Re: [Monotone-devel] mtn copy anyone?

[Markus Schiltknecht]

Hi,

Brian May wrote:
> I would prefer we stayed out of this mess. Don't explicitly track
> history, but maybe record some sort of pointer indicating where the
> file came from, if required.

Over time and considering importing or exporting to other VCSes, that 
will provoke an ever nastier mess, IMO. The purpose of a VCS is to keep 
track of history, not to write a textual log and let the user figure 
what history looks like.

Why can't we provide both copy-like operations with the different merge 
semantics and let the user choose what she wants?

 - a snapshot copying command, which does not merge future changes from 
the source file to the target file. (symmetric?)

 - a branch copying command, which - sort of - branches a single file 
and merges changes to the source. (asymmetric?)


Of course, both methods bring up some things to think about. I'm 
brainstorming somewhat on that:

What if a user no longer wants to merge the changes from the source 
file, i.e. 'detach' it? Can we do that for real branches? No, we don't 
really need to, because merging is an explicit operation and not done 
automatically. We could do the same for copied file and require further 
 merges to be done explicitly. OTOH, snapshot coping would always give 
you a way to 'detach' the file again.

Wouldn't the snapshot copying ability solve the DieDieDie merge problem? 
To revoke a deleted file, do a snapshot copy from the revision where the 
file still existed and merge that with your new head.

What do you think?

Regards

Markus