset's parent and the changes within the changeset.

> Start with an empty file, f0.  Apply some initial content, delta1.
>
> f0 + delta1 = f1

Let's say you get a manifest SHA of FEDC for this, and a changeset SHA1 
of 1234.

> It occurs to me that if I now apply a change, delta2, to this file, 
> (tree, subtree, repository, whatever):
>
> f1 + delta2 = f2


The manifest SHA1 of f2 is 5678, say.  To get the changeset ID, you now 
do SHA1(1234 + 5678) -> ABCD.

> And then I back out that change by applying the inverse delta2, delta2', 
> then:
>
> f2 + delta2' = f3 = f1.

Manifest is FEDC again.  Changeset, though, is SHA1(ABCD + FEDC), which 
is something different.


> Since the hash for f3 and f1 will be identical, we can no longer talk 
> meaningfully about a linear sequence of revisions.  That is, revision id 
> = SHA1(f3) = SHA1(f1) no longer has a single ancestor nor is there 
> necessarily even a deterministic path from f3 back to the empty file.  
> f3 and f1 are different not in content, but only in ancestry.

Changesets effectively encode the complete ancestry.  Clearly, I've been 
simplifying here.  change_set.[cc|hh] have more details in their comments.

--
Jon Bright
Silicon Circus Ltd.
http://www.siliconcircus.com